阅读说明：本技术 传感器校准 (Sensor calibration ) 是由 S·塔尔科玛 T·佩塔加 M·库尔马拉 J·库詹苏 于 2019-03-22 设计创作，主要内容包括：公开了一种用于通过校准过程在变化的操作环境中校准第一传感器的方法和装置。从第一传感器接收传感器数据,并且从已知校准传感器接收传感器值。维持用于第一传感器的传感器特定模型。在考虑到与已知校准传感器的传感器值的差值以及进一步考虑到第一传感器的传感器简档的情况下,通过估计漂移和误差来检测校准需求。使用上述差值来估计传感器数据的校正因子或校正模型,并且使用校正因子或校正模型来校正传感器。校正因子或校正模型是从传感器特定模型中导出的。(A method and apparatus for calibrating a first sensor in a varying operating environment through a calibration process is disclosed. Sensor data is received from a first sensor and sensor values are received from known calibration sensors. A sensor-specific model for the first sensor is maintained. The calibration requirement is detected by estimating drift and error taking into account the difference from the sensor value of the known calibration sensor and further taking into account the sensor profile of the first sensor. A correction factor or correction model of the sensor data is estimated using the above difference values, and the sensor is corrected using the correction factor or correction model. The correction factor or correction model is derived from the sensor-specific model.)

1. A method comprising calibrating a first sensor in a varying operating environment during a calibration process, the method comprising:

receiving sensor data from the first sensor;

receiving sensor values from known calibration sensors;

maintaining a sensor-specific model for the first sensor;

detecting a calibration requirement by estimating drift and error taking into account differences with sensor values of the known calibration sensor; and

estimating a correction factor or correction model for the sensor data using the difference values and correcting the sensor using the estimated correction factor or correction model;

wherein the correction factor or the correction model is derived from the sensor-specific model.

2. The method of claim 1, wherein the calibration sensor is a virtual sensor based on a model of the sensor data provided by one or more calibrated sensors.

3. The method of claim 2, wherein the virtual sensor data is developed based on long term sensor observables using determined spatiotemporal data received from the sensor network.

4. The method of any preceding claim, wherein different sensors are hierarchically ranked and a hierarchical model is used to perform the calibration of the first sensor with the calibration sensor ranked higher in the hierarchical model.

5. The method of claim 4, wherein:

the first sensor is configured to measure a given system;

maintaining a spatial model for one or more attributes of the system for which the first sensor is configured to measure;

using measurements performed outside the first sensor and under different conditions, the spatial model is used hierarchically to perform a derived calibration of the first sensor.

6. The method of claim 5, wherein the spatial model incorporates a plurality of air quality parameters.

7. The method of claim 6, wherein the spatial model is connected to a regional air quality model for additional data sources or data assimilation.

8. The method of any one of the preceding claims, wherein a plurality of sensors form a sensor group in communication with each other.

9. The method of claim 8, wherein multiple layers of calibration are performed within one node and between multiple nodes to improve accurate sensing capability.

10. The method of claim 9, wherein the sensor network comprises a mobile sensor configured to transfer the calibration from one or more reference viewing locations to a fixed sensor network.

11. The method of any preceding claim, wherein the detecting calibration requirements comprises further taking into account a sensor-specific model.

12. The method of any preceding claim, wherein the detecting a calibration requirement comprises further taking into account an operating environment of the sensor.

13. The method of any preceding claim, wherein the detecting a calibration requirement comprises further taking into account a sensor profile of the first sensor.

14. The method of any preceding claim, wherein the detecting a calibration requirement comprises further taking into account a sensor specific drift profile of the first sensor.

15. The method of any preceding claim, wherein the detection calibration requirements further take into account a sensor-specific error curve.

16. The method of any preceding claim, wherein the calibration sensor is controlled by a remote control.

17. The method of any preceding claim, wherein proxy variables are consolidated from the synthetic view and extended via a sensor network.

18. The method according to any one of the preceding claims, wherein the calibrated sensor is extended as an enhanced virtual sensor configured to detect environmental attributes that exceed hardware specifications of the at least one first sensor.

