阅读说明：本技术 感性反馈控制装置 (Inductive feedback control device ) 是由 山本透 木下拓矢 于 2019-10-25 设计创作，主要内容包括：本发明的感性反馈控制装置中,感性检测器检测与对象设备的输出相对应的操作人员的生命体征信息,基于该生命体征信息决定操作人员的舒适度。第1控制部基于关于舒适度的第1目标值与舒适度的差值来决定关于输出的第2目标值。第2控制部基于第2目标值与输出的差值来决定对对象设备的控制输入。δ设定部针对操作人员对对象设备的操作输入及控制输入分别进行与操作人员的操作水平相对应的加权。加法器将被加权后的操作输入和控制输入相加,并将所得的值输入到对象设备。(In the perceptual feedback control apparatus of the present invention, the perceptual detector detects the vital sign information of the operator corresponding to the output of the target device, and determines the comfort level of the operator based on the vital sign information. The 1 st control section determines a2 nd target value for output based on a1 st target value for comfort level and a difference between the comfort level. The 2 nd control unit determines a control input to the target device based on a difference between the 2 nd target value and the output. The delta setting unit performs weighting corresponding to the operation level of the operator for each of the operation input and the control input of the operator to the target device. The adder adds the weighted operation input and control input, and inputs the resultant value to the subject device.)

1. An inductive feedback control device, comprising:

an object device operated by an operator;

a perceptual detector for detecting vital sign information x (t) of the output u (t) of the operator with respect to the subject device, and for determining a comfort level y (t) of the operator based on the vital sign information x (t);

a1 st control unit for determining a2 nd target value w (t) for the output u (t) based on a difference between a1 st target value r (t) for the comfort level y (t) and the comfort level y (t);

a2 nd control unit for determining a control input v to the target device based on a difference between the 2 nd target value w (t) and the output u (t)_c(t)；

A weight setting unit configured to set a weight of the operation input v to the target device by the operator_h(t) and the control input v_c(t) weighting the operation levels corresponding to the operators, respectively; and the number of the first and second groups,

an adder for adding the operation inputs v weighted by the weight setting unit_h(t) and the control input v_c(t) adding and inputting the resultant value to the subject device.

2. The inductive feedback control device of claim 1 wherein:

the sensitivity detector detects 3 pieces of information, i.e., an emotional price, a liveness, and an expected feeling of the operator, as the vital sign information x (t), and determines a sensitivity value obtained from a correlation between the 3 pieces of information as the comfort level y (t).

3. The inductive feedback control device according to claim 1 or 2, further comprising:

a database for sequentially storing the 1 st target value r (t), the 2 nd target value w (t), and the comfort level y (t); wherein the content of the first and second substances,

the 1 st control unit determines the 2 nd target value w (t) while adjusting a control parameter, using data stored in the database.

4. The inductive feedback control device according to any of claims 1 to 3, wherein:

the target equipment is a construction machine.

5. The inductive feedback control device of claim 4 wherein:

the output u (t) is a response speed of a working device provided in the construction machine.

6. The inductive feedback control device of claim 3 wherein:

the control parameter is a proportional gain K_p(t) integral gain K_I(t) and a differential gain K_D(t)。

Technical Field

The present invention relates to an inductive feedback control device.

Background

According to the examination of the Japanese national condo, the general product weight (GDP) in Japan is high but the happiness feeling is low. That is, there is a large gap between "material wealth" regarding GDP and "mental wealth" regarding well-being.

As one of measures for compensating for such a gap, it is conceivable to increase the amount of mental wealth by allowing a highly advanced "substance" (welfare assistance device or the like) to perform an action in consideration of the human sense (for example, patent document 1). Further, studies have been conducted on a visualization technique of "perceptual (sensory)" in anticipation of social practices (for example, patent document 2).

However, most of the studies on the sensitivity are studies in a static field such as design evaluation and design of products, and there are few studies on dynamically controlling the sensitivity of an operator particularly for a device such as an automobile or a construction machine operated by the operator.

Documents of the prior art

Patent document

Patent document 1: japanese patent laid-open publication No. 2018-36773

Patent document 2: japanese patent laid-open publication No. 2017-74356.

Disclosure of Invention

The invention aims to provide a sensitivity feedback control device which can feedback control the sensitivity of an operator in equipment operated by the operator.

