阅读说明：本技术 车辆用学习系统、车辆用控制装置以及车辆用学习装置 (Learning system for vehicle, control device for vehicle, and learning device for vehicle ) 是由 桥本洋介 片山章弘 大城裕太 杉江和纪 冈尚哉 于 2020-08-19 设计创作，主要内容包括：提供一种车辆用学习系统、车辆用控制装置以及车辆用学习装置,所述车辆用学习系统具备搭载于车辆的第1执行装置、车辆的外部的第2执行装置及存储装置,存储装置存储映射数据,所述映射数据是规定映射的数据且包含通过机器学习而学习后的数据,所述映射是以基于车载传感器的检测值的输入数据为输入并输出输出值的映射,第1执行装置及第2执行装置通过协作执行取得处理、算出处理及关系评价处理,所述取得处理取得输入数据,所述算出处理将输入数据作为映射的输入而算出输出值,所述关系评价处理评价与输出值对应的变量以外的预定的变量与输出值的精度的关系,第1执行装置至少执行取得处理,第2执行装置至少执行所述关系评价处理。(Provided are a learning system for a vehicle, a control device for a vehicle, and a learning device for a vehicle, wherein the learning system for a vehicle is provided with a1 st execution device mounted on the vehicle, a2 nd execution device and a storage device outside the vehicle, the storage device stores mapping data that is data of a predetermined mapping and includes data learned by machine learning, the mapping being a mapping in which input data based on a detection value of an in-vehicle sensor is input and an output value is output, the 1 st execution device and the 2 nd execution device execute acquisition processing that acquires the input data as input of the mapping and calculates the output value by cooperation, calculation processing that evaluates a relationship between a predetermined variable other than the variable corresponding to the output value and an accuracy of the output value, the 1 st execution device executes at least the acquisition process, and the 2 nd execution device executes at least the relationship evaluation process.)

1. A learning system for a vehicle is characterized by comprising:

a1 st actuator mounted on a vehicle;

a2 nd actuator external to the vehicle; and

a storage device, wherein,

the storage device is configured to store mapping data that is data specifying a mapping that is input with input data based on detection values of in-vehicle sensors and outputs an output value having information relating to a predetermined state of the vehicle, and that includes data learned by machine learning,

the 1 st execution device and the 2 nd execution device are configured to execute an acquisition process of acquiring the input data, a calculation process of calculating the output value using the input data acquired by the acquisition process as an input of the map, and a relationship evaluation process of evaluating a relationship between a predetermined variable other than the variable corresponding to the output value and the accuracy of the output value in cooperation with each other,

the 1 st execution device is configured to execute at least the acquisition process,

the 2 nd execution device is configured to execute at least the relationship evaluation processing.

2. The learning system for a vehicle according to claim 1,

the predetermined variable includes a variable other than the variable corresponding to the input data.

3. The learning system for a vehicle according to claim 1 or 2,

the mapping is made up of a function approximator that is input with the input data and outputs the output value,

the 2 nd execution device is configured to execute an update process of updating the mapping data based on an evaluation result of the relationship evaluation process,

the update processing includes a division processing that is a processing of dividing the map data utilized into respectively divided data according to the value of the predetermined variable.

4. The learning system for a vehicle according to claim 1 or 2,

the mapping is made up of a function approximator that is input with the input data and outputs the output value,

the storage device is configured to store a plurality of data as the mapping data,

the calculation processing includes selection processing for selecting the mapping data used for the calculation of the output value in accordance with the value of the predetermined variable,

the 2 nd execution device is configured to execute an update process of updating the mapping data based on an evaluation result of the relationship evaluation process,

the update processing includes change processing for changing a relationship between the value of the predetermined variable and the map data.

5. The learning system for a vehicle according to claim 1 or 2,

the 2 nd execution device is configured to execute an update process of updating the mapping data based on an evaluation result of the relationship evaluation process,

the update processing includes addition processing of adding the predetermined variable to the input variable to the map.

6. The learning system for a vehicle according to any one of claims 3 to 5,

the predetermined variable includes a position information variable that is a variable representing position information of the vehicle.

7. The learning system for a vehicle according to any one of claims 3 to 6,

the vehicle is provided with an internal combustion engine,

the detection value of the in-vehicle sensor is a value that detects the state of the internal combustion engine,

the predetermined state is a state of the internal combustion engine.

8. The learning system for a vehicle according to claim 7,

the predetermined variable includes an intake air temperature variable representing a temperature of intake air of the internal combustion engine.

9. The learning system for a vehicle according to claim 7 or 8,

the predetermined variable includes a variable representing atmospheric pressure at a site where the vehicle is located, that is, an atmospheric pressure variable.

10. The learning system for a vehicle according to claim 7,

the predetermined variable includes a rotational speed variable that is a variable indicating a rotational speed of a crankshaft of an internal combustion engine mounted on the vehicle.

11. The learning system for a vehicle according to any one of claims 3 to 10,

the storage device is configured to store 1 st map data that is data defining a1 st map that receives 1 st input data based on detection values of the in-vehicle sensors and outputs 1 st output value that is an output value having information relating to the predetermined state, and store 2 nd map data that is data defining a2 nd map that receives 2 nd input data based on detection values of the in-vehicle sensors and outputs 2 nd output value that is an output value having information relating to the predetermined state,

the 1 st execution device and the 2 nd execution device are configured to execute the following processing in cooperation: a1 st acquisition process of acquiring the 1 st input data based on a detection value of the in-vehicle sensor; 1 st calculation processing for calculating the 1 st output value by using the 1 st input data acquired by the 1 st acquisition processing as an input of the 1 st map; a2 nd acquisition process of acquiring the 2 nd input data based on a detection value of the in-vehicle sensor; a2 nd calculation process of calculating the 2 nd output value by using the 2 nd input data acquired by the 2 nd acquisition process as an input of the 2 nd map; a coincidence determination process of determining whether or not the 1 st output value and the 2 nd output value coincide; and a validity determination process of determining validity of the 2 nd output value based on a determination result of the coincidence determination process,

the relationship evaluation processing includes processing for obtaining a distribution of frequencies at which the validity determination processing determines that the validity is lacking, with respect to the values of the predetermined variables.

12. The learning system for a vehicle according to claim 11,

the 1 st execution device is configured to execute the 1 st acquisition process, the 1 st calculation process, the 2 nd acquisition process, the 2 nd calculation process, the match determination process, and a vehicle-side transmission process of transmitting the 2 nd input data determined to be a mismatch by the match determination process,

the 2 nd execution device is configured to execute the off-vehicle reception process, the relationship evaluation process, and the update process, which receive the data transmitted by the vehicle-side transmission process.

13. The learning system for a vehicle according to claim 12,

the 2 nd execution device is configured to execute an update data transmission process of transmitting the 2 nd mapping data updated by the update process,

the 1 st execution device is configured to execute update data reception processing for receiving the data transmitted by the update data transmission processing.

14. A vehicle control device is characterized by comprising:

an execution device mounted on a vehicle; and

a storage device, wherein,

the storage device is configured to store 1 st map data that is data defining a1 st map that receives 1 st input data based on detection values of in-vehicle sensors and outputs 1 st output value that is an output value having information relating to a predetermined state, and store 2 nd map data that is data defining a2 nd map that receives 2 nd input data based on detection values of the in-vehicle sensors and outputs 2 nd output value that is an output value having information relating to the predetermined state,

the 1 st mapping is composed of a function approximator that is input with the 1 st input data and outputs the 1 st output value,

the 2 nd mapping is composed of a function approximator that is input with the 2 nd input data and outputs the 2 nd output value,

the execution device is configured to execute: a1 st acquisition process of acquiring the 1 st input data based on a detection value of the in-vehicle sensor; 1 st calculation processing for calculating the 1 st output value by using the 1 st input data acquired by the 1 st acquisition processing as an input of the 1 st map; a2 nd acquisition process of acquiring the 2 nd input data based on a detection value of the in-vehicle sensor; a2 nd calculation process of calculating the 2 nd output value by using the 2 nd input data acquired by the 2 nd acquisition process as an input of the 2 nd map; a coincidence determination process of determining whether or not the 1 st output value and the 2 nd output value coincide; and a vehicle-side transmission process of transmitting the 2 nd input data when it is determined by the coincidence determination process that there is no coincidence.

15. A learning device for a vehicle is characterized in that,

an execution device provided outside a vehicle, the execution device being configured to execute: a1 st acquisition process of acquiring 1 st input data based on a detection value of an in-vehicle sensor in the vehicle; a1 st calculation process of calculating a1 st output value by using the 1 st input data acquired by the 1 st acquisition process as an input of a1 st map; a2 nd acquisition process of acquiring 2 nd input data based on a detection value of the in-vehicle sensor; a2 nd calculation process of calculating a2 nd output value by using the 2 nd input data acquired by the 2 nd acquisition process as an input of a2 nd map; a coincidence determination process of determining whether or not the 1 st output value and the 2 nd output value coincide; an off-vehicle reception process of receiving the 2 nd input data transmitted from the vehicle when it is determined by the coincidence determination process that there is no coincidence; a validity determination process of determining validity of the 2 nd output value based on a determination result of the coincidence determination process; a relation evaluation process of evaluating a relation between predetermined variables other than the variable corresponding to the 2 nd output value and the accuracy of the 2 nd output value; and an update process of updating the 2 nd mapping data based on an evaluation result of the relationship evaluation process,

the relationship evaluation processing includes processing for obtaining a distribution of frequencies at which the validity determination processing determines that the validity is lacking with respect to the values of the predetermined variables,

the update process includes a division process that is a process of dividing the utilized 2 nd mapping data into respectively divided data according to the value of the predetermined variable.

Technical Field

The invention relates to a learning system for a vehicle, a control device for a vehicle, and a learning device for a vehicle.

Background

For example, japanese patent application laid-open No. 4-91348 proposes a device including a neural network (neural network) that receives a rotational fluctuation amount, which is a change amount of a rotational speed, and outputs a value indicating whether or not a misfire (misfire) has occurred in each of a plurality of cylinders of an internal combustion engine.

Disclosure of Invention

In general, in order to improve the reliability of a learning completion model learned by machine learning, it is necessary to perform learning in advance using training data in various situations. However, before being mounted on a vehicle, it is not always possible to obtain sufficient training data for various situations that may occur when actually mounted on the vehicle. In addition, when sufficient training data is not available, it is difficult to verify whether or not the neural network outputs an accurate value in various situations when the neural network is mounted on a vehicle.

The present invention provides a learning system for a vehicle, a control device for a vehicle, and a learning device for a vehicle for solving the above problems.

A learning system for a vehicle according to claim 1 of the present invention includes a1 st execution device mounted on a vehicle, a2 nd execution device outside the vehicle, and a storage device. The storage device stores mapping data that is data defining a mapping that is inputted with input data based on detection values of in-vehicle sensors and outputs an output value having information relating to a predetermined state of a vehicle and includes data learned by machine learning, the 1 st execution device and the 2 nd execution device execute acquisition processing, calculation processing, and relationship evaluation processing in cooperation, the acquisition processing being processing for acquiring the input data, the calculation processing being processing for calculating the output value with the input data acquired by the acquisition processing as input of the mapping, the relationship evaluation processing being processing for evaluating a relationship between a predetermined variable other than a variable corresponding to the output value and an accuracy of the output value, the 1 st execution device executing at least the acquisition processing, the 2 nd execution means executes at least the relationship evaluation processing.

According to the above-described aspect 1, the 2 nd execution device evaluates the relationship between the accuracy of the output value based on the map of the input data acquired by the 1 st execution device as the vehicle travels and a predetermined variable other than the variable corresponding to the output value. Therefore, the influence of the predetermined variable on the accuracy of the output value can be verified under various conditions that occur with the operation of the vehicle.

In the above-described claim 1, the predetermined variable may include a variable other than a variable corresponding to the input data.

According to the above-described aspect 1, since the relationship between the precision of the output value of the map and variables other than the variables corresponding to the input data is evaluated, it is possible to verify whether or not the other variables should be included in the input variables to the map, whether or not the map to be used should be divided according to the values of the other variables, whether or not the method of dividing the map to be used is appropriate when the map to be used is divided according to the values of the other variables, and the like.

In the above-described claim 1, the mapping may be configured by a function approximator that receives the input data and outputs the output value, the 2 nd execution device may execute an update process that updates the mapping data based on an evaluation result of the relationship evaluation process, and the update process may include a division process that divides the mapping data used into separate (independent) data according to a value of the predetermined variable.

In the case where a predetermined variable affects the accuracy of an output value, if it is assumed that the predetermined variable is not included in input variables to the function approximator, it is considered to add the predetermined variable to the input variables again in order to improve the accuracy. If it is assumed that a predetermined variable is included in the input variable to the function approximator, for example, when the function approximator is a neural network, it is considered that the influence of the predetermined variable on the output value can be grasped with higher accuracy by a method of complicating the structure of the function approximator in addition to the input dimension, such as increasing the number of layers in the intermediate layers. However, in any of these cases, the structure of the map becomes complicated, and the load of calculation required for calculating the output value may increase.

Therefore, according to the above-described aspect 1, by executing the division processing for dividing the map used into the respective divided maps according to the value of the predetermined variable, it is possible to alleviate the requirement for the sensitivity corresponding to the relationship between the value of the predetermined variable and the output value for each of the maps used, and further it is possible to improve the accuracy of the output value while suppressing the complication of the structure of the map.

In the above-described aspect 1, the mapping may be configured by a function approximator that receives the input data and outputs the output value, the storage device may store a plurality of data as the mapping data, the calculation process may include a selection process of selecting the mapping data used for calculating the output value based on a value of the predetermined variable, the 2 nd execution device may execute an update process of updating the mapping data based on an evaluation result of the relationship evaluation process, and the update process may include a change process of changing a relationship between a value of the predetermined variable and the mapping data.

In the case where a predetermined variable affects the accuracy of an output value, a method of increasing the number of layers of an intermediate layer, for example, in the case where the function approximator is a neural network, or in the case where a predetermined variable is included in an input variable of the function approximator, is considered to improve the accuracy. However, in any of these cases, the structure of the map becomes complicated, and the load of calculation required for calculating the output value may increase.

In this regard, according to the above-described aspect 1, by selecting the map to be used based on the value of the predetermined variable, it is possible to reduce the computational load for calculating the output value and improve the accuracy of the output value, as compared with the case where a single map is used regardless of the value of the predetermined variable. Further, according to the above-described aspect 1, by changing the relationship between the value of the predetermined variable and the map data used for calculating the output value based on the verification of the relationship between the accuracy of the output value calculated as the vehicle travels and the value of the predetermined variable, the relationship between the map used and the value of the predetermined variable can be updated to a more appropriate relationship.