19. A system for implementing distributed services in a cellular network as edge computing for performing the method of any preceding claim.

20. An apparatus for calibrating a first sensor in a varying operating environment, the apparatus comprising:

means for receiving sensor data from the first sensor;

means for receiving sensor values from known calibration sensors;

means for maintaining a sensor-specific model for the first sensor;

means for detecting a calibration requirement by estimating drift and error taking into account differences with sensor values of the known calibration sensor; and

means for correcting the sensor by using the difference to estimate a correction factor or correction model of the sensor data;

wherein the correction factor or the correction model is derived from the sensor-specific model.

21. The apparatus of claim 20, the apparatus configured to perform the method of any of claims 1-18.

Technical Field

The present invention generally relates to sensor calibration.

Background

This section illustrates useful background information without an admission that any of the art described herein represents prior art.

Sensors typically produce a given output based on a measured attribute of something, and based on a given behavior. As a simplified example, a temperature sensor may vary resistance linearly over a given temperature range. The behavior of the thermistor can typically be modeled with a slope and an offset, i.e., using a line equation.

Some sensors, such as carbon monoxide sensors or radioactive dose sensors, may be configured to calculate a measure of aggregation or integrate the concentration of a measured property over time. Sometimes, the measured property has a non-linear relationship with the output. In addition, the gain of some sensors is also unstable compared to common thermistors. Some sensors perform internal filtering (such as band-pass or high-pass filtering) to eliminate interference, in which case some of the initial raw measurement data is inevitably lost in the sensor.

Sensors are typically calibrated by exposure to different conditions in a standardized manner, and by adjusting sensor operation to achieve a desired sensor behavior. If the sensor includes digital circuitry, calibration can be performed digitally simply by forming the desired transfer function to set the desired output according to the measured property. It is envisioned that such digital calibration may also be assigned by digital modification of sensor values on other elements besides the sensor, for example, by a processing element downstream of the sensor.

In the case of an analog sensor, calibration may be performed, for example, by increasing or decreasing a resistance coupled in series with the sensor, or by adjusting the bias of an amplifier transistor to amplify the output amplifier. However, such standardized calibration requires collecting the sensors to a calibration station or moving the standardized calibration station to the sensors.

It is an object of the invention to simplify the calibration of the sensor. It is another object of the present invention to be able to calibrate the sensor more frequently than in the prior art. It is a further object of the present invention to provide an alternative to the prior art.

Disclosure of Invention

According to a first example aspect, there is provided a method comprising calibrating a first sensor in a calibration process, the method comprising:

receiving sensor data from a first sensor;

performing at least one of: receiving sensor values from known calibration sensors; or maintaining a sensor-specific model for the first sensor;

detecting a calibration requirement by estimating drift and error taking into account differences in sensor values from a known calibration sensor or sensor specific model, and further taking into account any one or more of an operating environment of the sensor, a sensor profile, a sensor specific drift curve and a sensor specific error curve; and

calibrating the sensor by estimating a correction factor or correction model of the sensor data using the difference;

wherein the correction factor or correction model is derived from the sensor-specific model.

The method may further comprise: the different sensors are hierarchically ranked and the hierarchical model is used to perform calibration of the first sensor with the calibration sensor ranked higher in the hierarchical model.

The method may further comprise:

a spatial model of one or more attributes of the system for which the first sensor is configured to measure is maintained;

using measurements outside the first sensor and performed under different conditions, the spatial model is used hierarchically to perform a derived calibration of the first sensor.

The method may further comprise: the spatial model integrates a plurality of air quality parameters. The spatial model may be connected to the regional air quality model for additional data sources or data assimilation.

The method may further comprise: the plurality of sensors form a sensor group that communicates with each other. Multiple layers (e.g., two layers) of calibration may be performed within a node as well as between nodes to improve accurate sensing capability. The sensor network may include mobile sensors configured to transfer calibration from one or more reference viewing positions to a fixed sensor network.