An inductive feedback device according to an aspect of the present invention includes: an object device operated by an operator; an perceptual detector for detecting vital sign information x (t) of the output u (t) of the operator for the subject device based on which vital sign information x (t) is derivedInformation x (t) determines the comfort y (t) of the operator; a1 st control unit for determining a2 nd target value w (t) for the output u (t) based on a difference between a1 st target value r (t) for the comfort level y (t) and the comfort level y (t); a2 nd control unit for determining a control input v to the target device based on a difference between the 2 nd target value w (t) and the output u (t)_c(t); a weight setting unit configured to set a weight of the operation input v to the target device by the operator_h(t) and the control input v_c(t) weighting the operation levels corresponding to the operators, respectively; and an adder for adding the operation inputs v weighted by the weight setting unit_h(t) and the control input v_c(t) adding and inputting the resultant value to the subject device.

Drawings

Fig. 1 is a configuration diagram of an inductive feedback control apparatus according to an embodiment.

Fig. 2 is a schematic diagram of an inductive multi-axis model used in the inductive feedback control device according to the embodiment.

Fig. 3 is a diagram illustrating database drive type control used in the inductive feedback control device according to the embodiment.

Fig. 4 is a diagram showing effects obtained by the inductive feedback control device according to the embodiment.

Fig. 5 is a simplified diagram showing the configuration of the inductive feedback control device shown in fig. 1 in which δ is 1.

Fig. 6 is a block diagram of the FRIT used in the inductive feedback control device according to the embodiment.

FIG. 7 is a graph showing the velocity error e in the brain of the operator_h(t) a graph of the relationship to comfort y (t).

Fig. 8 is a graph showing comfort y (t), output u (t) (speed), and control input v (t) (torque) at each step (time) in the present numerical example.

Fig. 9 is a graph showing the control parameter (PID gain) adjusted in each step (time).

Fig. 10 is a configuration diagram of an inductive feedback control apparatus according to a comparative example.

Fig. 11 is a graph showing the result of the perceptual feedback control performed by the database drive type control while adjusting the control parameter of the control unit in the comparative example.

Detailed Description

Hereinafter, an inductive feedback control device according to an embodiment of the present invention will be described with reference to the drawings. The scope of the present invention is not limited to the following embodiments, and may be arbitrarily changed within the scope of the technical idea of the present invention.

Fig. 1 is a configuration diagram of an inductive feedback control apparatus according to an embodiment. As shown in fig. 1, the perceptual feedback control apparatus 100 includes a target device 10, a perceptual detector (sensitivity meter)20, a1 st control unit 30, a2 nd control unit 40, a δ setting unit 50, a database 60, a control input adjustment unit 70, and an operation input adjustment unit 80. The δ setting unit 50, the control input adjusting unit 70, and the operation input adjusting unit 80 are examples of the weight setting unit. The perceptual feedback control device 100 further includes subtracters a1 and a2, and an adder A3. The target device 10 is a device operated by an operator, such as an automobile or a construction machine. In fig. 1, the 1 st controller 30, the 2 nd controller 40, the sensitivity detector 20, the subtractors a1, a2, and the adder A3 are each constituted by a processor such as a CPU. The database 60 is constituted by a nonvolatile memory such as a solid state drive or a hard disk drive, for example.

The perceptual detector 20 detects vital sign information x (t) of the operator corresponding to the output u (t) of the subject device 10, and decides the comfort y (t) of the operator based on the vital sign information x (t). The output u (t) of the target device 10 is, for example, the response speed of the hydraulic excavator.

The subtractor a1 calculates a difference e (t) between the 1 st target value (target comfort level r (t)) and the comfort level y (t), and inputs the difference e (t) to the 1 st control unit 30.

The 1 st controller 30 determines the 2 nd target value (target output w (t)) for the output u (t) based on the difference e (t) calculated by the subtractor a1, and inputs the target value to the subtractor a 2.

The subtractor a2 calculates a difference value ∈ (t) between the target output w (t) and the output u (t), and inputs the difference value ∈ (t) to the 2 nd control unit 40.

The 2 nd control unit 40 determines the control input v to the target device 10 based on the difference epsilon (t) calculated by the subtractor a2_c(t) and inputs the result to the control input adjustment unit 70.

The δ setting unit 50 sets an operation input v to the target device 10 for the operator_h(t) weight value (1- δ). The delta setting unit 50 sets the control input v to be inputted_c(t) weight value (δ). The δ setting unit 50 sets a weight value (1- δ) and a weight value (δ) according to the operation level of the operator.