In the above-described claim 1, the 2 nd execution device may execute an update process of updating the map data based on an evaluation result of the relationship evaluation process, and the update process may include an addition process of adding the predetermined variable to an input variable to the map.

When the accuracy of the output value is affected by the value of the predetermined variable, it is considered that the correlation between the value of the predetermined variable and the output value is large. Therefore, according to the above-described aspect 1, by executing the addition processing, it is possible to add a predetermined variable to the input variable when it is determined that the correlation is large based on the evaluation result of the relationship evaluation processing. Further, the output value of the map can be calculated based on the value of the predetermined variable, and thus the output value can be calculated with higher accuracy than a case where the predetermined variable is not used as the input variable.

In the above-described claim 1, the predetermined variable may include a position information variable that is a variable indicating position information of the vehicle.

The situation in which the vehicle is located may vary depending on the region, and may affect the accuracy of the output value due to various factors. Therefore, according to the above-described means 1, by evaluating the relationship between the positional information variable and the accuracy of the output value, it is possible to verify whether or not the accuracy of the output value has changed due to various factors specific to the region, and it is possible to update the mapping data based on the verification result.

In the above-described claim 1, the vehicle may be provided with an internal combustion engine, the detection value of the in-vehicle sensor may be a value that detects a state of the internal combustion engine, and the predetermined state may be the state of the internal combustion engine.

In the above-described claim 1, the predetermined variable may include an intake air temperature variable indicating a temperature of intake air of the internal combustion engine.

The combustion state of the air-fuel mixture in the combustion chamber of the internal combustion engine and the like change depending on the intake air temperature. Therefore, according to the above-described aspect 1, by including the intake air temperature in the evaluation target of the relationship with the accuracy of the output value, it is possible to verify whether or not the accuracy of the output value changes depending on the intake air temperature, and it is possible to update the map data based on the verification result.

In the above-described claim 1, the predetermined variable may include an atmospheric pressure variable that is a variable indicating an atmospheric pressure at a location where the vehicle is located.

The atmospheric pressure may affect the combustion state of the air-fuel mixture in the combustion chamber due to, for example, the pressure in the exhaust passage varying depending on the atmospheric pressure. Therefore, according to the above-described means 1, by including atmospheric pressure in the evaluation target in the relationship with the accuracy of the output value, it is possible to verify whether or not the accuracy of the output value changes depending on the atmospheric pressure, and it is possible to update the map data based on the verification result.

In claim 1, the predetermined variable may include a rotational speed variable that is a variable indicating a rotational speed of a crankshaft of an internal combustion engine mounted on the vehicle.

For example, since the control of the internal combustion engine is generally performed according to an operating point defined by the rotation speed and the load, the combustion control of the internal combustion engine tends to differ according to the rotation speed. Therefore, according to the above-described aspect 1, by including the rotation speed in the evaluation target of the relationship with the accuracy of the output value, it is possible to verify whether or not the accuracy of the output value changes depending on the rotation speed, and it is possible to update the map data based on the verification result.

In the above-described aspect 1, the storage device may store 1 st map data that is data defining a1 st map that receives 1 st input data based on detection values of the in-vehicle sensors and outputs 1 st output values that are output values having information relating to the predetermined state, and store 2 nd map data that is data defining a2 nd map that receives 2 nd input data based on detection values of the in-vehicle sensors and outputs 2 nd output values that are output values having information relating to the predetermined state, and the 1 st execution device and the 2 nd execution device may cooperate to execute: a1 st acquisition process of acquiring the 1 st input data based on a detection value of the in-vehicle sensor; 1 st calculation processing for calculating the 1 st output value by using the 1 st input data acquired by the 1 st acquisition processing as an input of the 1 st map; a2 nd acquisition process of acquiring the 2 nd input data based on a detection value of the in-vehicle sensor; a2 nd calculation process of calculating the 2 nd output value by using the 2 nd input data acquired by the 2 nd acquisition process as an input of the 2 nd map; a coincidence determination process of determining whether or not the 1 st output value and the 2 nd output value coincide; and a validity determination process of determining validity of the 2 nd output value based on a determination result of the matching determination process, wherein the relationship evaluation process may include a process of obtaining a frequency distribution in which the validity determination process determines that the validity is deficient with respect to the value of the predetermined variable.

According to the above-described claim 1, the validity of the 2 nd output value can be determined by the determination process of whether or not the 1 st output value matches the 2 nd output value. Further, according to the above-described means 1, by obtaining a distribution of frequencies determined to be deficient in the validity, it is possible to appropriately evaluate the relationship between the accuracy of the predetermined variable and the accuracy of the output value.

In the above-described claim 1, the 1 st execution device may execute the 1 st acquisition process, the 1 st calculation process, the 2 nd acquisition process, the 2 nd calculation process, the coincidence determination process, and the vehicle-side transmission process for transmitting the 2 nd input data determined to be out of coincidence by the coincidence determination process, and the 2 nd execution device may execute the off-vehicle reception process for receiving the data transmitted by the vehicle-side transmission process, the relationship evaluation process, and the update process.

In the above-described claim 1, the 2 nd execution device may execute update data transmission processing for transmitting the 2 nd mapping data updated by the update processing, and the 1 st execution device may execute update data reception processing for receiving data transmitted by the update data transmission processing.

According to the above-described aspect 1, the 2 nd map data can be updated on the vehicle side by the update data reception processing.

A vehicle control device according to claim 2 of the present invention includes an execution device mounted on a vehicle, and a storage device. The storage device is configured to store 1 st mapping data which is data defining a1 st mapping and to store 2 nd mapping data which is data defining a2 nd mapping, the 1 st map having 1 st input data based on detection values of in-vehicle sensors as input and outputting a1 st output value as an output value having information on a predetermined state, the 2 nd map having 2 nd input data based on detection values of the in-vehicle sensors as input and outputting a2 nd output value as an output value having information on the predetermined state, the 1 st mapping is composed of a function approximator that is input with the 1 st input data and outputs the 1 st output value, the 2 nd map is configured by a function approximator that receives the 2 nd input data and outputs the 2 nd output value, and the execution device is configured to execute: a1 st acquisition process of acquiring the 1 st input data based on a detection value of the in-vehicle sensor; 1 st calculation processing for calculating the 1 st output value by using the 1 st input data acquired by the 1 st acquisition processing as an input of the 1 st map; a2 nd acquisition process of acquiring the 2 nd input data based on a detection value of the in-vehicle sensor; a2 nd calculation process of calculating the 2 nd output value by using the 2 nd input data acquired by the 2 nd acquisition process as an input of the 2 nd map; a coincidence determination process of determining whether or not the 1 st output value and the 2 nd output value coincide; and a vehicle-side transmission process of transmitting the 2 nd input data when it is determined by the coincidence determination process that there is no coincidence.

A learning device for a vehicle according to claim 3 of the present invention includes an execution device outside the vehicle. The execution device is configured to execute: a1 st acquisition process of acquiring 1 st input data based on a detection value of an in-vehicle sensor in the vehicle; a1 st calculation process of calculating a1 st output value by using the 1 st input data acquired by the 1 st acquisition process as an input of a1 st map; a2 nd acquisition process of acquiring 2 nd input data based on a detection value of the in-vehicle sensor; a2 nd calculation process of calculating a2 nd output value by using the 2 nd input data acquired by the 2 nd acquisition process as an input of a2 nd map; a coincidence determination process of determining whether or not the 1 st output value and the 2 nd output value coincide; an off-vehicle reception process of receiving the 2 nd input data transmitted from the vehicle when it is determined by the coincidence determination process that there is no coincidence; a validity determination process of determining validity of the 2 nd output value based on a determination result of the coincidence determination process; a relation evaluation process of evaluating a relation between predetermined variables other than the variable corresponding to the 2 nd output value and the accuracy of the 2 nd output value; and update processing of updating the 2 nd mapping data based on an evaluation result of the relationship evaluation processing, the relationship evaluation processing including processing of obtaining a distribution of frequencies at which the validity determination processing determines that the validity is lacking with respect to the value of the predetermined variable, the update processing including division processing of dividing the 2 nd mapping data used into respective pieces of data according to the value of the predetermined variable.

Drawings

Features, advantages, and technical and industrial significance of exemplary embodiments of the present invention will be described below with reference to the accompanying drawings, in which like reference numerals represent like elements, and wherein:

fig. 1 is a diagram showing a configuration of a learning system for a vehicle according to embodiment 1.

Fig. 2 is a flowchart showing steps of processing executed by the control device according to embodiment 1.

Fig. 3 is a flowchart showing steps of processing executed by the control device according to embodiment 1.

Fig. 4 is a flowchart showing steps of processing executed by the system according to embodiment 1.

Fig. 5 is a diagram showing transmission data according to embodiment 1.

Fig. 6 is a flowchart showing steps of processing executed by the system according to embodiment 1.

Fig. 7 is a diagram illustrating an example of the division processing of the map data according to embodiment 1.

Fig. 8 is a diagram illustrating an example of the division processing of the map data according to embodiment 1.

Fig. 9 is a diagram illustrating a process of dividing map data according to embodiment 1.

Fig. 10 is a flowchart showing the procedure of the selection process according to embodiment 2.

Fig. 11 is a flowchart showing steps of processing executed by the system according to embodiment 2.

Fig. 12 is a diagram showing area division of evaluation map data before update.

Fig. 13 is a flowchart showing steps of processing executed by the system according to embodiment 3.

Fig. 14 is a block diagram showing a process executed by the learning system for a vehicle according to embodiment 4.

Fig. 15 is a block diagram showing a process executed by the learning system for a vehicle according to embodiment 5.

Fig. 16 is a block diagram showing a process executed by the learning system for a vehicle according to embodiment 6.

Fig. 17 is a block diagram showing a process executed by the learning system for a vehicle according to embodiment 7.

Detailed Description

Embodiment 1

Hereinafter, embodiment 1 relating to a learning system for a vehicle will be described with reference to the drawings.

In an internal combustion engine 10 mounted on a vehicle VC1 shown in fig. 1, a throttle valve 14 is provided in an intake passage 12. The air taken in from the intake passage 12 flows into the combustion chamber 18 of each of the cylinders #1 to #4 by opening the intake valve 16. Fuel is injected into the combustion chamber 18 through the fuel injection valve 20. In the combustion chamber 18, an air-fuel mixture of air and fuel is used for combustion by spark discharge of the ignition device 22, and energy generated by the combustion is extracted as rotational energy of the crankshaft 24. The air-fuel mixture used for combustion is discharged as exhaust gas to the exhaust passage 28 as the exhaust valve 26 opens. A catalyst 30 having an oxygen storage capacity is provided in the exhaust passage 28. The intake passage 12 and the exhaust passage 28 are connected by an EGR passage 32, and the flow path cross-sectional area of the EGR passage 32 is adjusted by an EGR valve 34.

An input shaft 56 of a transmission 54 can be coupled to the crankshaft 24 of the internal combustion engine 10 via the torque converter 50. The torque converter 50 includes a lock-up clutch 52, and the crankshaft 24 and the input shaft 56 are coupled by engaging the lock-up clutch 52. A drive wheel 60 is mechanically coupled to an output shaft 58 of the transmission 54.

A crank rotor 40 is coupled to the crankshaft 24, and a tooth 42 indicating each of a plurality of rotation angles of the crankshaft 24 is provided on the crank rotor 40. In the present embodiment, 34 teeth 42 are exemplified. The crank rotor 40 is provided with teeth 42 at intervals of substantially 10 ° CA, but is provided with 1-point tooth-missing portion 44, which is a portion where the interval between adjacent teeth 42 is 30 ° CA. This is used to indicate a rotation angle that becomes a reference of the crankshaft 24.

The control device 70 controls the internal combustion engine 10, and operates operation portions of the internal combustion engine 10 such as the throttle valve 14, the fuel injection valve 20, the ignition device 22, and the EGR valve 34 in order to control torque, an exhaust gas component ratio, and the like as control amounts thereof. The control device 70 controls the torque converter 50, and operates the lock-up clutch 52 to control the engagement state of the lock-up clutch 52 as a control amount. The control device 70 controls the transmission 54, and operates the transmission 54 to control the gear ratio as a control amount. Fig. 1 shows the operation signals MS1 to MS6 for the throttle valve 14, the fuel injection valve 20, the ignition device 22, the EGR valve 34, the lock-up clutch 52, and the transmission 54, respectively.

When controlling the control amount, the control device 70 refers to the output signal Scr of the crank angle sensor 80 that outputs pulses at every angular interval between the teeth 42 provided at every 10 ° CA except for the missing teeth 44, the intake air amount Ga detected by the air flow meter 82, and the temperature of the intake air (intake air temperature Ta) taken into the intake passage 12 detected by the intake air temperature sensor 84. The control device 70 refers to the temperature of the exhaust gas flowing into the catalyst 30 (exhaust gas temperature Tex) detected by the exhaust gas temperature sensor 86, the upstream air-fuel ratio Afu detected by the upstream air-fuel ratio sensor 88 provided on the upstream side of the catalyst 30, and the downstream air-fuel ratio Afd detected by the downstream air-fuel ratio sensor 90 provided on the downstream side of the catalyst 30. The control device 70 refers to the temperature of the cooling water (water temperature THW) of the internal combustion engine 10 detected by the water temperature sensor 92, the shift position Vsft of the transmission 54 detected by the shift position sensor 94, and the atmospheric pressure Pa detected by the atmospheric pressure sensor 96. The control device 70 refers to the traveling speed (vehicle speed V) of the vehicle VC1 detected by the vehicle speed sensor 98, the vertical acceleration Gy of the vehicle VC1 detected by the acceleration sensor 100, and the position data Pgps based on the global positioning system (GPS 102).

The control device 70 includes a CPU72, a ROM74, a storage device 76 as an electrically rewritable nonvolatile memory, a communicator 77, and a peripheral circuit 78, and can communicate with each other via a local network 79. The peripheral circuit 78 includes a circuit that generates a clock signal that defines an internal operation, a power supply circuit, a reset circuit, and the like. Further, the storage device 76 stores the utility map data 76a and the evaluation map data 76 b. Here, the utility map data 76a is data actually used for monitoring the misfire of the internal combustion engine 10. In contrast, the evaluation map data 76b is data to be evaluated for reliability, and is not data used for monitoring the misfire of the internal combustion engine 10. The evaluation map data 76b is attached to the control device 70 in a state in which learning by machine learning is performed to some extent.