The calibration sensor may be a motion sensor.

The calibration sensor may be controlled by a remote control. The remote control may comprise a server computer. Alternatively, the remote control may comprise distributed functionality. The distributed functionality may be implemented through cloud computing. The remote control may include an edge server located at the edge of the computer network.

The remote control may perform a runtime calibration of the plurality of sensors.

The calibrated sensor may be a virtual sensor based on a model of sensor data provided by the calibrated sensor. Virtual sensor data can be developed based on long term sensor observables using determined spatiotemporal data received from the sensor network.

The method may further comprise: proxy variables that are consolidated from the composite view and extended via the sensor network.

For example, SO₂Solar radiation and particulate matter can provide data on the concentration of sulfuric acid in the gas phase. New agents may be developed via artificial intelligence and/or other data mining techniques.

The method may further comprise: the calibrated sensor is extended to an enhanced virtual sensor configured to detect environmental attributes that exceed the hardware specifications of the at least one first sensor used.

The extension of calibrated sensors to enhanced virtual sensors may use a calibration model. The enhanced virtual sensor may also use an environmental model. The environment model may be generated by a remote control.

According to a second example aspect, there is provided a system for implementing distributed services in a cellular network, the system comprising:

a remote controller; and

a plurality of sensors communicatively connected to a remote controller;

wherein the remote control and the plurality of sensors are configured to perform the method of the first example aspect.

According to a third example aspect, there is provided a computer program comprising computer executable program code configured to, when executed, cause an apparatus to perform at least the method of the first example aspect.

The computer program may be stored on a storage medium. The storage medium may be a non-transitory storage medium.

According to a fourth example aspect, there is provided an apparatus for calibrating a first sensor, the apparatus comprising:

means for receiving sensor data from a first sensor;

means for performing at least one of: receiving sensor values from known calibration sensors; or maintaining a sensor-specific model for the first sensor;

means for detecting calibration requirements by estimating drift and error taking into account differences from sensor values of a known calibration sensor or sensor specific model, and further taking into account any one or more of operating environment of the sensor, sensor profile, sensor specific drift curve and sensor specific error curve; and

means for calibrating the sensor by estimating a correction factor or correction model of the sensor data using the difference;

wherein the correction factor or correction model is derived from the sensor-specific model.

Any of the foregoing storage media may include digital data storage such as a data disk, optical storage, magnetic storage, or magneto-optical storage. The storage medium may be formed as a device having no other basic functionality than storing memory, or may be formed as part of a device having other functionality, including but not limited to the memory of a computer, a chipset, and subcomponents of an electronic device.

The foregoing has described various non-limiting exemplary aspects and embodiments of the present invention. The foregoing embodiments are merely illustrative of selected aspects or steps that may be utilized in implementations of the present invention. Some embodiments may be presented only with reference to certain example aspects of the invention. It should be understood that corresponding embodiments may also be applied to other example aspects.

Drawings

Some example embodiments of the invention will be described with reference to the accompanying drawings, in which:

FIG. 1 shows a schematic diagram of a system according to an embodiment of the invention;

FIG. 2 shows a flow diagram of a process according to an example embodiment;

FIG. 3 shows a block diagram of the terminal of FIG. 1;

figure 4 shows primary signalling according to an embodiment of the invention; and

fig. 5 shows primary signaling according to another embodiment of the invention.

Detailed Description

In the following description, like reference numerals denote like elements or steps.

FIG. 1 shows a schematic diagram of a system 100 according to an embodiment of the invention. The system 100 includes a first sensor 110 in a changing operating environment. The first sensor 110 has a sensor profile defined by a given calibration parameter and associated with a respective sensor profile identity.

The system 100 also includes a calibration sensor 120 that is either a virtual sensor or a physical sensor. The calibration sensor 120 has a sensor profile defined by a given calibration parameter and associated with a respective sensor profile identity.