The control input adjusting unit 70 sets the delta setting unit 50 for the control input v_c(t) multiplying the weight value (δ) by the control input v_c(t) and inputs the result to adder a 3.

The operation input adjustment unit 80 controls the δ setting unit 50 to set the target operation input v_h(t) multiplying the weight value (1- δ) by the operation input v_h(t) and inputs the result to adder a 3.

The adder a3 adds δ · v input from the control input adjustment unit 70_c(t) and (1-delta) · v input from the operation input adjustment unit 80_h(t) add up, and input the added value to the subject apparatus 10.

The subject device 10 acquires the added value input from the adder a3, and calculates an output u (t) from the added value.

Next, the details of the sensitivity detector 20 will be described.

(sensitivity detector)

In this embodiment, the perceptual detector 20 calculates the comfort level y (t) based on the vital sign information x (t) of the operator corresponding to the output u (t) of the subject device. The vital sign information x (t) of the operator is, for example, facial expression of the operator, skin resistance of the operator, upper limb myoelectricity of the operator, voice information indicating voice of the operator, and the like. The perceptual detector 20 may detect the facial expression of the operator, for example, by a CCD camera. The perceptual detector 20 may detect the heart rate, e.g. by a heart rate sensor. The perceptual detector 20 may detect the respiration (number or depth), for example by means of a CCD camera. The inductive detector 20 detects the skin resistance via the skin impedance sensor. The sensory detector 20 may detect upper limb myoelectricity and lower limb myoelectricity by a myoelectricity sensor, for example. The perceptual detector 20 may detect the speech information by means of a microphone.

The emotions or emotions of a mandible sparrow leap or joy or heart lifting gallbladder or \2428andthe feeling or emotion of the medicine core are sometimes embraced by people who see or hear what or touch. Such feelings or emotions are caused based on the complex and advanced brain activities of human beings. The formation of emotion or emotion is closely related to the somatic nervous system including motor nerves and sensory nerves, the autonomic nervous system including sympathetic nerves and parasympathetic nerves, and memory or experience. Therefore, perceptual (sensory) is defined as a higher-level brain function that overlooks, at a higher level, an emotional response that is generated by comparing information obtained by integrating external sensory information (information on the somatic nervous system) and internal sensory information (information on the autonomic nervous system) with past experience and memory. In other words, perceptual perception is a high-level brain function that is swiftly understood based on a comparison of the gap between prediction (impression) and result (sensory information) with experience and knowledge.

The perception of such high-level brain functions needs to be comprehensively understood from various viewpoints. The feeling of comfort can be understood from the viewpoint of "pleasant/unpleasant" whether a person feels pleasant, comfortable, or conversely, whether a person feels unpleasant, or unsmooth. The feeling of sensitivity can be understood from the viewpoint of whether a person is awake, excited, or lively, or conversely, whether a person is blurred, calm, or inactive, "lively/inactive". The feeling of interest can be understood from the viewpoint of "feeling of interest" such as whether or not the person expects or expects what to look like and jump like, or conversely whether or not the feeling of jumping like a peaceful bird of the person is obstructed.

A rosmarin (Russell) ring model that expresses pleasantness/unpleasantness and liveness/inactiveness on 2 axes is known. Emotions can be represented by the ring model. However, perceptual perception is a high-level brain function that compares the difference between predictions (impressions) and outcomes (sensory information) with experience and knowledge. Therefore, the perceptual quality cannot be sufficiently expressed by the existing ring model composed of 2 axes which are pleasant/unpleasant and active/inactive. Therefore, in the present embodiment, for example, a perceptual multi-axis model in which a time axis (e.g., expected perceptual) as the 3 rd axis is added to the phoneme loop model is used to grasp perceptual.