The control device 70 executes the control of the above-described control amount by the CPU72 executing a program stored in the ROM 74. For example, the CPU72 operates the fuel injection valve 20 such that the upstream air-fuel ratio Afu becomes a target value. In addition, for example, the CPU72 operates the EGR valve 34 in accordance with the operating state of the internal combustion engine 10 to open-loop control the flow rate (EGR amount) of the exhaust gas flowing from the exhaust passage 28 into the intake passage 12 via the EGR passage 32. Further, the ROM74 stores a misfire detection program 74a and a relearning subroutine 74 b. Here, the relearning subroutine 74b is a routine for performing relearning of the evaluation mapping data 76 b.

The communication device 77 is a device for communicating with the data analysis center 120 via the network 110 outside the vehicle VC 1. The data analysis center 120 analyzes data transmitted from a plurality of vehicles VC1, VC 2. The data analysis center 120 includes a CPU122, a ROM124, a storage device 126, a communication device 127, and a peripheral circuit 128, and can communicate with each other via a local area network 129. ROM124 stores a relearning main program 124a, and this relearning main program 124a specifies a process of relearning evaluation map data 76b based on data transmitted from a plurality of vehicles VC1, VC2, ·. Further, the storage device 126 stores data 126a for relearning, which is transmitted from the plurality of vehicles VC1, VC2, · · for relearning the map defined by the evaluation map data 76b, for relearning, in the data 126a for relearning.

Fig. 2 shows a part of the processing realized by the CPU72 executing the misfire detection program 74a stored in the ROM 74. The process shown in fig. 2 is a process that utilizes utility mapping data 76 a. The processing shown in fig. 2 is repeatedly executed at predetermined cycles, for example. In the following, the step number of each process is represented by a numeral denoted by "S" at the head.

In the series of processes shown in fig. 2, the CPU72 first acquires the minute rotation time T30 (S10). The minute rotation time T30 is the time required for the crankshaft 24 to rotate by 30 ° CA, and is calculated by the CPU72 based on the output signal Scr of the crank angle sensor 80. Next, the CPU72 sets the latest minute rotation time T30 obtained in the processing of S10 to the minute rotation time T30[0], and sets the variable "m" of the minute rotation time T30[ m ] to a larger value as the value becomes the past value (S12). That is, "m" is 1, 2, 3, ·, and the minute rotation time T30[ m-1] immediately before the processing of S12 is set to the minute rotation time T30[ m ]. Thus, for example, the minute rotation time T30 obtained by the processing of S10 when the processing of fig. 2 was executed last time becomes the minute rotation time T30[1 ]. Further, of the minute rotation times T30[0], T30[1], T30[2], · · chronologically adjacent minute rotation times T30 mutually represent the time required for rotation at mutually adjacent angular intervals of 30 ° CA, the angular intervals having no overlapping portions.

Next, the CPU72 determines whether or not the minute rotation time T30 acquired in the process of S10 is the time required for rotation at an angular interval of 30 ° CA to compression top dead center before compression top dead center of any of the cylinders #1 to #4 (S14). When the CPU72 determines that it is the time required for the rotation at the angular interval to the compression top dead center (YES in S14)), the CPU substitutes "T30 [0] -T30[6 ]" for the rotational fluctuation amount Δ ω (i) of the cylinder # i to be determined in order to determine whether or not the cylinder having the compression top dead center is misfiring (S16). That is, the amount of rotational variation Δ ω is quantified by subtracting the time required for the rotation of the angular interval from 30 ° CA before compression top dead center to compression top dead center of the cylinder to be judged for misfire, which is one cylinder before the cylinder to be judged for misfire, to 30 ° CA before compression top dead center to compression top dead center.

Next, the CPU72 determines whether or not the rotational fluctuation amount Δ ω (i) is equal to or greater than a predetermined amount Δ ω th (S18). This process is a process of determining whether or not misfire has occurred in the cylinder to be determined as the misfire. Here, the CPU72 variably sets the predetermined amount Δ ω th based on the rotation speed NE and the charging efficiency η that define the operating point of the internal combustion engine 10.

Specifically, the CPU72 performs the mapping operation on the predetermined amount Δ ω th in a state where the memory device 76 stores mapping data in advance, the mapping data having the rotational speed NE and the inflation efficiency η as input variables and having the predetermined amount Δ ω th as an output variable. The mapping data is a data set of discrete values of the input variable and values of the output variable corresponding to the values of the input variable. The mapping operation may be, for example, the following processing: when the values of the input variables and the values of the input variables of the map data match, the values of the output variables of the corresponding map data are used as the calculation results, whereas when the values do not match, the values obtained by interpolation processing of the values of the plurality of output variables included in the map data are used as the calculation results.

The rotation speed NE is calculated by the CPU72 based on the output signal Scr of the crank angle sensor 80. Here, the rotation speed NE is an average value of rotation speeds at an angular interval where the crankshaft 24 rotates greater than the appearance interval of the compression top dead center (180 ° CA in the present embodiment). Preferably, the rotation speed NE is an average value of rotation speeds when the crankshaft 24 rotates by a rotation angle of 1 rotation or more of the crankshaft 24. The average value here is not limited to a simple average value, and may be, for example, an exponential moving average process, or in short, a low-frequency component from which a high-order component that fluctuates by the appearance interval of the compression top dead center is removed may be calculated. The charging efficiency η is calculated by the CPU72 based on the rotation speed NE and the intake air amount Ga.

The processing of S16 and S18 is processing using the utility map data 76 a. That is, the utility map data 76a defines a map in which the minute rotation time T30[0] and the minute rotation time T30[6] are input, and a logical value corresponding to whether or not a misfire occurred in the cylinder to be determined is output as an output value. The logical value here is a value related to whether the proposition that the rotational fluctuation amount Δ ω (i) is equal to or greater than the predetermined amount Δ ω th is true or false.

When the CPU72 determines that the engine misfire occurred in the cylinder # i (S18: YES), it determines that the engine misfire occurred in the cylinder # i (S20). Subsequently, the CPU72 increments the misfire count cn (i) of the cylinder # i (S22). Then, the CPU72 determines whether the logical or of "a predetermined period has elapsed since the process of S18 was first executed in the state where the misfire count cn (i) was initialized" and "a predetermined period has elapsed since the process of S28 described later" is true (S24). If the CPU72 determines that the logical or is true (yes in S24), it determines whether or not the misfire count cn (i) is equal to or greater than the threshold CNth (S26). When the CPU72 determines that the value is smaller than the threshold CNth (NO in S26), the CPU72 initializes the misfire count cn (i) (S28).

On the other hand, if the CPU72 determines that the threshold CNth is equal to or greater than the threshold CNth (yes in S26), it operates the warning lamp 104 shown in fig. 1 to notify the user of the occurrence of an abnormality (S30).

When the process of S28 or S30 is completed or when the process of S14 or S24 determines no, the CPU72 once ends the series of processes shown in fig. 2. Fig. 3 shows a part of the processing performed by the CPU72 executing the misfire detection program 74a stored in the ROM 74. The processing shown in fig. 3 is processing using the evaluation mapping data 76 b.

In the series of processing shown in fig. 3, the CPU72 first acquires the minute rotation time T30(1), T30(2), · · T30(24), the rotation speed NE, and the inflation efficiency η (S40). Here, the minute rotation times T30(1), T30(2),. cndot.are different from the minute rotation times T30[1], T30[2 ]. cndot.of fig. 2, and in particular, the larger the numbers in parentheses are, the later the values are. The minute rotation times T30(1) to T30(24) are rotation times at 24 angular intervals obtained by equally dividing a rotation angle region of 720 ° CA by 30 ° CA.

Next, the CPU72 substitutes the values obtained in the processing of S40 into the input variables x (1) to x (26) of the map defined by the evaluation map data 76b (S42). Specifically, assuming that "s" is 1 to 24 ", the CPU72 substitutes the minute rotation time T30(s) into the input variable x(s). That is, the input variables x (1) to x (24) are time series data of the minute rotation time T30. The CPU72 substitutes the rotation speed NE into the input variable x (25) and substitutes the inflation efficiency η into the input variable x (26).

Next, the CPU72 calculates the values of the misfire variables P (1) to P (5) by inputting the input variables x (1) to x (26) to the map defined by the evaluation map data 76b (S44). Here, if "i" is 1 to 4 ", the misfire variable p (i) is a variable having a larger value when the probability that a misfire occurs in the cylinder # i is high than when the probability is low. The misfire variable P (5) is a variable that has a larger value when the probability that no misfire occurs in any of the cylinders #1 to #4 is high than when the probability is low.

Specifically, the map defined by the evaluation map data 76b is a neural network having an intermediate layer of 1. The neural network includes coefficients w (1) ji (j is 0 to n, i is 0 to 26) and an activation function h1(x), and the activation function h1(x) is a nonlinear map in which outputs of linear maps defined by the coefficients w (1) ji are each nonlinearly transformed. In the present embodiment, a hyperbolic tangent function is exemplified as the activation function h1 (x). w (1) j0, etc. are bias parameters, and the input variable x (0) is defined as "1".

The neural network includes coefficients w (2) kj (k is 1 to 5, and j is 0 to n) and a softmax function that takes prototype variables y (1) to y (5) as outputs of a linear map defined by the coefficients w (2) kj as inputs and outputs misfire variables P (1) to P (5).

Next, the CPU72 specifies the largest variable among the misfire variables P (1) to P (5) (S46). Then, the CPU72 determines whether the largest misfire variable P (q) is any one of the misfire variables P (1) to P (4) or the misfire variable P (5) (S48). When the CPU72 determines that the misfire variable P (1) to P (4) is any one of the misfire variables (S48: YES), it determines that a misfire occurred in the cylinder # q (S50).

When the process at S50 is completed or the process at S48 is determined as no, the CPU72 once ends the series of processes shown in fig. 3. Fig. 4 shows a part of the procedure of the process related to the relearning of the evaluation map data 76b according to the present embodiment. The process shown on the left side of fig. 4 is realized by the CPU72 executing the relearning subroutine 74b shown in fig. 1, which is stored in the ROM 74. In addition, the processing shown on the right side of fig. 4 is realized by the CPU122 executing the relearning main program 124a stored in the ROM 124.

In a series of processing shown on the left side of fig. 4, the CPU72 first determines whether or not it is during verification to evaluate the reliability of the map data 76b (S60). Specifically, in the present embodiment, the following period is referred to as a verification period.

(a) The period in which the water temperature THW is equal to or lower than a predetermined temperature: when the water temperature THW is low, combustion is likely to become unstable, and it is difficult to improve the misfire detection accuracy as compared with the case where the water temperature THW is high, so this period is included in the verification period.

(b) A period in which the intake air temperature Ta is equal to or lower than a predetermined temperature: when the intake air temperature Ta is low, combustion is likely to become unstable, and it is difficult to improve the misfire detection accuracy as compared with the case where the intake air temperature Ta is high, so this period is included in the verification period.

(c) During the execution of the preheating process of the catalyst 30: during the execution period of the warm-up process of the catalyst 30, since combustion is performed with reduced combustion efficiency, the combustion is likely to be unstable, and it is difficult to improve the detection accuracy of misfire as compared with after warm-up of the catalyst 30, and this period is included in the verification period.

(d) The period in which the charging efficiency η is equal to or less than a predetermined value: in the light load, the combustion is likely to be unstable compared to the case where the load is high, and it is difficult to improve the misfire detection accuracy compared to the medium and high loads, so this period is included in the verification period.

(e) A period during which the change amount Δ NE of the rotation speed NE per predetermined time is equal to or greater than a predetermined value: in the transient operation, the misfire detection accuracy is likely to be lower than that in the steady operation, and therefore this period is included in the verification period.

If the CPU72 determines that it is during verification (S60: yes), it determines whether the flag F is "1" (S62). Here, the flag F is "1" when the result of the misfire determination based on the processing shown in fig. 2 does not match the result of the misfire determination based on the processing shown in fig. 3, and is "0" when the result of the misfire determination matches the result of the misfire determination based on the processing shown in fig. 3. When the flag F is determined to be "0" (S62: no), the CPU72 determines whether or not the result of the misfire determination based on the processing shown in fig. 2 and the result of the misfire determination based on the processing shown in fig. 3 do not match (S64). In a case where the determination results of 4 times by the processing of S18 of fig. 2 in the same combustion cycle do not match the determination result by the processing of S46 of fig. 3, the CPU72 determines that they do not match. That is, for example, when it is determined in the process of S18 that the amount of rotational fluctuation Δ ω (1) of the cylinder #1 is equal to or greater than the predetermined amount Δ ω th, but P (5) is selected in the process of S46, the CPU72 determines that the amounts of rotational fluctuation Δ ω (1) and P (5) do not match.

If the CPU72 determines that there is no match (yes in S64), the CPU72 substitutes "1" for the flag F (S66). Next, the CPU72 increments the count C (S68). On the other hand, if the CPU72 determines that the flag F is "1" (S62: YES), it determines whether or not the result of misfire determination based on the processing shown in FIG. 2 and the result of misfire determination based on the processing shown in FIG. 3 match (S70). If it is determined that the signals do not match (no in S70), the CPU72 proceeds to the process of S68, and if it is determined that the signals match (yes in S70), the CPU substitutes "0" for the flag F (S72). Then, the CPU72 determines whether the count C is larger than the maximum value C0 (S74). If the CPU72 determines that the value is larger than the maximum value C0 (S74: yes), the maximum value C0 is updated to the current value of the count C, and the rotation time set GrT30 and the extra information set GrE are updated (S76).

Specifically, as shown in fig. 5, the rotation time set GrT30 is a set of minute rotation times T30(1) to T30(72) of 3 combustion cycles. However, the minute rotation times T30(49) to T30(72) are updated to correspond to the combustion cycle in which the result of determination of misfire based on the processing shown in fig. 2 and the result of determination of misfire based on the processing shown in fig. 3 are determined to match by the latest processing of S70. Here, when the maximum value C0 is "2" or more, the minute rotation time T30(1) to T30(24) and the minute rotation time T30(25) to T30(48) correspond to a combustion cycle in which the result of determination of misfire based on the processing shown in fig. 2 does not match the result of determination of misfire based on the processing shown in fig. 3. Further, the initial value of the maximum value C0 is zero.

The extra information set GrE is made up of the rotation speed NE, the inflation efficiency η, the combustion control mode value MC, the intake air temperature Ta, the atmospheric pressure Pa, the shift stage Vsft of the transmission 54, the engagement state value Vrc, the value of a variable indicating the state of the road surface on which the vehicle VC1 is traveling (road surface state value SR), and the value of a position variable that is a variable indicating the position of the vehicle VC1 (position data Pgps). Here, the engagement state value Vrc is a logical value of a variable indicating the engagement state of the lock-up clutch 52, and is "1" when the lock-up clutch is in the engagement state and is "0" when the lock-up clutch is not in the engagement state. Further, road surface state value SR becomes "1" when the unevenness of the road surface on which vehicle VC1 runs is conspicuous, and becomes "0" when the unevenness is inconspicuous. Note that the road surface state value SR may be "1" when the average value of the absolute value of the acceleration Gy in a predetermined period is equal to or greater than a predetermined value. Preferably, each data constituting the extra information set GrE is a value in a combustion cycle before the combustion cycle determined as yes in the process of S70.