The sensors of the system 100 may be capable of communicating with each other or forming a sensor network, as illustrated by the arrows between the first sensor 110 and the calibration sensor 120. Such communication may be used by the first sensor 110 to detect a calibration condition, for example, from a measurement difference above a given threshold exceeding the measurement values of other sensors in proximity (referred to herein as calibration sensor 120).

In some embodiments, calibration sensor 120 and first sensor 110 may each function as a first sensor or calibration sensor, however in some other embodiments, the calibration sensor is hierarchically located above the first sensor. For example, calibration sensor 120 may have better accuracy than the first sensor, or calibration sensor 120 may be calibrated more recently and/or more frequently than first sensor 110.

The first sensor 110 and the calibration sensor 120 are depicted in fig. 1 as being communicatively coupled to a processing element, such as an edge module 130. In some embodiments, the edge module 130 is co-located with or located in a base station of a cellular network or a public land mobile network. The edge module may be operable to control calibration of the first sensor using sensing data telemetry received from the first sensor 110 and measurement data or (more generally, data relating to calibration) received from the calibration sensor 120. That is, using data received directly from the first sensor 110, the calibration sensor 120 can easily generate some calibration-related data based on observations made in the calibration sensor 120 or at the calibration sensor 120.

The edge module 130 is communicatively connected to a back end 140, which back end 140 may be implemented using, for example, a cloud calibration module or one or more server computers. The edge module 130 provides aggregated sensed data and telemetry to the back end, including measurements of the first sensor 110 and measurements of the calibration sensor 120.

In an embodiment, the backend 140 calculates or updates and maintains a sensor data map based on all received sensor measurements and sensor profiles (e.g., in a local or distributed database). In an embodiment, the sensor map also includes the locations or relative locations of the different sensors of the system 100. The sensor profile identity may be used to associate a particular sensor profile with a respective sensor.

After the sensor map has been calculated or updated, the back end 140 provides the edge module 130 with a local model/map and sensor profiles of the first sensor 110. The edge module 130 may then calculate or update a partial sensor data model or physical model map of the first sensor 110 accordingly and issue calibration data (new or updated) to the first sensor 110 as another part of the calibration process.

In an embodiment, for a first sensor, data and telemetry from the first sensor 110 is received 310 taking into account an operating environment, a sensor profile, a drift curve, and/or an error curve, and the data and telemetry is compared 320 with calibration reference data from the calibration sensor 120. If the comparison results in invalid data being found, the process proceeds to step 380 and the invalid data is analyzed in step 390. Unless invalid data discovery is performed, the process proceeds to enhance 330 the first sensor data based on a model for improving accuracy of the first sensor. Again, in this case, the process may continue to step 380, which may be the discovery of invalid data. Alternatively, where a new calibration configuration may be formed, the process may proceed to step 360, or the process may proceed to step 340.

In step 360, it is analyzed whether the new calibration configuration improves the first sensor accuracy. If so, a calibration configuration is established 370 and correction parameters including the physical location of the first sensor are formed. If not, the process continues to step 340, where the parameter values are analyzed to optimize the first sensor accuracy, and a calibration configuration is searched for that is expected to improve the first sensor accuracy. If found, the process continues to step 370, otherwise in step 350, no improvement is deemed possible.

The creation of the correction model and/or the implementation of the new configuration for the first sensor 110 may be performed at the edge module 120, which performs the data collection.

The term "virtual sensor" refers to a computational sensor that simulates or simulates a real sensor based on data generated by different sensors and based on knowledge of the environment/operating environment, the modeled behavior of the sensor, and possibly also based on artificial intelligence known from similar other systems and earlier histories within the same system 100. For example, taking an outdoor temperature measurement as an example, two relatively close sensors will typically produce fairly similar values unless one is subjected to a particular cooling or heating effect, such as localized rain or otherwise absorbing sunlight shining through gaps in a dense cloud. However, in general, the temperature between the two sensors can simply be interpolated. With knowledge of other parameters, such as wind speed and direction, and knowledge of the temperature development of other sensors around and the normal development of temperature over time, it will be relatively simple to infer how the temperature will be distributed. Artificial intelligence can be used to determine associations and potential causal relationships between different attributes, thereby forming a new cross-parameter model. This can be used to form enhanced sensor data to indicate attributes that exceed the actual capabilities of the sensor. The enhanced sensor will be further described with reference to fig. 4.