Fig. 2 is a schematic diagram of an inductive multi-axis model used in the inductive feedback control device according to the embodiment. As shown in fig. 2, the perceptual multi-axis model can be expressed with "pleasant/unpleasant" (sentiment price) as the 1 st axis, with "liveness/inactivity" (liveness) as the 2 nd axis, and with "time (expectation)" as the 3 rd axis. The advantage of making the perceptual multi-axis modelling is: by calculating evaluation values of the respective axes and integrating them, it is possible to quantitatively evaluate a wide range of feelings with a blurred concept, that is, to realize visualization. Specifically, brain physiological index values (EEG pleasure, EEG liveness, and EEG expectation) for respective axes are obtained based on brain physiological information for the respective axes of the perceptual multi-axis model. Furthermore, subjective psychological axes representing weighting coefficients (a, b, c) of respective axes of the perceptual multi-axis model are obtained based on subjective statistical data of the subject person. Then, the perceptual value can be evaluated by using the brain physiological index value and the subjective psychological axis and using the following calculation formula. The subjective statistical data of the subject is data indicating a relationship between the brainwaves of the subject in a certain state and the response (subjective) of the subject to the state, and is data learned in advance.

Sensory value = [ subjective psychometric axis ]/[ brain physiological index value ] = a + EEG pleasure + b + EEG liveness + c + EEG expectation sensory

The perceptual detector 20 calculates the brain physiological index value by inputting the vital sign information x (t) detected by the various sensors described above into a predetermined arithmetic expression, for example. Then, the sensitivity detector 20 calculates the sensitivity value by inputting the calculated brain physiological index value into the above calculation formula. The perceptual detector 20 outputs the calculated perceptual value as comfort y (t).

Next, the details of the 1 st control unit 30 will be described.

(1 st control part)

The 1 st control part 30 executes the operation with K_P、K_I、K_DPID control represented by the following formula (1) is used as each control parameter.

Formula 1

The target output w (t) is an operation amount in the PID control. K_P、K_I、K_DThe proportional gain, the integral gain, and the differential gain in the PID control are provided. The proportional gain, integral gain, and differential gain are referred to as PID gains. The difference e (t) is a deviation in PID control. Here, w (t) is adjusted in such a way that the difference e (t) approaches zero.

In the present embodiment, the comfort level y (t) is increased by approximating the comfort level y (t) to the target comfort level r (t), but it is difficult to set the target output w (t) suitable for each operator in the feedback control. For this reason, the present embodiment employs a cascade control system as shown in fig. 1. Thus, the present embodiment can automatically generate the target output w (t) suitable for each operator by giving the target comfort level r (t). In this case, human perception is considered to be a time-varying system or a nonlinear system. Therefore, as shown in FIG. 1, the 1 st control unit 30 performs feedback control using the database 60 (database-driven control: design and measurement control of a nonlinear PID control system by FRIT method, for example, refer to "flank dip" and the like, Vol.52, No.10, pp.885-891(2013) "). The database 60 sequentially stores data necessary for adjusting the PID gain of the 1 st controller 30, such as the target comfort level r (t), the comfort level y (t), and the target output w (t). The target comfort level r (t) takes a predetermined value, for example.

Fig. 3 is a diagram illustrating database drive type control used in the inductive feedback control device according to the embodiment. The database drive type control of the 1 st control unit 30 will be described in detail in the following embodiments. The database drive control according to the present embodiment may use, for example, a learning function using a gradient method, may use offline learning that combines a gradient method and a FRIT (computational Reference Iterative) method, or may use online learning that combines a gradient method and a FRIT method.

For example, in the present embodiment, the control parameter expressed by equation (1) is calculated based on neighborhood data in the neighborhood of the current system state among the data stored in the database 60. The control parameters calculated here are control parameters obtained by on-line learning described in the later-described embodiments. However, this is merely an example, and the control parameter expressed by the equation (1) may be a control parameter obtained by off-line learning described in the embodiment described later.

Next, the details of the 2 nd control unit 40 will be described.

(the 2 nd control part)

The 2 nd control part 40 executes the operation with K_P'、K_I'、K_D' PID control represented by the following formula (2) as a control parameter.

Formula 2

Control input v_c(t) is an operation amount in the PID control. K_P’、K_I’、K_D' proportional gain, integral gain, and differential gain in PID control, respectively. The difference epsilon (t) is a deviation in PID control. Here, the control input v is adjusted so that the difference epsilon (t) becomes zero_c(t)。

The control parameter of the 2 nd control unit 40 is, for example, an existing parameter corresponding to the target device 10. The existing parameters may be fixed constants or variable parameters.