Returning to fig. 4, if the process at S76 is completed or the process at S74 is determined as no, the CPU72 initializes the count C (S78). When the processing of S68 and S78 is completed or when the processing of S60 and S64 determines no, the CPU72 determines whether the end of the trip is reached (S80). Here, the trip refers to 1 time period during which the vehicle running permission signal is in the on state. In the present embodiment, the running permission signal corresponds to an ignition signal. When the CPU72 determines that the trip is ended (yes in S80), the operation communicator 77 transmits the information "q" relating to the largest variable among the misfire variables P (1) to P (5), the maximum value C0, the rotation time set GrT30, and the extra information set GrE to the data analysis center 120 (S82). Then, the CPU72 initializes the maximum value C0 (S84), and temporarily ends a series of processing shown on the left side of fig. 4. If the CPU72 determines no in the process of S80, the process shown on the left side of fig. 4 is also terminated once.

In contrast, as shown on the right side of fig. 4, the CPU122 receives the information "q" relating to the largest variable among the misfire variables P (1) to P (5), the maximum value C0, the rotation time set GrT30, and the additional information set GrE (S90). Then, the CPU122 displays the waveform data relating to the rotational behavior of the crankshaft 24 represented by the rotation time set GrT30 on the display device 112 shown in fig. 1, and displays the maximum value C0, the extra information set GrE (S92). This is a process of providing information to a skilled person that the skilled person can determine whether or not a misfire occurred. That is, if the user is skilled, the presence or absence of misfire can be determined with high accuracy by looking at the waveform data. At this time, by referring to the information of the extra information group GrE, it becomes more reliable to determine whether or not the misfire occurred. Thus, the skilled person can determine whether or not the misfire determination using the evaluation map data 76b is an erroneous determination based on the determination of whether or not the misfire occurred.

When the judgment result is input by the skilled person operating the interface 114 shown in fig. 1, the CPU122 acquires the judgment result (S94). Then, the CPU122 determines whether or not the determination result input by the operation of the interface 114 is "the misfire determination using the evaluation map data 76b is an erroneous determination" (S96). If it is determined that the determination is "false determination" (yes in S96), the CPU122 stores the minute rotation time T30(25) to T30(48), the extra information set GrE, and the determination result of whether or not the misfire occurred by the skilled person in the data received in the processing of S90 as the data 126a for re-learning (S98). The data 126a for re-learning includes data based on data received not only from the vehicle VC1 but also from another vehicle VC2, which is equipped with an internal combustion engine of the same specification as the internal combustion engine 10.

If it is determined in S96 that the process is no or S98 is completed, CPU122 once ends the series of processes shown on the right side of fig. 4. Through the above processing, the input data and the data related thereto in the case where the evaluation map data 76b is erroneously determined are stored in the storage device 126 as the data 126a for relearning. Then, when the data 126a for re-learning becomes a predetermined amount or more, the CPU122 executes the processing shown in fig. 6.

Fig. 6 shows a part of the procedure of the process related to the relearning of the evaluation map data 76b according to the present embodiment. The process shown on the left side of fig. 6 is realized by the CPU72 executing the relearning subroutine 74b stored in the ROM74 shown in fig. 1. In addition, the processing shown on the right side of fig. 6 is realized by the CPU122 executing the relearning main program 124a stored in the ROM 124. In the following, the processing shown in fig. 6 will be described in terms of the sequence of the processing related to the relearning of the evaluation map data 76 b.

In a series of processing shown on the right side of fig. 6, CPU122 first classifies re-learning data 126a for each region where the corresponding vehicle is located when transmitting information, and measures the number of data whose output of the map defined by evaluation map data 76b is an erroneous output for each region after the classification, thereby creating erroneous output distribution data for each region (S100). The data that is erroneously output is determined to be erroneously determined in the processing at S96. In the present embodiment, the region is defined by a country or a region smaller than the country. For example, when the land area is equal to or smaller than a predetermined value, areas such as korea and taiwan are set as one division. In contrast, in a case of a country having a large land area, for example, a tibetan autonomous region, a sichuan province, or a province that divides the country is set as one division.

This process is a process for verifying whether there is a difference in the misfire determination accuracy due to a difference in the conditions in which the vehicles VC are located in each region. That is, it is verified whether, for example, there is a case where: the accuracy of the determination process using the evaluation map data 76b in a specific region is lower than that in other regions due to the locality of the nature of the fuel supplied to the internal combustion engine 10 and some unexpected locality.

Then, CPU122 determines whether or not there is a region in which the error output rate is equal to or greater than threshold Rth1 in the regions divided by the processing of S100 (S102). When it is determined that there is a region equal to or greater than the threshold Rth1 (S102: yes), CPU122 divides the region using the evaluation mapping data 76b into a region equal to or greater than the threshold Rth1 and other regions (S104). That is, it is determined to update the evaluation map data 76b used in the area equal to or greater than the threshold Rth1 and the evaluation map data 76b used in the other areas.

Fig. 7 illustrates a case where the error output rate is equal to or higher than the threshold Rth1 in the region a 1. In this case, the CPU122 determines to generate evaluation map data 76b dedicated to the area a1 that is used in the area a1 and is not used outside the area a 1.

Returning to fig. 6, when the process of S104 is completed or when it is determined that no is present in the process of S102, the CPU122 classifies the data constituting the data 126a for re-learning into regions divided by the intake air temperature Ta and the atmospheric pressure Pa, and creates erroneous output distribution data by measuring the number of data erroneously output for each of the classified regions (S106). Here, when the process of S104 has already been executed, error output distribution data in a region divided by the intake air temperature Ta and the atmospheric pressure Pa is created for a region determined to be equal to or higher than the threshold Rth1 in the process of S102 and for other regions.

This process is a process for verifying whether or not a difference in misfire determination accuracy occurs due to a difference in intake air temperature Ta and atmospheric pressure Pa. That is, when the intake air temperature Ta is low, the combustion of the internal combustion engine 10 is likely to become unstable as compared with the case where the intake air temperature Ta is high, and it is verified whether the detection accuracy of the misfire is reduced. Further, when the atmospheric pressure Pa is different, the pressure in the exhaust passage 28 is different, and even if the opening degree of the EGR valve 34 is the same, the flow rate of the exhaust gas flowing from the exhaust passage 28 into the intake passage 12 via the EGR passage 32 is different. Therefore, when the EGR valve 34 is operated to open-loop control the flow rate of the exhaust gas flowing from the exhaust passage 28 into the intake passage 12 via the EGR passage 32 as in the present embodiment, the flow rate of the exhaust gas flowing from the exhaust passage 28 into the intake passage 12 via the EGR passage 32 changes according to the atmospheric pressure Pa, and the combustion state differs. It is verified whether the detection accuracy of the misfire decreases according to the combustion state.

Then, the CPU122 determines whether or not there is a region in which the error output rate becomes equal to or greater than the threshold Rth2 in the region partitioned by the intake air temperature Ta and the atmospheric pressure Pa (S108). When determining that there is a region equal to or larger than the threshold Rth2 (S108: yes), the CPU122 divides the region using the evaluation mapping data 76b into a region equal to or larger than the threshold Rth2 and other regions (S110). That is, it is determined to update the evaluation map data 76b used in the region equal to or greater than the threshold Rth2 and the evaluation map data 76b used in the other regions.

For example, as shown in fig. 8, the evaluation map data 76b is determined to be updated separately from the other regions with respect to the region a2 of the regions divided by the intake air temperature Ta and the atmospheric pressure Pa. In addition, when the area determined as being the yes in the processing at S108 is present in the area determined as being the yes in the processing at S102, it may be determined that the evaluation map data 76b is updated for each of the area, which is the area determined as being the yes in the processing at S108 and the other area, and the area other than the area determined as being the yes in the processing at S102. Thus, for example, the evaluation mapping data 76b illustrated in fig. 7 and dedicated to the area a1 is divided into data dedicated to the area a2 and data dedicated to areas other than the area a2 illustrated in fig. 8.

Returning to fig. 6, when the process at S110 is completed or when it is determined that the process at S108 is no, CPU122 classifies the data constituting the data 126a for re-learning by the regions divided by the rotation speed NE and the combustion control mode, and creates erroneous output distribution data by measuring the number of data erroneously output for each of the classified regions (S112). Then, CPU122 determines whether or not there is a region in which the error output rate becomes equal to or greater than threshold Rth3 in the regions divided by the rotation speed NE and the combustion control mode (S114).

This process is a process for verifying whether or not there is a difference in misfire determination accuracy due to the rotation speed NE and the combustion control mode. That is, for example, because the control for reducing the combustion efficiency and increasing the exhaust gas temperature is performed in the warm-up control mode of the catalyst 30, it is verified whether the determination accuracy of the misfire is reduced because the combustion is likely to be unstable as compared with the case where such control is not performed. Further, for example, when the rotation speed NE is low, the rotation is likely to be unstable because the inertial energy of the crankshaft 24 is small, and it is verified whether the misfire determination accuracy is reduced depending on the rotation speed NE due to various operating portions operated based on the rotation speed NE and the inflation efficiency η.

Fig. 9 illustrates a region defined according to the rotation speed NE and the combustion control mode in the present embodiment. In the present embodiment, the mode of the idle operation, the mode of the warming-up control of the catalyst 30, and the other modes are classified according to the combustion control mode value MC, and the other modes are classified into the low rotation speed region and the other regions.

When the process of S104 has already been executed, CPU122 creates erroneous output distribution data for each of the region determined to be equal to or greater than threshold Rth1 in the process of S102 and the other regions. When the process of S110 has already been executed, CPU122 creates erroneous output distribution data for each of the region determined to be equal to or greater than threshold Rth2 in the process of S108 and the other regions.

Returning to fig. 6, if CPU122 determines that there is a region equal to or greater than threshold Rth3 in the region shown in fig. 9 (S114: yes), then, for the region equal to or greater than threshold Rth3, the erroneous output distribution is examined for each shift position Vsft, engagement state value Vrc, inflation efficiency η, and road surface state value SR (S116).

This process is a process for verifying that the accuracy of the misfire determination varies depending on the gear ratio of the transmission 54, the engagement state of the lock-up clutch 52, the inflation efficiency η, and the presence or absence of irregularities on the road surface on which the vehicle VC1 is traveling. That is, since the moment of inertia (moment of inertia) from the crankshaft 24 to the output shaft 58 of the transmission 54 differs depending on the gear ratio of the transmission 54 and the engagement state of the lock-up clutch 52, the rotational behavior of the crankshaft 24 differs, and it is verified whether or not the precision of determining the misfire is lowered at a predetermined moment of inertia. Further, since the rotational behavior of the crankshaft 24 changes according to the irregularities of the road surface, it is verified whether the precision of the misfire determination is significantly reduced when the irregularities are conspicuous as compared with when the vehicle travels on a flat road surface. In addition, when the charging efficiency η is small, the rotational behavior of the crankshaft 24 is likely to become unstable due to the small combustion energy in the combustion chamber 18 as compared with the case where the charging efficiency η is large, and therefore, it is verified whether the precision of determining the misfire is significantly reduced.

Then, the CPU122 divides the area using the evaluation map data 76b into an area where the output error rate is equal to or higher than the threshold Rth3 and other areas (S118). Specifically, it is determined that the evaluation map data 76b used in the region equal to or greater than the threshold Rth3 and the evaluation map data 76b used in the other regions are updated.

The following is illustrated in fig. 9: in a region where the engine speed is low and neither the idle operation mode nor the catalyst warm-up control mode, if the inflation efficiency η is equal to or less than the predetermined value η 1, the shift position Vsft is at the predetermined position, the engagement state value Vrc is at the predetermined value, and the road surface state value SR is "1" region A3, the output error rate exceeds the threshold Rth 3. In addition, when the area A3 exists in the area a1 illustrated in fig. 7, for example, the evaluation mapping data 76b dedicated to the area a1 may be divided into data dedicated to the area A3 and data other than the data. For example, when the area A3 exists in the area a2 illustrated in fig. 8, the evaluation mapping data 76b dedicated to the area a2 may be divided into data dedicated to the area A3 and data other than the data. Further, for example, when the area A3 exists in the area a2 illustrated in fig. 8 of the area a1 illustrated in fig. 7, the dedicated evaluation mapping data 76b belonging to both the area a1 and the area a2 may be divided into data dedicated to the area A3 and data other than the data.

Returning to fig. 6, if the process at S118 is completed or if the process at S114 determines no, CPU122 relearns the evaluation map data 76b for each of the regions divided by the processes at S104, S110, and S118 (S120). That is, for example, in the case of updating the evaluation mapping data 76b used in the area a1, the CPU122 updates the evaluation mapping data 76b using only the data transmitted in the area a1 among the data 126a for relearning. In addition, for example, when the evaluation map data 76b used in the case of belonging to the area a1 and belonging to the area a2 is updated, the evaluation map data 76b is updated using only the data transmitted in the case where the driving environment of the vehicle located in the area a1 belongs to the area a2 among the data 126a for relearning. If the CPU122 determines no in the processing of S102, S108, and S114, it updates the evaluation map data 76b with all the data 126a for relearning.

Specifically, the CPU72 selectively uses the corresponding data 126a for re-learning as training data to update the coefficients w (1) ji, w (2) kj as parameters for which learning has been completed, which are the evaluation map data 76 b. That is, the CPU72 calculates the misfire variables P (1) to P (5) using the minute rotation time T30, the rotation speed NE, and the inflation efficiency η in the training data as input variables x (1) to x (26), and generates teacher data based on data on the determination result of whether or not the skilled person has misfired. For example, if the skilled judgment means that the cylinder #1 is misfiring, P (1) is set to 1 and P (2) to P (5) are set to 0. For example, if the skilled person determines that the determination is normal, P (1) to P (4) are set to 0 and P (5) is set to 1. The coefficients w (1) ji and w (2) kj are updated by a known method so that the absolute value of the difference between the teacher data and the misfire variables P (1) to P (5) output by the neural network is reduced.

In addition, the processing of calculating the misfire variables P (1) to P (5) in the data analysis center 120 requires information on the coefficients w (1) ji and w (2) kj, the activation function h1, and information on the effect that the softmax function is used in the output layer of the neural network. The information may be, for example, an instruction to "transmit data relating to the information" from CPU122 to control device 70 before the processing on the right side of fig. 6, or may be stored in storage device 126 in advance.