If a physical sensor is close to or has been close (at some physical distance) to a calibrated sensor, the data values from the physical sensor may be used with the current spatiotemporal data available for the area and the sensor device profile.

According to one embodiment illustrated by FIG. 3, if no data is available within a certain physical distance between the calibrated sensor and the calibrated physical sensor, the system 100 creates a virtual calibrated sensor at the location of the first sensor 110. The values of the calibration sensors 120 are inferred based on the high quality space-time diagram and the data of any physical sensors in the area in the past.

FIG. 4 illustrates a process by which the hardware sensing capabilities of a sensor can be extended by software. Spatio-temporal models and sensor models are used to detect properties, such as pollutants and environmental properties, that exceed the hardware specifications of the sensor device. The process analyzes 430 the sensor based on the spatio-temporal model to detect attributes or additional parameters that exceed the capabilities of the sensor hardware. If the analysis finds that such an extension is not possible, the process proceeds to step 480 and thereafter continues to analyze 390 the invalid data.

If a new enhanced sensor configuration has been established since the previous calibration run, then in step 460, it is analyzed whether the new calibration improves the accuracy of the first sensor 110 and ground truth is estimated for the enhanced first sensor. If not, the process continues to step 440, otherwise to step 470 to establish calibration configurations and correction parameters based on the new enhanced sensor configuration. In step 440, the first sensor is enhanced to detect new attributes, such as new pollutants, based on the first sensor data, and the environmental model and ground truth and calibration for the enhanced sensor are estimated. If in step 440, a possible improvement is found to exist for the first sensor configuration, the process proceeds to step 470 for use, otherwise to step 450, no possible improvement.

In an embodiment, ground truth is based on data from the more accurate calibration sensor 120 identifying correlations for detecting attributes that exceed hardware specifications. The sensor enhancement model may be designed in a laboratory environment; however, alternatively or additionally, the system 100 may generate such a model at runtime based on ground truth data from the physical calibration sensor 120. An enhanced sensor configuration may then be implemented and established at the physical sensor. The process may estimate the behavior of the new enhanced sensor configuration based on available ground truth data.

Returning to a higher level, let us turn to fig. 2, fig. 2 shows a flow chart of a calibration procedure according to an example embodiment for calibrating the first sensor. The process may be performed by a remote controller (i.e., one or more entities such as edge module 130 or cloud calibration module 140).

The process includes receiving 210 sensor data from a first sensor;

performing at least one of: receiving sensor values from known calibration sensors; or maintaining a sensor-specific model for the first sensor;

detecting 220 a calibration requirement by estimating drift and error taking into account a difference from a sensor value of a known calibration sensor or sensor specific model, and further taking into account any one or more of an operating environment of the sensor, a sensor profile, a sensor specific drift curve and a sensor specific error curve; and

calibrating 230 the sensor by estimating a correction factor or correction model of the sensor data using the difference;

wherein the correction factor or correction model is derived from the sensor-specific model.

In an embodiment, the different sensors are hierarchically ranked and the hierarchical model is used to perform calibration of the first sensor with the calibration sensors ranked higher in the hierarchical model.

In an embodiment, the method further comprises:

a spatial model of one or more attributes of the system for which the first sensor is configured to measure is maintained; and

using measurements outside the first sensor and performed under different conditions, the spatial model is used hierarchically to perform a derived calibration of the first sensor.

In an embodiment, the spatial model integrates a plurality of air quality parameters. The spatial model may be connected to the regional air quality model for additional data sources or data assimilation.

In an embodiment, the plurality of sensors form a sensor group in communication with each other. Multiple layers of calibration may be performed both within and between nodes to provide improved accurate sensing capabilities. The sensor network may include mobile sensors configured to transfer calibration from one or more reference viewing locations to a fixed sensor network.