(delta setting unit)

Basically, the δ setting unit 50 sets a relatively low weight value (δ) for a highly skilled operator, and sets a relatively high weight value (δ) for a less skilled operator. The δ setting unit 50 holds skill data in advance, for example, in which skill levels of an operator, which are evaluated in advance at 5 levels, are associated with weight values (δ) corresponding to the skill levels. After the skill level of the operator is input, the δ setting unit 50 may determine a weight value (δ) corresponding to the input skill level based on the skill data. Alternatively, the δ setting unit 50 may evaluate the level of the operator based on the result of an experiment performed by the operator when performing the experiment operation, and set the weight value (δ) based on the result of the evaluation.

(Effect)

As described above, in the perceptual feedback control apparatus 100 according to the present embodiment, the perceptual detector 20 detects the vital sign information x (t) of the operator corresponding to the output u (t) of the target device 10, and determines the comfort level y (t) of the operator based on the vital sign information x (t). The 1 st control unit 30 determines a target output w (t) of the output u (t) based on a difference e (t) between the 1 st target value (target comfort level) r (t) and the comfort level y (t) with respect to the comfort level y (t). The 2 nd control unit 40 determines the control input v to the target device 10 based on the difference epsilon (t) between the target output w (t) and the output u (t)_c(t) of (d). In this way, the sensitivity feedback control apparatus 100 can perform control of feeding back the sensitivity (specifically, comfort level y (t)) of the target device 10 by the operator using the sensitivity detector 20, the 1 st control unit 30, and the 2 nd control unit 40.

Fig. 4 is a diagram showing effects obtained by the inductive feedback control apparatus 100 according to the present embodiment. In the example of fig. 4, the sensory feedback control device 100 is applied to the target equipment 10, which is constituted by a hydraulic shovel, for example. Thus, the target device 10 shown in fig. 4 can perform control for feeding back the sensitivity of the operator. Therefore, the sensory feedback control apparatus 100 can cause the target device 10, which is originally designed regardless of the skill level of the operator, to generate output responses corresponding to operators having various skill levels. Therefore, the sensory feedback control device 100 can improve the comfort of the operator having various skill levels.

Further, the delta setting unit 50, the control input adjusting unit 70, and the operation input are includedThe weight setting section of the entry adjustment section 80 sets an operation input v to the target device 10 by the operator_h(t) and control input v_c(t) weighting the operation input v according to the operation level of the operator_h(t) and control input v_c(t) add up, and input the addition result to the object apparatus 10. Therefore, the weight setting unit can set an appropriate degree of the perceptual feedback according to the operation level of the operator. Thus, the weight setting unit can reduce the degree of reflection of the operation input of the operator in the target device 10 by, for example, increasing the degree of the perceptual feedback with respect to the unskilled operator. In contrast, the weight setting unit can improve the degree of reflection of the operation input of the operator in the target device 10 by reducing the degree of the perceptual feedback with respect to the skilled operator.

In the perceptual feedback control apparatus 100 according to the present embodiment, the perceptual detector 20 may detect 3 pieces of information, i.e., the emotional price (value), the liveness, and the expected feeling of the operator, as the vital sign information x (t), and may determine a perceptual value obtained from the correlation between the 3 pieces of information, as the comfort level y (t). In this case, since the "comfort level" as the sensitivity is quantitatively evaluated, the accuracy of the sensitivity feedback control is improved. In this case, the 3 pieces of information and the perceptual value obtained based on the correlation of the 3 pieces of information are calculated, for example, by machine learning using AI.

The perceptual feedback control apparatus 100 according to the present embodiment may further include a database 60 that sequentially stores the target comfort level r (t), the target output w (t), and the comfort level y (t), and the 1 st control unit 30 may determine the target output w (t) while adjusting the control parameter using the data stored in the database 60. In this case, without modeling the sensitivity of the operator as a nonlinear system, the sensitivity feedback control can be performed by using the database-driven control.

In the inductive feedback control apparatus 100 according to the present embodiment, when the target device 10 is a construction machine, the operator can operate the actual construction machine so that the output u (t) equivalent to the target output of the construction machine (for example, the target response speed of the hydraulic shovel) in the brain can be obtained.

In the sensory feedback control device 100 according to the present embodiment, the learning function may be used for the database drive type control of the 1 st control unit 30. In this case, as the operation time of the subject apparatus 10 by the operator increases, the subject apparatus 10 gradually changes to be able to perform an output response more suitable for the operator. Thereby, the sensory feedback control device 100 can further improve the comfort of the operator.