Then, CPU122 operates communicator 127 to transmit updated coefficients w (1) ji, w (2) kj as parameters for which relearning is completed to a target vehicle of vehicles VC1, VC2, · (S122). That is, for example, the coefficients w (1) ji, w (2) kj dedicated to the area a1 illustrated in fig. 7 in the updated evaluation map data 76b are selectively transmitted to the vehicle located in the area a1 among the vehicles VC1, VC2, ·. Further, when the process of S122 is completed, the CPU122 temporarily ends a series of processes shown on the right side of fig. 6.

On the other hand, as shown on the left side of fig. 6, the CPU72 determines whether or not the relearning-completed parameter is transmitted from the data analysis center 120 (S130). When the CPU72 determines that there is a parameter for which relearning is completed (yes in S130), it receives the coefficients w (1) ji and w (2) kj (S132), and updates the evaluation map data 76b stored in the storage device 76 (S134).

When the process at S134 is completed or when the process at S130 is determined as no, the CPU72 once ends the series of processes shown on the left side of fig. 6. Here, the operation and effect of the present embodiment will be described.

The CPU72 executes the processing shown in fig. 2 based on the utility map data 76a to monitor the presence or absence of a misfire in the internal combustion engine 10, and if a misfire frequently occurs, executes notification processing to cope with the situation. Further, the CPU72 executes the processing shown in fig. 3 based on the evaluation map data 76b and executes misfire determination based on the evaluation map data 76 b. The CPU72 determines whether or not the misfire determination result using the evaluation map data 76b matches the misfire determination result using the utility map data 76a, and if it is determined that the misfire determination result does not match the utility map data, transmits the position data Pgps, the intake air temperature Ta, the atmospheric pressure Pa, and the like to the data analysis center 120 in addition to the input data for the misfire determination of the evaluation map data 76 b. This is a process for evaluating a case where learning of the mapping data 76b may be insufficient.

On the other hand, the CPU122 displays the input data and the like transmitted from the CPU72 on the display device 112. Thus, the skilled person determines whether or not the misfire occurred based on the waveform data indicating the rotational behavior of the crankshaft 24, and determines whether or not the determination of the presence or absence of the misfire using the evaluation map data 76b is an erroneous determination based on the determination. When the judgment result of the skilled person is a judgment that the presence or absence of the misfire is judged erroneously using the evaluation map data 76b, the CPU122 stores at least a part of the data transmitted from the vehicle side in the storage device 126 as the data 126a for relearning.

When the data 126a for re-learning becomes equal to or more than a predetermined amount, the CPU122 determines whether or not the accuracy of the determination process using the evaluation map data 76b is particularly low in a specific region among the regions partitioned by the position data Pgps, the intake air temperature Ta, the atmospheric pressure Pa, and the like. When there is a particularly low area, CPU122 updates evaluation map data 76b using only the training data relating to the area, thereby generating dedicated evaluation map data 76b to be used when entering the area. This makes it possible to improve the misfire determination accuracy without complicating the structure of the map defined by the evaluation map data 76 b.

That is, for example, when the misfire determination accuracy is particularly low in the region a1 illustrated in fig. 7, it is considered that the determination accuracy is lowered in the region a1 in a specific manner. In contrast, even if a large amount of the data 126a for relearning is accumulated and relearning is performed using all of these data, it is not always easy to improve the misfire determination accuracy without complicating the structure of the evaluation map data 76 b. This is because, if there is a characteristic case in the region a1 in which the misfire determination accuracy is degraded, the optimal values of the coefficients w (1) ji and w (2) kj may differ between the region a1 and the other regions. Of course, if the number of layers of the intermediate layers is increased, or if the number of dimensions input is increased such as by adding the position data Pgps to the input to the map defined by the evaluation map data 76b, the misfire determination accuracy may be improved in both the region a1 and the other regions. However, in this case, the structure of the map defined by the evaluation map data 76b becomes complicated, and the calculation load of the misfire determination process using this map becomes large. In contrast, in the present embodiment, when there is a region where the misfire determination accuracy is low, the dedicated evaluation map data 76b used in the region is generated, whereby the misfire determination accuracy can be improved without complicating the structure of the map defined by the evaluation map data 76 b.

Then, CPU122 transmits updated coefficients w (1) ji and w (2) kj to the corresponding one of vehicles VC1, VC2 · · as data for which relearning is completed. For example, the CPU122 transmits the coefficients w (1) ji, w (2) kj specific to the area a1 to the vehicles located in the area a 1. Thus, in the vehicle located in the area a1, the evaluation map data 76b is updated to the evaluation map data 76b in which the evaluation accuracy is improved in the area a 1.

Then, when it is found by the judgment of the skilled person when the inconsistency occurs by the processing of fig. 4 that the reliability of the evaluation map data 76b is higher than that of the utility map data 76a, the updated evaluation map data 76b can be used as the utility map data 76a for monitoring the misfire in the vehicle to which the updated coefficients w (1) ji and w (2) kj are transmitted. Furthermore, a learning completion model (map data) based on raw data mounted on a plurality of vehicles VC1, VC2, · · may be mounted as practical map data from the beginning on a control device mounted on a newly developed vehicle having an internal combustion engine with the same number of cylinders.

Embodiment 2

Hereinafter, the following description will focus on differences from embodiment 1, and embodiment 2 will be described with reference to the drawings.

In the present embodiment, the evaluation map data 76b is composed of the low rotation speed map data and the high rotation speed map data, and the accuracy of misfire judgment using these data is evaluated. Fig. 10 shows a procedure of the selection processing of the evaluation map data 76 b. The process shown in fig. 10 is realized by the CPU72 repeatedly executing the misfire detection program 74a stored in the ROM74, for example, at predetermined cycles.

In the series of processing shown in fig. 10, the CPU72 determines whether the rotation speed NE is equal to or higher than a predetermined speed NEth (S124). When the CPU72 determines that the speed is equal to or higher than the predetermined speed NEth (yes in S124), it selects the high-speed map data in the evaluation map data 76b (S126). The high rotation speed map data is data obtained by learning input variables x (1) to x (26) when the rotation speed NE is equal to or greater than a predetermined speed NEth as training data. On the other hand, if the CPU72 determines that the speed is less than the predetermined speed NEth (no in S124), the CPU selects the low-rotation-speed map data in the evaluation map data 76b (S128). The low rotation map data is obtained by learning input variables x (1) to x (26) when the rotation speed NE is smaller than the predetermined speed NEth as training data.

When the processing in S126 and S128 is completed, the CPU72 once ends the series of processing shown in fig. 10. Fig. 11 shows a part of the procedure of the process related to the relearning of the evaluation map data 76b according to the present embodiment. The process shown on the left side of fig. 11 is realized by the CPU72 executing the relearning subroutine 74b stored in the ROM74 shown in fig. 1. In addition, the processing shown on the right side of fig. 11 is realized by the CPU122 executing the relearning main program 124a stored in the ROM 124. In the processing shown in fig. 11, for convenience, the same step numbers are assigned to the processing corresponding to the processing shown in fig. 6.

In a series of processing shown on the right side of fig. 11, CPU122 first classifies re-learning data 126a for each rotation speed NE, and measures the number of erroneously determined data for each rotation speed NE classified as such, thereby creating erroneous output distribution data for rotation speed NE (S140). Here, the rotation speed NE for defining the erroneous output distribution has a smaller particle size than a particle size of a binary value of whether or not the rotation speed NE is equal to or greater than the predetermined speed NEth. Next, the CPU72 determines whether or not there is a region in which the output error rate becomes equal to or greater than the threshold Rth4 among the plurality of regions divided by the rotation speed NE (S142).

When the CPU72 determines that there is an area equal to or larger than the threshold Rth4 (S142: yes), the CPU changes the area of the division evaluation mapping data 76b (S144). That is, for example, when the region equal to or higher than the threshold Rth4 is a part of the region smaller than the predetermined speed NEth, the region smaller than the predetermined speed NEth is divided into the region equal to or higher than the threshold Rth4 and the other regions.

When the process at S144 is completed, the CPU72 executes the processes at S120 and S122, and when the process at S142 determines no, the CPU once ends the series of processes shown on the right side of fig. 11.

Fig. 12 shows an example of updating the evaluation mapping data 76b according to the present embodiment. The left side of fig. 12 shows the division of the usage area of the evaluation map data 76b before updating, and either the high rotation speed map data or the low rotation speed map data is used for misfire determination depending on whether or not the usage area is equal to or greater than the predetermined speed NEth. In contrast, the right side of fig. 12 shows the division of the usage region of the updated evaluation map data 76b, and any one of the three map data is used for misfire determination in accordance with the rotation speed NE. In this example, near the predetermined speed NEth, the error output ratio is equal to or greater than the threshold Rth4 in both the misfire determination using the high rotation speed map data and the misfire determination using the low rotation speed map data, and therefore the intermediate rotation speed map data is newly added.

As described above, according to the present embodiment, in the setting using any one of the plurality of maps according to the rotation speed NE, when the setting itself, which is used according to the rotation speed NE, is not satisfactory in maintaining the misfire determination accuracy high, the correspondence relationship itself between the rotation speed NE and the map for misfire determination is changed. Thus, when the map for misfire determination is divided according to the rotation speed NE, a more appropriate division can be found.

Embodiment 3

Hereinafter, the following description will focus on differences from embodiment 1, and embodiment 3 will be described with reference to the drawings.

In the present embodiment, when the error output rate becomes significant in a part of the region divided by the variables with respect to the variables x (1) to x (26) which are not mapped to the evaluation mapping data 76b, the variables are added to the input variables.

Fig. 13 shows a part of the procedure of the process related to the relearning of the evaluation map data 76b according to the present embodiment. The process shown on the left side of fig. 13 is realized by the CPU72 executing the relearning subroutine 74b stored in the ROM74 shown in fig. 1. In addition, the processing shown on the right side of fig. 13 is realized by the CPU122 executing the relearning main program 124a stored in the ROM 124. In the processing shown in fig. 13, for convenience, the same step numbers are assigned to the processing corresponding to the processing shown in fig. 6.

In a series of processing shown on the right side of fig. 13, the CPU122 first classifies the data 126a for relearning for each intake air temperature Ta, and measures the number of erroneously determined data for each intake air temperature Ta after the classification, thereby creating erroneous output distribution data for the intake air temperature Ta (S150). Next, the CPU72 determines whether or not there is a region in which the output error rate becomes equal to or greater than the threshold Rth5 among the plurality of regions divided by the intake air temperature Ta (S152). When the CPU72 determines that there is a region equal to or larger than the threshold Rth5 (S152: yes), it adds a variable relating to the intake air temperature to the input variable of the map defined by the evaluation map data 76b (S154). That is, the input variable is changed from 26 dimensions to 27 dimensions.

When the process at S154 is completed, the CPU72 executes the processes at S120 and S122. In the processing of S120 in the present embodiment, all the data 126a for re-learning are used as training data. Further, for example, a dynamometer (dynameter) or the like may be connected to the crankshaft 24 of the internal combustion engine having the same specification as the internal combustion engine 10, and the fuel injection from the fuel injection valve 20 may be stopped under a predetermined condition to operate the internal combustion engine 10, thereby generating misfire, and training data may be generated on a test bench and additionally used.

Embodiment 4

Hereinafter, the following description will focus on differences from embodiment 1, and embodiment 4 will be described with reference to the drawings.

The practical map data 76a and the evaluation map data 76b according to the present embodiment are data that define a map that outputs values of an unevenness variable Inb that is a variable indicating unevenness between actual air-fuel ratios when the fuel injection valve 20 is operated to control the air-fuel ratios of the air-fuel mixtures in the plurality of cylinders #1 to #4 to be equal to each other.

Fig. 14 shows a process executed by the learning system for a vehicle according to the present embodiment. The processing shown in fig. 14 is processing realized by the CPU72 executing a program stored in the ROM74, and processing realized by the CPU122 executing a program stored in the ROM 124.

The unevenness detection processing M10(1) is processing for calculating the value of the unevenness variable Inb (1) using the practical mapping data 76 a. The unevenness detecting process M10(1) includes a process of calculating a value when the unevenness variable Inb (1) exhibits a rich side based on the amount of change per predetermined time of the upstream air-fuel ratio Afu. The unevenness detecting process M10(1) includes a process of calculating a value when the unevenness variable Inb (1) shows a lean value based on the variation of the minute rotation time T30.

The unevenness detection processing M10(2) is processing for calculating the value of the unevenness variable Inb (2) using the evaluation mapping data 76 b. The unevenness detecting process M10(2) is a process of calculating the value of the unevenness variable Inb (2) using a neural network that receives as input the time series data composed of the minute rotation times T30(1) to T30(24) and the time series data of the upstream air-fuel ratio Afu in the period, and outputs the value of the unevenness variable Inb (2).

The re-learning sub-process M12 includes the following processes: when the absolute value of the difference between the uneven variable Inb (1) and the uneven variable Inb (2) is equal to or greater than a predetermined value, it is determined that the determination result using the practical mapping data 76a does not match the determination result using the evaluation mapping data 76 b. In addition, the sub-process for re-learning M12 includes a process of transmitting predetermined data to the data analysis center 120. Here, the data to be transmitted includes, in addition to the minute rotation time T30 and the upstream air-fuel ratio Afu used for calculating the variation variable Inb (2) when the mismatch occurs, time series data of the minute rotation time T30 and time series data of the upstream air-fuel ratio Afu adjacent to them in time series. This is to enable a skilled person to more accurately determine whether the variation variable Inb (2) is an accurate value based on the rotational behavior of the crankshaft 24 and the behavior of the upstream air-fuel ratio Afu.

The data to be transmitted includes the rotation speed NE and the charging efficiency η, which are values of operating point variables, which are variables indicating the operating point of the internal combustion engine 10, when a mismatch occurs. The value of the operating point variable is data for allowing a skilled person to more accurately judge whether the unevenness variable Inb (2) is a correct value, and is data for verifying whether the accuracy of the unevenness variable Inb (2) differs depending on the operating point.

The data to be transmitted includes the combustion control mode value MC. In the present embodiment, as the combustion control mode value MC, a combustion control mode value MC that identifies a mode corresponding to a value of the target air-fuel ratio, a mode in which control for increasing the exhaust gas temperature is performed for the regeneration process of the catalyst 30, and other modes is exemplified. This is because the influence of the variation in the actual air-fuel ratio among the cylinders on the rotational behavior of the crankshaft 24 and the influence of the upstream air-fuel ratio Afu differs depending on the difference in the combustion control mode. Therefore, the combustion control mode value MC is data for allowing a skilled person to more accurately determine whether or not the unevenness variable Inb (2) is a correct value, and is data for verifying whether or not the accuracy of the unevenness variable Inb (2) varies depending on the value.