In an embodiment, the calibration sensor is a mobile sensor, such as a portable sensor, a vehicle-mounted sensor, or a sensor configured to be transported by the drone.

In an embodiment, the first sensor is a movement sensor, such as a portable sensor, a vehicle-mounted sensor, or a sensor configured to be transported by a drone.

In an embodiment, the calibration sensor is controlled by a remote control. The remote control may comprise a server computer. Additionally or alternatively, the remote control may include distributed functionality. The distributed functionality may be implemented through cloud computing. The remote control may include an edge server located at the edge of the computer network.

The remote control may perform a runtime calibration of the plurality of sensors.

In an embodiment, the calibrated sensor is a virtual sensor based on a model of sensor data provided by the calibrated sensor. Virtual sensor data can be developed based on long term sensor observables using the determined spatiotemporal data received from the sensor network.

The virtual sensor may be automatically formed, for example using artificial intelligence, so that the virtual sensor used to calibrate the first sensor may be used to form or improve the calibrated sensor. Alternatively, the virtual sensor may be formed at least partially manually based on a selection of real sensors to be used and/or a parameterization to be applied with the real sensors.

In an embodiment, proxy variables are merged from the synthetic view and extended via the sensor network.

For example, SO₂Solar radiation and particulate matter can provide data on the concentration of sulfuric acid in the gas phase. New agents may be developed via artificial intelligence and/or other data mining techniques.

In an embodiment, the calibrated sensor is extended as an enhanced virtual sensor configured to detect environmental attributes that exceed the hardware specifications of the at least one first sensor used. The extension of calibrated sensors to enhanced virtual sensors may use a calibration model. The enhanced virtual sensor may also use an environmental model. The environment model may be generated by a remote control.

Fig. 5 also illustrates an analysis of the (possibly weak) correlation between the measured properties. Given that sufficient data is available, the method can be generalized to any number of attributes. State-of-the-art machine learning techniques can be used to predict attributes that exceed the original hardware specifications. 510, one sensor measures attribute a. The correlation between measurements a and B (sensed by another sensor) is analyzed based on the available data 520, and a model is developed for predicting the presence of the property given the other properties. If a correlation is found, then in step 530, a model is generated for predicting the attribute given the other attributes; otherwise the process ends. As an exception, a correlation may be found, but the correlation is not statistically significant (550), so the process terminates. If a model is generated, further attempts are made to validate the model in step 530. If the model is verified based on the available data, the process proceeds to step 540 and the model is adopted, otherwise the model is not verified and will not be adopted 560.

According to an embodiment, a model of one natural system is parameterized to adapt to another natural system. For example, a pollutant gas model for one city or country may be used for another city or country.

The processing in the system may be, for example, distributed, such that one or more remote entities (such as remote controls) are able to process a large data set, and only sensor readings and necessary local processing are performed at the sensors. Alternatively, the processing may be distributed in other ways per implementation. For example, the virtual network entity may perform the processing.

In an embodiment, the calibration of the first sensor is dynamically adjusted. The adaptation may be achieved, for example, by changing a parameter, position or orientation of the basic sensor. Additionally or alternatively, the adaptation may be achieved by adapting a model used to derive the correction factor or correction model.

Various embodiments have been proposed. It should be understood that in this document, the words "comprise", "include" and "contain" are used as open-ended expressions and are not intended to be exclusive.

The foregoing description has provided by way of non-limiting examples of particular implementations and embodiments of the invention a full and informative description of the best mode presently contemplated by the inventors for carrying out the invention. It is however clear to a person skilled in the art that the invention is not restricted to details of the embodiments presented above, but that it can be implemented in other embodiments using equivalent means or in different combinations of embodiments without deviating from the characteristics of the invention.

Furthermore, some of the features of the foregoing embodiments of this invention could be used to advantage without the corresponding use of other features. As such, the foregoing description should be considered as merely illustrative of the principles of the present invention, and not in limitation thereof. The scope of the invention is therefore intended to be limited solely by the appended patent claims.