(examples)

An embodiment of the inductive feedback control device 100 is described below. In this embodiment, δ is set to "1", and only the control input v is controlled_c(t) is input to the object device 10. In this embodiment, the database drive type control performed by the 1 st control unit 30 will be mainly described.

Fig. 5 is a diagram illustrating a configuration in the case where δ is set to 1 in the inductive feedback control device 100 shown in fig. 1. In fig. 5, the same components as those in fig. 1 are denoted by the same reference numerals. In fig. 5, illustration of the δ setting unit 50, the control input adjustment unit 70, the operation input adjustment unit 80, and the adder a3 is omitted.

In the present embodiment, the target device 10 is a hydraulic excavator. The output u (t) of the target device 10 is the response speed of the bucket of the hydraulic excavator. Control input v to the subject device 10_c(t) is torque. The response speed is measured, for example, by a sensor. The response speed may be a response speed of other members other than the bucket, such as a boom and an arm, instead of the response speed of the bucket. That is, the response speed is a response speed of the working device such as the bucket, the arm, and the boom.

In the present embodiment, the comfort level y (t) of the operator as the control target is regarded as a nonlinear system, and the database drive type control is applied to the 1 st control unit 30. In addition, the 2 nd control part 40 of the inner ring is set to K_P’=1.5、K_I’=0.1、K_D' =0.1 performs PID control shown in equation (2). Therefore, in the present embodiment, the control parameter of the 2 nd control unit 40 is not the adjustment target.

In the present embodiment, the comfort level y (t) to be controlled is represented by a discrete-time nonlinear system (hereinafter, simply referred to as "system") represented by the following formula (3).

Formula 3

In equation (3), the comfort level y (t) is the output of the system. f () is a non-linear function. Phi (t-1), called the information vector, represents the state of the system before time t. The information vector φ (t-1) is defined by the following equation (4).

Formula 4

Here, the target output w (t) is the input to the system for control of the comfort y (t). n is_yIs the number of outputs of the system. n is_wIs the number of inputs to the system. In the database drive type control, each operation data is stored in the database 60 (see fig. 3) in the form of equation (4). Phi (t) is an information vector that represents the current state of the system and is called a request point (query).

As a control law of the 1 st control unit 30, a velocity type I-PD control law represented by the following formula (5) is used.

Formula 5

In the formula (5), e (t) is a control error signal. When r (t) is the target comfort level, e (t) is defined by the following formula (6).

Formula 6

In the formula (5), K_P(t)、K_I(t)、K_D(t) is in each step (time t)Proportional gain, integral gain, and differential gain. And, Δ (: = 1-z)^-1) Is the difference operator. z is a radical of^-1Is a delay operator.

In the present embodiment, the 1 st control unit 30 uses the FRIT method and uses the input/output data w obtained by 1 operation experiment₀(t) and y₀(t) and pseudo-reference inputs to r (t) generated based on these experimental data, and directly calculates the control parameters of the 1 st control unit 30. Fig. 6 is a block diagram of FRIT used in the present embodiment. In fig. 6, the System includes the 2 nd control unit 40, the target device 10, and the sensitivity detector 20. C (z)^-1) Is a controller. The controller can be represented by the following formula (7).

Formula 7

In the formula (7), n is the number of control laws. In the case of the PID control law, n = 2. As shown in FIG. 6, C (z)^-1) The input/output relationship (c) is expressed by the following expression (8).

Formula 8

Pseudo-reference input based on equation (8)^～r is calculated as the following expression (9).

Formula 9

In the FRIT method, as shown in fig. 6, a reference model Gm (z) having a desired feature designed in advance is used^-1). Based on input of pseudo-reference^～reference model Gm (z) of r^-1) Output of (2)^～y_m(t) the 1 st control part 30 outputs^～y_m(t) and y₀The control parameter is calculated so that the error of (t) becomes small. Reference model Gm (z)^-1) The characteristic polynomial of the following formula (10) or (11)The formula is shown.

Formula 10

Formula 11

In the equation (11), α represents the rising characteristic of the control system. σ represents a parameter having a relationship with the attenuation characteristic. α and σ can be set arbitrarily. T is_sIs the sampling time. Specifically, σ is the time until the output of the control system reaches about 60% of the target value of the step (step) expression. Further, α is preferably set in a range of 0 to 2.0. α =0 represents the response of Binomial model. α =1 represents the response of the Butterworth model.

Hereinafter, a design procedure of the database-driven PID controller will be described.