The data to be transmitted includes an intake air temperature Ta, atmospheric pressure Pa, shift stage Vsft, engagement state value Vrc, road surface state value SR, and position data Pgps. The significance of these data is the same as in embodiment 1.

The CPU122 executes processing based on the right side of fig. 4 and the right side of fig. 6 as relearning processing M14. Further, as the processing based on the processing of fig. 4, the following processing is included: if the absolute value of the difference between the value of the unevenness variable Inb that is considered appropriate in the data analysis center 120 and the output value of the map defined by the evaluation map data 76b is equal to or greater than a predetermined value, an erroneous output is set. In addition, this embodiment shows an example in which after the error output rate of each region is evaluated, the error output rate of each region divided by the operating point variable is evaluated. The following is illustrated in fig. 14: the output error rate of the vehicle in the region a4 becomes equal to or higher than the threshold value due to the difference in the output error rate based on the position data Pgps, and even in the region a4, when the region divided by the operating point variable enters the region a5, the output error rate becomes particularly large. In detail, the following example is shown: in the region a5, the output error rate becomes particularly large in a predetermined region divided by the combustion control mode value MC, the intake air temperature Ta, the atmospheric pressure Pa, the shift position Vsft, the engagement state value Vrc, and the road surface state value SR. Therefore, in the case of being in the area a5 and entering a predetermined area and in other cases, the CPU122 updates the evaluation mapping data 76b dedicated to the area a4 to separate (independent) data each.

Embodiment 5

Hereinafter, the following description will focus on differences from embodiment 1, and embodiment 5 will be described with reference to the drawings.

The utility map data 76a and the evaluation map data 76b according to the present embodiment are data that define a map that outputs the value of the degradation variable RD indicating the degree of degradation of the catalyst 30. Fig. 15 shows a process executed by the learning system for a vehicle according to the present embodiment. The processing shown in fig. 15 is processing realized by the CPU72 executing a program stored in the ROM74, and processing realized by the CPU122 executing a program stored in the ROM 124.

The catalyst degradation detection process M20(1) is a process of calculating the value of the degradation variable RD (1) using the utility map data 76 a. The catalyst degradation detection process M20(1) is a process of: the active control is executed at the timing at which the downstream side air-fuel ratio Afd is inverted from lean to rich so that excess oxygen is present in the exhaust gas flowing into the catalyst 30, and the value of the degradation variable RD (1) is calculated based on the amount of inflow of oxygen into the catalyst 30 during the period until the downstream side air-fuel ratio Afd is inverted from rich to lean.

The catalyst degradation detection process M20(2) is a process of calculating the value of the degradation variable RD (2) using the evaluation map data 76 b. The catalyst degradation detection process M20(2) is a process of calculating the value of the degradation variable RD (2) by a neural network that takes as input the time-series data of the upstream air-fuel ratio Afu of the catalyst 30, the time-series data of the downstream air-fuel ratio Afd, the rotation speed NE, the air-charging efficiency η, and the temperature of the catalyst 30 (catalyst temperature Tcat), and outputs the degradation variable RD (2). The catalyst temperature Tcat may be a1 st order lag (japanese: 1 times れ) treatment value of the exhaust gas temperature Tex, for example.

The re-learning sub-process M22 includes the following processes: when the absolute value of the difference between the value of the degradation variable RD (1) and the value of the degradation variable RD (2) calculated in the same trip becomes equal to or greater than a predetermined value, it is determined that the determination result using the utility map data 76a does not match the determination result using the evaluation map data 76 b. In addition, the sub-process for re-learning M22 includes a process of transmitting predetermined data to the data analysis center 120. Here, the data to be transmitted includes the value of the input variable used for calculating the degradation variable RD when a mismatch occurs. The data to be transmitted includes time series data of the upstream air-fuel ratio Afu and time series data of the downstream air-fuel ratio Afd that are chronologically adjacent to the upstream air-fuel ratio Afu and the downstream air-fuel ratio Afd used for calculating the degradation variable RD when a mismatch occurs. This is to enable the data analysis center 120 to more accurately determine whether or not the degradation variable RD (2) is an accurate value based on the behavior of the upstream air-fuel ratio Afu and the downstream air-fuel ratio Afd.

The transmission target includes an intake air amount Ga. This is because: since the behavior of the downstream air-fuel ratio Afd is easily affected by the intake air amount Ga, it is verified whether the accuracy of the degradation variable RD (2) is significantly reduced depending on the value of the intake air amount Ga.

The transmission destination includes a variation Δ η that is a difference between the maximum value and the maximum value in the predetermined time of the inflation efficiency η. This is because: since the air-fuel ratio of the air-fuel mixture to be burned in the combustion chamber 18 changes according to the variation in the charging efficiency η, it is verified whether the accuracy of the degradation variable RD (2) is significantly reduced according to the variation Δ η.

In addition, the transmission target includes the position data Pgps. This is because: since fuel properties such as the alcohol concentration in the fuel vary from region to region, it is verified whether the accuracy of the value of the degradation variable RD (2) is significantly reduced from region to region.

The CPU122 executes processing based on the right side of fig. 4 and the right side of fig. 6 as relearning processing M24. Further, as the processing based on the processing of fig. 4, the following processing is included: if the absolute value of the difference between the value of the degradation variable RD, which is considered appropriate in the data analysis center 120, and the output value of the map defined by the evaluation map data 76b is equal to or greater than a predetermined value, an erroneous output is assumed. In addition, in the present embodiment, an example is shown in which after the error output rate of each region is evaluated, the error output rate of each region divided by the intake air amount Ga is evaluated. The following is illustrated in fig. 15: the output error rate of the vehicle located in the region a6 becomes equal to or higher than the threshold value according to the difference in the output error rate based on the position data Pgps, and in the region a6, the output error rate becomes particularly large in one region a7 of the three regions divided according to the intake air amount Ga. Specifically, an example is shown in which the output error rate is particularly large in a region where the amount of change Δ η in the region a7 is equal to or greater than a predetermined amount. Therefore, when the variation Δ η is equal to or greater than a predetermined amount in the region a7 of the three regions divided by the intake air amount Ga, and when otherwise, the CPU122 updates the evaluation map data 76b dedicated to the region a 6.

Embodiment 6

Hereinafter, the following description will be made of embodiment 6, focusing on differences from embodiment 1, with reference to the drawings.

In the present embodiment, the catalyst 30 is exemplified by a catalyst having a filter for trapping Particulate Matter (PM) mounted thereon, and the practical map data 76a and the evaluation map data 76b are data defining a map for outputting a PM deposition amount DPM, which is an amount of PM deposited on the filter.

Fig. 16 shows a process executed by the learning system for a vehicle according to the present embodiment. The processing shown in fig. 16 is processing realized by the CPU72 executing a program stored in the ROM74, and processing realized by the CPU122 executing a program stored in the ROM 124.

The deposition amount estimation processing M30(1) is processing for calculating the PM deposition amount DPM (1) using the practical map data 76 a. The deposit amount estimation process M30(1) is a process of calculating the PM deposit amount DPM (1) using map data that specifies the relationship between the rotation speed NE and the inflation efficiency η and the base value of the PM deposit amount DPM, map data that specifies the relationship between the ignition timing aig and the correction amount of the PM deposit amount DPM, and map data that specifies the relationship between the water temperature THW and the correction amount of the PM deposit amount DPM.

The deposition amount estimation processing M30(2) is processing for calculating the PM deposition amount DPM (2) using the evaluation map data 76 b. The deposit amount estimation process M30(2) is a process of calculating the PM deposit amount DPM (2) using a neural network that outputs the amount of change in the PM deposit amount DPM (2) with the rotation speed NE, the inflation efficiency η, the upstream air-fuel ratio Afu, the catalyst temperature Tcat, the ignition timing aig, and the water temperature THW as inputs.

The re-learning sub-process M32 includes the following processes: when the absolute value of the difference between the PM accumulation amount DPM (1) and the PM accumulation amount DPM (2) is equal to or greater than a predetermined value, it is determined that the determination result using the utility map data 76a does not match the determination result using the evaluation map data 76 b. The sub-process M32 for re-learning includes a process of transmitting, as a transmission target, time-series data or the like including input data used for calculating the PM accumulation amount DPM (2) until a time point when a mismatch occurs after the internal combustion engine 10 is started. The reason why the time-series data is transmitted is to calculate the PM accumulation amount DPM based on the time-series data in the data analysis center 120 and determine the validity of the PM accumulation amount DPM (2).

The data to be transmitted includes a combustion control mode value MC. In the present embodiment, as the combustion control mode value MC, a combustion control mode value MC that identifies a mode corresponding to a value of a target air-fuel ratio, a mode in which control for increasing the exhaust gas temperature is performed for the regeneration process of a filter, and other modes is exemplified. This is to verify whether the accuracy of estimation of the PM accumulation amount DPM (2) changes depending on the difference in the combustion control mode.

The data to be transmitted includes an elapsed time Ts from the start of the internal combustion engine 10. This is because: immediately after the start, since fuel is likely to adhere to the cylinder wall surface and the like and PM is likely to be generated, it is verified whether or not a difference occurs in the estimation accuracy of the PM accumulation amount DPM (2) between immediately after the start and the period other than the start.

The data to be transmitted includes the position data Pgps and the intake air temperature Ta. The reason is the same as the reason for transmitting these data by the sub-process for relearning M12.

The CPU122 executes processing based on the right side of fig. 4 and the right side of fig. 6 as relearning processing M34. Further, as the processing based on the processing of fig. 4, the following processing is included: if the absolute value of the difference between the value considered appropriate as the value of the PM accumulation amount DPM and the output value of the map defined by the evaluation map data 76b is equal to or greater than the predetermined value in the data analysis center 120, it is assumed to be an erroneous output. In addition, this embodiment shows an example in which after the error output rate of each region is evaluated, the error output rate of each region divided by the elapsed time Ts is evaluated. The following is illustrated in fig. 16: the output error rate of the vehicle located in the region A8 becomes equal to or higher than the threshold value according to the difference in the output error rate based on the position data Pgps, and the output error rate becomes particularly large in the region A8 in the region a9 in which the elapsed time Ts is equal to or less than the predetermined time Ts 1. Fig. 16 shows an example in which the output error rate is particularly large in a region a9 in which the combustion control mode value MC becomes a predetermined value even in a region in which the elapsed time Ts is greater than the predetermined time Ts 1. Therefore, the CPU122 updates the evaluation map data 76b specific to the region a8 to separate data in each of the region in which the elapsed time Ts is equal to or less than the predetermined time Ts1, the region in which the elapsed time Ts is greater than the predetermined time Ts1 and the combustion control mode value MC is a predetermined value, and other regions.

Embodiment 7

Hereinafter, the following description will focus on differences from embodiment 1, and embodiment 7 will be described with reference to the drawings.

The practical map data 76a and the evaluation map data 76b according to the present embodiment are data that define a map of the temperature of the output catalyst 30 (catalyst temperature Tcat). Fig. 17 shows a process executed by the learning system for a vehicle according to the present embodiment. The processing shown in fig. 17 is processing realized by the CPU72 executing a program stored in the ROM74, and processing realized by the CPU122 executing a program stored in the ROM 124.

The catalyst temperature estimation process M40(1) is a process of calculating the catalyst temperature Tcat (1) using the utility map data 76 a. The catalyst temperature estimation process M40(1) is a process of executing a1 st order lag filter process with the exhaust gas temperature Tex as an input and setting the output value thereof as the catalyst temperature Tcat (1).

The catalyst temperature estimation process M40(2) is a process of calculating the catalyst temperature Tcat (2) using the evaluation map data 76 b. The catalyst temperature estimation process M40(2) is a process of calculating the catalyst temperature Tcat (2) using a neural network to which the time-series data of the exhaust gas temperature Tex, the rotation speed NE, the charging efficiency η, and the upstream air-fuel ratio Afu, and the last value of the catalyst temperature Tcat (2) are input.

The sub-process for re-learning M42 includes a process of determining that the determination result using the utility map data 76a does not match the determination result using the evaluation map data 76b when the absolute value of the difference between the catalyst temperature Tcat (1) and the catalyst temperature Tcat (2) is equal to or greater than a predetermined value. In addition, the sub-process for re-learning M42 includes a process of sending data to the data analysis center 120.

The data to be transmitted includes time series data made up of input data used for calculating the catalyst temperature Tcat (2) until a time point when a mismatch occurs after the internal combustion engine 10 is started. Here, the reason why the time-series data is transmitted is to calculate the catalyst temperature Tcat from the time-series data and determine the validity of the catalyst temperature Tcat (2) in the data analysis center 120.

The data to be transmitted includes the combustion control mode value MC. In the present embodiment, as the combustion control mode value MC, a combustion control mode value MC that identifies a mode corresponding to a value of the target air-fuel ratio, a mode in which control for increasing the exhaust gas temperature is performed for the regeneration process of the catalyst 30, and other modes is exemplified. This is to verify the influence of the difference in the combustion control pattern on the estimation accuracy of the catalyst temperature Tcat (2).

The data to be transmitted includes the intake air amount Ga. This is because: since the heat exchange between the catalyst 30 and the air differs depending on the intake air amount Ga, the influence of the intake air amount Ga on the estimation accuracy of the catalyst temperature Tcat (2) is verified.

The data to be transmitted includes a vehicle speed V. This is to verify the influence of the traveling wind on the estimation accuracy of the catalyst temperature Tcat (2). The data to be transmitted includes the intake air temperature Ta. Here, the intake air temperature Ta is a substitute for the temperature of the outside air. This is because: since the thermal gradient between the catalyst 30 and the outside air changes depending on the outside air temperature, the influence of the outside air temperature on the estimation accuracy of the catalyst temperature Tcat (2) is verified.

The CPU122 executes processing based on the right side of fig. 4 and the right side of fig. 6 as relearning processing M44. Further, as the processing based on the processing of fig. 4, the following processing is included: when the absolute value of the difference between the value of the catalyst temperature Tcat considered to be appropriate in the data analysis center 120 and the output value of the map defined by the evaluation map data 76b is equal to or greater than a predetermined value, an erroneous output is set. In addition, this embodiment shows an example in which the error output rate of each of the three regions divided by the intake air amount Ga is first evaluated. The following is illustrated in fig. 13: in one of the three regions a11 defined by the intake air amount Ga, the output error rate becomes particularly large when the vehicle speed V is equal to or higher than a predetermined speed and the intake air temperature Ta is equal to or lower than a predetermined temperature. Therefore, the CPU122 updates the dedicated evaluation map data 76b when the vehicle speed V is equal to or higher than the predetermined speed and the intake air temperature Ta is equal to or lower than the predetermined temperature in the region a 11.