< Generation of initial database >

In the database-driven control, when there is no past stored data, the design of a local controller (local controller) cannot be performed in principle. Therefore, in the present embodiment, based on the input/output data obtained around a certain balance point, the Zieglar & nichols (zn) method, the Chien, Hrones & rewick (chr) method, or the like is used to calculate the PID gain (control parameter), and an initial database including an information vector (represented by the following expression (12)) composed of the PID gain and the aforementioned input/output data is generated. The ZN method is disclosed in the literature "J.G.Zieglar et al, optics settings for automatic controllers, trans.ASME, Vol.64, No.8, pp.759-768 (1942)". The CHR method is disclosed in the literature "K.L.Chien et al, On the Automatic Control of Generalized Passive Systems, trans.ASME, Vol.74, pp.175-185 (1972)". The input-output data obtained around the balance point is, for example, a data set of the same characteristics.

Formula 12

In formula (12), j =1, 2, · M, N (0), i =1, 2, · M,^-phi (j) and K (j) are given by the following expressions (13) and (14), respectively.

Formula 13

Formula 14

N (0) is the initial data amount (the number of information vectors in the initial database). M is the number of elements. Since the PID gain in the initial database is fixed, K (1) = K (2) = · · · = K (N (0)).

When the system is not operating, the above processing for generating the initial database is executed by using the operation data stored in the database 60 during the operation of the system.

< calculation of distance, selection of neighborhood >)

Request point^-Phi (t) and information vector stored in database^-The distance between phi (j) is weighted by L represented by the following equation (15)₁The norm.

Formula 15

N (t) is the amount of data (the number of information vectors) stored in the database at time t.^-φ_l(j) Is the l-th element of the j-th information vector.^-φ_l(t) is the l-th element of the request point at time t. max^-φ_l(m) is all information vectors present in the database: (m)^-φ (j): j = the largest element of the l-th elements of 1, 2, ·, n (t)). min^-φ_l(m) isThe smallest element among the ith elements.

In this embodiment, k information vectors are selected from the database in ascending order of distance d as shown in equation (15), and the selected data set is defined as neighborhood data.

< architecture of local controller >

Next, in the present embodiment, a local controller is configured by applying a local Weighted linear Average (LWA) represented by the following formula (16) to the neighborhood data selected as described above.

Formula 16

Here, w_iThe weight of k (i) included in the i-th information vector of the neighborhood data is given by the following expression (17).

Formula 17

The PID gains at the respective times t are calculated by the above steps. In addition, in order to enable the database-driven control system to appropriately adjust the PID gain at each balance point, it is necessary to perform learning of the database (update of the control parameter). For this reason, in the present embodiment, FRIT is applied, and the PID gains in the respective data sets within the database are updated offline by learning from the initial data used for the construction of the database. The offline state is a state before the system is operated, and is, for example, a non-operation state of the target device 10.

Offline learning of database-driven control using FRIT

The off-line learning of the database-driven control by FRIT will be specifically described below. First, to calculate a request point in closed-loop data^-φ₀PID gain in (t), the distance between the request point and the information vector in the database 60 is calculated by equation (15), and k neighborhood data are selected from the calculation result. Then, byEquation (16) calculates the PID gain. The steepest descent method represented by the following equations (18) and (19) is applied to the calculated PID gain. Thus, PID gain K is performed^old(t) learning to derive a new K^new。

Formula 18

Formula 19

In the expressions (18) and (19), η is a learning coefficient, and J (t +1) is an evaluation criterion defined by the following expressions (20) and (21).

Formula 20

Formula 21

Wherein the content of the first and second substances,^～y_m(t) is designed as shown in the following formula (22).

Formula 22

In the formula (22), G_m(1)=1+p₁+p₂(see formula (10)).

The partial differential values of the right-hand term 2 of equation (18) are developed as equation (23).

Formula 23

In the formula (23), Γ (t) is represented as the following formula (24).

Formula 24

The pseudo-reference input is included in equation (23)^～r (t), equation (18) becomes an offline learning based on FRIT. Using K obtained from formula (18)^newAnd (t) updating each neighborhood data in the database. This step is repeated until the evaluation criterion expressed by the formula (20) becomes sufficiently small. Thereby, an optimal database is obtained. When the database-driven control is applied to the system, the local controller is configured by the steps described in "generation of initial database", "calculation of distance, selection of neighborhood", and "configuration of local controller" for each step (time). Thereby, more effective control performance for the nonlinear system can be obtained.