Corresponding relation

The correspondence between the matters in the above embodiment and the matters described in the above section of "summary of the invention" is as follows. The 1 st execution device corresponds to the CPU72 and the ROM74, and the 2 nd execution device corresponds to the CPU122 and the ROM 124. The storage device corresponds to the storage device 76. The mapping data corresponds to the evaluation mapping data 76 b. The acquisition process corresponds to the process of S40. The relationship evaluation processing corresponds to the processing of S100, S102, S106, S108, S112, S114, S116. The values of the predetermined variables correspond to the position data Pgps, the intake air temperature Ta, the atmospheric pressure Pa, the combustion control mode value MC, the shift stage Vsft, the engagement state value Vrc, the charging efficiency η, and the road surface state value SR in fig. 6, the rotation speed NE in fig. 11, the intake air temperature Ta in fig. 13, and the like. The function approximator corresponds to a neural network, the update process corresponds to the processes of S104, S110, S118, and S120, and the division process corresponds to the processes of S104, S110, and S118. The function approximator corresponds to the neural network, the update processing corresponds to the processing of S120 and S144, and the change processing corresponds to the processing of S144. The update processing corresponds to the processing of S120 and S154, and the addition processing corresponds to the processing of S154. The 1 st mapping data corresponds to the utility mapping data 76a, the 1 st acquiring process corresponds to the process of S10, and the 1 st calculating process corresponds to the processes of S16 and S18. The conformity determination processing corresponds to the processing of S64, S70. The validity determination process corresponds to the processes from S92 to S96. The vehicle-side transmission processing corresponds to the processing of S82, and the vehicle-outside reception processing corresponds to the processing of S90. The update data transmission process corresponds to the process of S122, and the update data reception process corresponds to the process of S132. The vehicle control device corresponds to the control device 70. The vehicle learning device corresponds to the data analysis center 120.

Other embodiments

The present embodiment can be modified and implemented as follows. The present embodiment and the following modifications can be implemented in combination with each other within a range not technically contradictory.

Predetermined state of the vehicle

The predetermined state of the vehicle including the information in the output of the map is not limited to the state illustrated in the above embodiment. For example, the following state may be adopted as the state of the internal combustion engine.

(a) State associated with deterioration of responsiveness of air-fuel ratio sensor

In this case, in the deterioration determination process using the practical map data 76a as the 1 st map data, active control may be used in which the air-fuel ratio is alternately changed to lean and rich largely, instead of the normal air-fuel ratio feedback control. The practical map data 76a may be data for calculating the value of the degradation variable based on the time required until the upstream air-fuel ratio Afu is inverted from rich to lean or from lean to rich by active control. The evaluation map data 76b as the 2 nd map data may be data that defines a neural network that receives the time series data of the injection amount and the time series data of the upstream air-fuel ratio Afu and outputs a value of a degradation variable indicating the presence or absence of degradation. In this case, the process of calculating the value of the degradation variable based on the 2 nd map may be performed when the active control is not executed. The data other than the input data among the data transmitted from the vehicle to the data analysis center 120 may include the rotation speed NE and the inflation efficiency η. Thus, the data analysis center 120 can verify whether or not there is an operating point variable having a high false output rate.

(b) State related to oxygen storage amount of catalyst

In this case, the practical map data 76a as the 1 st map data may be map data in which the difference between the average value of the upstream air-fuel ratio Afu and the average value of the downstream air-fuel ratio Afd is used as an input variable and the value of a storage amount variable, which is a variable indicating the oxygen storage amount, is used as an output variable. The evaluation map data 76b as the 2 nd map data may be data that defines a neural network that takes as input the excess/deficiency amount of the actual fuel amount with respect to the fuel amount that reacts little or no with oxygen, the integrated value during the predetermined period of the temperature of the catalyst, and the last value of the storage amount variable, and outputs the value of the storage amount variable. The data other than the input data among the data transmitted from the vehicle to the data analysis center 120 may include the flow rate of the fluid in the catalyst 30. Thus, the data analysis center 120 can verify whether or not there is a traffic volume with a high false output rate.

(c) State related to presence or absence of knocking (knocking) of internal combustion engine

In this case, the practical map data 76a as the 1 st map data may be data that defines a map that outputs a logical value indicating whether knocking occurred or not by comparing the magnitude of the determination value with the integrated value of the detection values of the knock sensor. The evaluation map data 76b as the 2 nd map data may be data that defines a neural network that receives time series data of detection values of the knock sensor and outputs a peak value of the pressure in the combustion chamber 18. In this case, it is sufficient to determine that knocking has occurred when the peak value is equal to or greater than the threshold value. The data other than the input data among the data transmitted from the vehicle to the data analysis center 120 may include, for example, the rotation speed NE and the inflation efficiency η. Thus, CPU122 can verify whether or not there is an operating point region with a high false output rate.

(d) State related to temperature of fuel supplied to the fuel injection valve 20

In this case, the utility map data 76a as the 1 st map data may be map data having the rotation speed NE, the charging efficiency η, and the water temperature THW as input variables and the temperature of the fuel as an output variable. The evaluation map data 76b as the 2 nd map data may be a map that defines a neural network that receives the previous values of the rotation speed NE, the inflation efficiency η, the fuel injection amount by the fuel injection valve 20, the intake air temperature Ta, the vehicle speed V, and the fuel temperature and outputs the fuel temperature. The data transmitted from the vehicle to the data analysis center 120 may be output values and input data of the neural network from the start of the internal combustion engine 10. This enables the data analysis center 120 to estimate the time-lapse temperature of the fuel. Further, the transmitted data may include, for example, an ignition timing, a water temperature THW, and the like. Thus, CPU122 can verify whether or not the error output rate is high in a specific region divided by the ignition timing and the water temperature THW. However, even if only the output value and the input data of the neural network from the start of the internal combustion engine 10 are transmitted, it can be verified in the data analysis center 120 whether or not the error output rate is high in a specific region divided by the operating point variable.

(e) Presence or absence of exception of purge system

In this case, in a purge system including a canister that traps fuel vapor in a fuel tank and a purge valve that adjusts a flow path cross-sectional area of a purge path between the canister and an intake passage, a map that is determined to be abnormal when a hole is present in the purge path may be considered. In this case, the utility map data 76a as the 1 st map data may be data that defines a map that outputs a logical value indicating that there is an abnormality when the rate of increase of the pressure when the purge valve is closed after the purge valve is opened to reduce the pressure in the tank is equal to or greater than a threshold value. The evaluation map data 76b as the 2 nd map data may be data that defines a neural network that receives time series data of the pressure in the tank and the atmospheric pressure Pa and outputs an output value corresponding to the presence or absence of holes. In addition to the output value and the input data of the neural network, the data transmitted from the vehicle to the data analysis center 120 may be the remaining amount of fuel in the fuel tank. Thus, the influence of the remaining fuel amount on the accuracy of the output value can be verified.

(f) EGR rate

Here, the EGR rate is a ratio of the flow rate of the fluid flowing from the EGR passage 32 into the intake passage 12 to the flow rate of the fluid flowing from the intake passage 12 into the combustion chamber 18. In this case, the practical map data 76a as the 1 st map data may be map data having the rotation speed NE and the charging efficiency η as input variables and the EGR rate as an output variable. The evaluation map data 76b as the 2 nd map data may be data that defines a neural network that outputs the EGR rate with the rotation speed NE, the charging efficiency η, the pressure in the intake passage 12, and the intake air amount Ga as input variables.

In this case, the data transmitted from the vehicle to the data analysis center 120 may include, for example, the atmospheric pressure Pa, the intake air temperature Ta, and the water temperature THW, in addition to the input data used for calculating the EGR rate when a mismatch occurs. This makes it possible to verify the influence of the atmospheric pressure Pa, the intake air temperature Ta, and the water temperature THW on the accuracy of the output value.

(g) State relating to presence or absence of leakage of blowby gas delivery passage

Here, it is assumed that a blowby gas delivery passage is provided which connects the crankcase of the internal combustion engine 10 and the intake passage 12. In this case, a pressure sensor may be provided in the blowby gas delivery passage, and the practical map data 76a as the 1 st map may be data that outputs a value indicating the presence or absence of an abnormality based on a comparison between the pressure detected by the pressure sensor and a determination value based on the rotation speed NE and the inflation efficiency η. The evaluation map data 76b as the 2 nd map data may be data that defines a neural network that outputs values indicating the presence or absence of an abnormality, with the rotation speed NE, the inflation efficiency η, and the difference between the intake air amount Ga and the intake air amount that has passed through the throttle 14 as input variables.

In this case, by transmitting input data used for calculating a value indicating whether or not a mismatch has occurred from the vehicle to the data analysis center 120, it is possible to verify, for example, the influence of the operating point variable on the accuracy of the output value.

The predetermined state of the vehicle is not limited to the state of the internal combustion engine. For example, in a vehicle provided with a rotating electric machine as described in the column "with respect to the vehicle" below, the battery may be in a state of storing electric power supplied to the rotating electric machine.

For vehicle-side transmission processing

(a) In the event of fire

In the processing of fig. 4, the time series data of the minute rotation time T30 of 3 combustion cycles is transmitted, but the present invention is not limited thereto. For example, the time series data may be 2 combustion cycle amounts, i.e., the minute rotation times T30(25) to T30(48) when the determination result using the utility map data 76a does not match the determination result using the evaluation map data 76b, and the minute rotation times T30(49) to T30(72) when the state is shifted from the state determined to be not matching to the state determined to be matching.

In the processing of fig. 4, the minute rotation times T30(49) to T30(72) when the state is shifted from the non-coincidence determined state to the coincidence determined state are transmitted in addition to the minute rotation times T30(25) to T30(48) when the determination result using the utility map data 76a and the determination result using the evaluation map data 76b do not coincide with each other, but the present invention is not limited thereto. For example, the time series data of the minute rotation time T30 in the state determined to match and the time series data of the minute rotation time T30 when the state is shifted from the state determined to match to the state determined to not match may be transmitted.

The time-series data of the minute rotation time T30 at the time of transition to the state determined to match among the time-series data to be transmitted is not limited to the time-series data of 1 combustion cycle amount. For example, as described in the column of "2 nd map data", when the value of the misfire variable of only 1 cylinder is output based on the output value of 1 input, and the input data itself is time series data of the minute rotation time T30 in a period shorter than 1 combustion cycle, the corresponding time series data is used. Of course, the time series data of the minute rotation time T30 constituting the input variable to the map and the time series data of the minute rotation time T30 when the transition to the state of matching determination is made are not necessarily the minute rotation time T30 in the section having the same length.

In the processing of fig. 4, the time series data of the minute rotation time T30 of 3 combustion cycles corresponding to the case where the number of consecutive determinations of non-compliance is the maximum is transmitted 1 time in 1 trip, but the present invention is not limited thereto. For example, the 1-cycle time series data of 1 combustion cycle amount may be transmitted, which corresponds to 1 time when the number of consecutive determinations of non-agreement is the maximum, and which includes all the minute rotation time T30 in the period of consecutive determinations of non-agreement and the minute rotation time T30 when the state of non-agreement is shifted from the state of determination of non-agreement to the state of determination of agreement. For example, the time series data of 1 combustion cycle amount may be transmitted for all the minute rotation time T30 in the period in which the determination of the mismatch is 1 time and the minute rotation time T30 when the state of the determination of the mismatch shifts to the state of the determination of the mismatch in each of these periods, for 1 stroke.

The data other than the input data to the map and the minute rotation time T30 in the data to be transmitted are not limited to the data exemplified by the extra information set GrE. It is not essential to transmit the input data to the map and the data other than the minute rotation time T30.

(b) Integral body

The data to be transmitted, which is related to the output value of the 2 nd map specified by the 2 nd map data when it is determined that the data does not match, is not limited to the output value of the 2 nd map itself. For example, in the processing of fig. 4, the output value of the map defined by the utility map data 76a may be used. In this case, for example, when the skilled person determines that the output value of the map defined by the practical map data 76a is correct in the processing of S92 to S94, the processing of S96 may determine that the result is yes. Of course, even if such data is not transmitted, the output value of the map defined by the evaluation map data 76b can be calculated on the data analysis center 120 side by transmitting the input data.

Regarding the 1 st mapping, the 1 st mapping data

In fig. 1 and 14 to 17, the practical mapping data 76a as the 1 st mapping data is exemplified by data for adjusting the determination value and the like without using machine learning, but is not limited thereto. For example, the data may be data that defines a neural network in which the dimension of the input variable is smaller than the mapping defined by the evaluation mapping data 76b as the 2 nd mapping data. Of course, the dimension of the input of the 1 st map does not have to be smaller than the dimension of the input of the 2 nd map. For example, the number of input dimensions may be the same and the number of layers in the middle layer may be smaller than the number of layers in the 2 nd mapping. For example, the number of input dimensions and the number of intermediate layers may be the same as those of the 2 nd mapping, and the activation functions may be different from each other.

The function approximator of the learning completion model learned by machine learning is not limited to a neural network. For example, a regression expression or the like may be used when the information on the state is represented by a value of 3 or more, or a discrimination function may be used when a determination value of normality or abnormality is output. The regression expression may be equivalent to the expression obtained by removing the intermediate layer in the neural network, and may be an expression using a kernel function (japanese: カーネル Seki number), for example. The identification function may be a function having an output value of a logical sigmoid function (logical signature) in which an output of the regression expression is an independent variable.

About the 2 nd mapping data

(a) Integral body

The evaluation mapping data 76b as the 2 nd mapping data is not limited to data that specifies a neural network whose middle layer is layer 1. For example, the 2 nd mapping data may be data for specifying a neural network having an intermediate layer of 2 or more layers. The activation function h1 is not limited to the hyperbolic tangent function, and may be a logical sigmoid function or ReLU, for example. The ReLU is a function of the output input value and the value of the value that is not smaller than "0".

The function approximator of the learning completion model learned by machine learning is not limited to a neural network. For example, a regression expression or the like may be used when the information on the state is represented by a value of 3 or more, or a discrimination function may be used when a determination value of normality or abnormality is output. The regression expression here may be equivalent to the expression obtained by removing the intermediate layer in the neural network, and may be an expression using a kernel function, for example. The identification function may be a function having as an output value of a logical sigmoid function in which an output of the regression expression is an independent variable.

(b) Condition of misfire detection

The number of nodes in the output layer of the neural network, i.e., the dimension, is not limited to "(number of cylinders) + 1". For example, the number of cylinders may be equal to the number of cylinders, and it may be determined that misfire is present when a value exceeding a threshold value is present in each output value. For example, the number of cylinders to be subjected to misfire determination based on 1 output of the neural network may be 1, and the number of nodes of the output layer may be 1. In this case, the output layer preferably normalizes a range of desirable values of the output value by a logical sigmoid function or the like.

Regarding the division processing

In the above embodiment, the region in which the output error rate is lower than the threshold value may be set as a common region by division. That is, for example, in a region other than the region a2 in fig. 8 within the region a1 in fig. 7, if the error output rate is smaller than the threshold Rth1, only the evaluation map data 76b dedicated to the region a2 within the region a1 and only the evaluation map data 76b in all the regions other than the region a2 may be updated.

For example, the order of creating the distribution of the error output rate is not limited to the order illustrated in the above embodiment. Further, for example, when the regions are divided according to the values of the plurality of variables, it may be determined whether or not the error output rate is equal to or higher than a threshold value, and the total number of the regions that are eventually equal to or higher than the threshold value and the regions that are not equal to or higher than the threshold value may be minimized.

In the above configuration, the classification of the error output rate based on the position data Pgps is performed on the basis of the country or a predetermined region into which the country is subdivided, but the present invention is not limited thereto. For example, all the regions may be in units of countries, or for example, the regions may be in units of regions larger than the countries. Conversely, all the regions into which the country is subdivided may be used as units.

Further, for example, the division may be performed by using a statistical method. This may be performed using a clustering method such as K-means. For example, the error output distribution may be approximated by a mixture gaussian distribution, which is a linear combination of gaussian distributions whose average values become the center of the error output distribution, and the regions may be divided by using the approximated error output distribution. That is, the mixture coefficient, the average value, and the variance in the mixture gaussian distribution may be learned based on the erroneously output position data Pgps, and the region may be divided for each region in which the value of one gaussian distribution is larger than the other values. The target of the statistical method is not limited to the classification of the false output rate based on the position data Pgps, and for example, the classification of the false output rate based on the value of the environmental variable such as the intake air temperature Ta and the classification of the false output rate based on the value of the operation state variable such as the combustion control mode value MC may be targeted.

In the above-described embodiment, there is no particular mention of the case where the division processing is executed in the proportion of the region where the error output rate is equal to or greater than the threshold value, but in the case where the proportion of the region where the error output rate is equal to or greater than the threshold value is large, the value of the variable may not contribute much to the reduction of the error output rate, so the division processing is preferably executed in the proportion of the predetermined value or less. Instead of quantizing the distribution of the error output rates referred to when the division process is performed to the ratio of the absolute number of error outputs, a histogram may be used instead, which is a graph showing the relative error output rates of the respective levels with the divided regions as the levels.

About change processing

Fig. 12 illustrates an example of a change process for further dividing the area when the map data used is divided for each area, but the change process is not limited to this. For example, the boundary between the regions may be changed in a state where the number of divisions itself is fixed. Of course, the present invention is not limited to this, and for example, the number of divisions may be changed to be reduced.

For additional processing

The addition process is not limited to the process of adding the intake air temperature Ta.

About update processing

For example, the division processing and the addition processing may be executed by executing the processing of S106 in addition to the processing of S154 in fig. 12.

Processing for consistency determination

The verification period as the process of S60 is not limited to the period exemplified in the above embodiment. In the processing of fig. 4, the matching or non-matching of the misfire identification result based on the practical map data 76a and the misfire identification result based on the evaluation map data 76b is determined only during the verification period, but the present invention is not limited thereto, and the above determination may be performed at all times.

Parameters on completion of relearning

In fig. 6, 11, and 13, the parameters whose relearning is completed are transmitted to the vehicles VC1, VC2, · · via the network 110 after updating, but the present invention is not limited thereto. For example, the data may be sent to a vehicle store, and the data in the storage device 76 may be updated when each vehicle VC1, VC2, · · · · · · get shipped to the store. Even in this case, the reliability of the evaluation map data 76b updated with the relearning-completed parameters can be further evaluated and updated.

Of course, it is not necessary per se to provide the vehicle that provides the data used in the relearning with the parameters that have completed the relearning. The evaluation map data 76b may be updated using the relearning completed parameters, and the updated evaluation map data 76b may be installed only in a newly developed vehicle. In this case, when outputting the map indicating the output value of the information on the state of the internal combustion engine 10, it is preferable that the difference between the exhaust gas amount of the internal combustion engine mounted on the newly developed vehicle and the exhaust gas amount of the internal combustion engine mounted on the vehicle that transmitted the data for relearning is equal to or less than a predetermined amount. In the case where the evaluation map data outputs the misfire variable according to the probability of the misfire occurring in each cylinder as in the above-described embodiment, the number of cylinders of the internal combustion engine mounted on the newly developed vehicle is preferably the same as the number of cylinders of the internal combustion engine mounted on the vehicle that transmitted the data for relearning.

Further, in fig. 6, 11, and 13, the evaluation map data 76b may be updated using the relearning completed parameters, and then the utility map data 76a may be overwritten.

Relating to the display device

In the above embodiment, the display device 112 is disposed in the data analysis center 120, but the present invention is not limited thereto, and may be disposed in a place different from the place where the storage device 126 and the like are disposed.

Processing for validity determination

In fig. 4, the input data used for calculating the misfire variable p (j) calculated using the evaluation map data 76b and the associated data are displayed on the display device 112, whereby the skilled person evaluates whether or not the judgment is erroneous, but the invention is not limited thereto. For example, the evaluation may be performed using map data having higher precision of misfire determination than the map defined by the utility map data 76 a. For example, a neural network may be used in which all elements of the rotation time set GrT30 and the additional information set GrE received in the processing of S90 are input variables or more elements are input variables and the misfire variable is output variables. In this case, it is preferable that the number of layers of the intermediate layer is larger than the number of layers of the intermediate layer of the neural network specified by the evaluation mapping data 76 b.

Of course, it is not essential to use a subject having higher accuracy than the mappings defined by the evaluation mapping data 76b and the utility mapping data 76a to determine the validity of the determination result of the mapping defined by the evaluation mapping data 76 b. For example, the validity of the determination result of the map defined by the evaluation map data 76b may be determined by a majority of the determination result defined by the evaluation map data 76b and the determination result based on two or more other maps. For example, instead of the determination result based on the map, one of the determination results based on the two or more other maps may be determined by a skilled person.

In the processing of fig. 4, for convenience of explanation, the processing of S92 is executed each time the processing of S82 is executed, but is not limited thereto. For example, the process of S92 may be executed when the data determined to be inconsistent accumulates a predetermined amount. For example, data considered to be inconsistent may be accumulated every time, and the process of S92 may be executed in response to a request from a skilled person.

Determination processing of validity of output value of map

The validity is determined by other methods not only after the determination process of determining whether or not the output value of the map defined by the utility map data 76a as the 1 st map data matches, but also when it is determined that the output value does not match. For example, when it is determined that the output value of the map defined by the evaluation map data 76b does not match the output value of the map defined by the practical map data 76a or the output value of the map with higher accuracy than the output value of the map defined by the practical map data 76a, the output value of the map defined by the evaluation map data 76b may be determined to be inappropriate.

For example, the in-vehicle sensors may include sensors that detect values of variables related to output values of the map defined by the evaluation map data 76b, and the detection values may be used as teacher data to determine whether or not the output values match, and when it is determined that the output values do not match, the output values of the map may be determined to be inappropriate. This can be achieved, for example, by: when the map of the peak value of the pressure in the output combustion chamber 18 described in (c) of "the predetermined state of the vehicle" is the map defined by the evaluation map data 76b, the sensor that detects the pressure in the combustion chamber 18 is included in the vehicle-mounted sensor.

About update processing

In the above embodiment, when learning the evaluation mapping data 76b used only in a certain area, only the training data of the area is used, but the invention is not limited thereto. For example, in the case of relearning the evaluation map data 76b used in a region other than the region a1 in fig. 7, the training data of the region a1 may be mixed at a predetermined ratio.

About environmental variables

The environment variables for verifying the relationship with the accuracy of the output value of the map defined by the evaluation map data 76b are not limited to the variables exemplified in the above embodiment. For example, the variable may indicate the weather around the vehicle VC1 and the wind speed. In particular, since the heat radiation condition of the catalyst 30 is affected by the weather and the wind speed with respect to the catalyst temperature Tcat or the like, it is effective to use these variables. Further, the value of the variable indicating weather and the value of the variable indicating wind speed may be transmitted from the vehicle VC1, but instead, the values may be acquired from a mechanism that provides weather-related information or the like in the data analysis center 120 based on the position data Pgps of the vehicle VC 1.

With respect to variables representing operating conditions

The variables representing the operating state for verifying the relationship with the accuracy of the output value of the map defined by the evaluation map data 76b are not limited to the variables exemplified in the above embodiment. In particular, the combustion control mode value MC is not limited to the above-described division, and for example, the division used in one of the above-described embodiments may be applied to the other embodiments.

Operation processing on output values based on the mapping specified by the 1 st mapping data

In the above embodiment, the notification process of operating the warning lamp 104 mounted on the vehicle is exemplified as the operation process of operating the predetermined hardware based on the output value of the map defined by the 1 st map data, but the notification process is not limited to this. For example, the communication device 77 may be operated to display information indicating that an abnormality has occurred in the user's mobile terminal.

The operation processing is not limited to notification processing. For example, the processing may be processing for operating an operation unit for controlling combustion of the air-fuel mixture in the combustion chamber 18 of the internal combustion engine 10 based on information indicating that a misfire has occurred. For example, as shown in fig. 14, when the evaluation map data 76b is a map that outputs a determination result of the presence or absence of the imbalance abnormality, the evaluation map data may be processed to operate the fuel injection valve to suppress the imbalance abnormality. For example, when the evaluation map data 76b is a map for outputting the PM amount as shown in fig. 16, the processing may be performed to burn and remove PM by operating an operation unit of the internal combustion engine 10 for raising the temperature of the filter. For example, when the evaluation map data 76b is a map for outputting the temperature of the catalyst 30 as shown in fig. 17, the evaluation map data may be a process for operating an operation unit of the internal combustion engine for increasing the temperature of the catalyst 30. The operation treatment in this case may be, for example, a regeneration treatment of the catalyst.

Role sharing with respect to the 1 st and 2 nd execution devices

For example, the CPU122 may execute the processing of fig. 3 by storing the utility map data 76a as the 1 st map data in the storage device 76, storing the evaluation map data 76b as the 2 nd map data in the storage device 126, and executing the processing of fig. 3. In this case, the process of fig. 4 may be changed as follows, for example.

1. The CPU72 side transmits the input data to the map defined by the evaluation map data 76b to the CPU122 side. The output value of the map is sent from the CPU122 side to the CPU72 side. The CPU72 determines whether or not the output value of the map defined by the utility map data 76a matches the output value of the map defined by the evaluation map data 76 b. If the CPU72 determines that the data do not match, it transmits the associated data.

2. In addition to the input data to the map defined by the evaluation map data 76b, the output value of the map defined by the utility map data 76a is transmitted from the CPU72 side to the CPU122 side. The CPU122 determines whether or not the output value of the map defined by the utility map data 76a matches the output value of the map defined by the evaluation map data 76 b. If it is determined that the input data does not match the input data, CPU122 may request CPU72 to transmit related data other than the input data.

Further, not only the evaluation mapping data 76b as the 2 nd mapping data but also the utility mapping data 76a as the 1 st mapping data may be stored in the storage device 126. In this case, the CPU72 side may transmit the input data to the map defined by the utility map data 76a and the input data to the map defined by the evaluation map data 76b to the CPU122 side.

For example, the 2 nd execution device may be configured by a CPU and a ROM provided in the portable terminal, in addition to the CPU122 and the ROM 124. This can be achieved, for example, by: in embodiment 1 described above, the process of fig. 3 is executed by the mobile terminal and the result is transmitted to the control device 70.

Learning device for vehicle

Instead of the data analysis center 120, the vehicle learning device may be configured using a portable terminal. This can be achieved, for example, by: the data of the map with higher predetermined accuracy than the practical map data described in the column "determination processing on validity" is stored in advance in the storage device of the mobile terminal, and the processing on the right side of fig. 4 and the processing on the right side of fig. 6 are executed by the mobile terminal. In this case, only the data relating to vehicle VC1 may be transmitted to the portable terminal of the user of vehicle VC 1.

Configuration of the 1 st and 2 nd execution devices

The 1 st and 2 nd execution devices are not limited to those provided with the CPU72(102) and the ROM74 (104) and executing software processing. For example, a dedicated hardware circuit (e.g., ASIC) may be provided for performing hardware processing on at least a part of the processing performed by software processing in the above-described embodiment. That is, the 1 st and 2 nd actuators may be configured as any of the following (a) to (c). (a) The processing device includes a processing device for executing all of the above-described processing in accordance with a program, and a program storage device such as a ROM for storing the program. (b) The apparatus includes a processing device and a program storage device for executing a part of the above processes in accordance with a program, and a dedicated hardware circuit for executing the remaining processes. (c) The apparatus includes a dedicated hardware circuit for executing all of the above processes. Here, a plurality of software execution devices and dedicated hardware circuits may be provided, each of which includes a processing device and a program storage device.

With respect to the storage device

In the above embodiment, the storage device 76 storing the evaluation map data 76b and the utility map data 76a and the ROM74 as a storage device storing the relearning subroutine 74b are provided as separate storage devices, but the present invention is not limited thereto. For example, the storage device 126 storing the data 126a for relearning and the ROM124 storing the main relearning program 124a are provided as separate storage devices, but the present invention is not limited thereto.

About internal combustion engines

In the above-described embodiment, the in-cylinder injection valve that injects fuel into the combustion chamber 18 is exemplified as the fuel injection valve, but the present invention is not limited thereto. For example, a port injection valve that injects fuel into the intake passage 12 may be used. For example, both the port injection valve and the in-cylinder injection valve may be provided.

The internal combustion engine is not limited to a spark ignition type internal combustion engine, and may be, for example, a compression ignition type internal combustion engine using light oil or the like as fuel. It is not necessary per se that the combustion engine constitutes a drive system. For example, the present invention may be mounted in a so-called series hybrid vehicle in which a crankshaft is mechanically coupled to an in-vehicle generator and power transmission to drive wheels 60 is blocked.

About the vehicle

The vehicle is not limited to a vehicle in which the device generating the propulsion force of the vehicle is only an internal combustion engine, and may be a parallel hybrid vehicle or a series-parallel hybrid vehicle, for example, in addition to the series hybrid vehicle described in the column "regarding the internal combustion engine". Further, the electric vehicle may be one not equipped with an internal combustion engine.

Others

The drive system device interposed between the crankshaft and the drive wheel is not limited to the stepped transmission device, and may be a continuously variable transmission device, for example.