Database-driven controlled online learning using FRIT

The PID gain obtained by the offline learning described above is further updated by the online learning described below. Online learning refers to learning while the system is online. Online is a state that indicates that the system is in operation.

In the online learning, the steepest descent method shown in equation (25) and equation (26) is applied to the PID gain obtained by the offline learning.

Formula 25

Formula 26

In expressions (25) and (26), η learning coefficient J (t +1) is an evaluation criterion defined by the following expressions (27) and (28). In formula (28), y is represented by formula (21)₀This point that (t) becomes y (t) is different from off-line learning。

y_m(t) is represented by formula (29).

Formula 27

Formula 28

Formula 29

Gm(z^-1) The characteristic polynomial of the reference model is expressed by the equations (30) and (31).

Formula 30

Formula 31

δ is the rising characteristic of the control system. σ is a parameter that has a relationship with the attenuation characteristics. δ and σ are set by the designer at will. The partial differential of the right-hand term 2 of equation (25) is represented by equation (32).

Formula 32

The PID gain calculated by the online learning is applied as a control parameter of equation (1).

< numerical examples >

Hereinafter, a numerical example of the perceptual feedback control according to the present embodiment will be described.

Here, the target equipment (equipment operated by an operator) 10 shown in fig. 5 is a hydraulic excavator. In this case, the target device 10 is represented by a first-order lag system of the following expression (33).

Formula 33

On the other hand, the comfort level y (t) is expressed by weber fisher's law expressed by the following formula (34) and formula (35) (see "i.p. herman, Physics of the Human Body: Biological and Medical Physics, Biological Engineering, spring-Verlag GmbH & co. kg (2007)"). In the formulas (34) and (35), the maximum value of comfort level y (t) is 1.

Formula 34

Formula 35

In the formulae (34) and (35), w_h(t) is the target speed of the hydraulic excavator that the operator has in the brain, however, this is unknown to the control system. e.g. of the type_h(t) is a speed error of the hydraulic shovel which is sensed by the brain of the operator.

According to equation (34), if the speed error e in the operator's brain_h(t) is completely zero, the comfort level y (t) becomes 1 at the maximum. E (t) is a variable related to the comfort level y (t) and has a value that varies depending on the operator. FIG. 7 is a graph showing the velocity error e in the brain of the operator_h(t) a graph of the relationship to comfort y (t). As shown in fig. 7, the speed error e_hThe greater (t) the less comfort y (t). In addition, the greater E (t), the greater the rate of decline of comfort y (t).

In the numerical examples described below, the respective setting parameters r =0.8, w_h=40，σ=10，α=0，η=［80、60、80］。

The existing control parameters are used for the 2 nd control unit 40 of the inner ring. For the adjustment of the control parameters of the outer ring 1 st control unit 30, database-driven control effective for the nonlinear system (refer to "design, measurement, and control of the nonlinear PID control system by FRIT method, vol.52, No.10, pp.885-891 (2013)") is used.

Furthermore, for obtaining initial data { u }₀、y₀The PID gain of the outer loop (the 1 st control unit 30) is K_P=3.5，K_I=0.5，K_D= 3.5. The PID gain of the inner loop (2 nd control part 40) is K_P’=1.5，K_I’=0.1，K_D’=0.1。

In the above setting, the perceptual feedback control is performed by the database drive type control while adjusting the control parameter of the 1 st control unit 30. FIG. 8 shows comfort y (t), output u (t) (speed), and control input v at each step (time) in the present numerical example_c(t) (torque) graph. Fig. 9 is a graph showing the control parameter (PID gain) adjusted in each step (time).

As shown in equation (34), the comfort y (t) is a nonlinear system. Therefore, as shown in the upper graph of fig. 8, the initial data y obtained in the state where the PID gain is fixed₀The target comfort level r (=0.8) is not followed. On the other hand, the present embodiment adjusts the PID gain by the database-driven type control. Therefore, as shown in the upper graph of fig. 8, the comfort level y follows the target comfort level r. As can be seen, the inductive feedback control is realized in the present embodiment. This is based on adjusting the PID gains accordingly as shown in fig. 9.

Target output w in the operator's brain_h(= 40: target speed) is unknown to the control system, but the target output w (target speed) obtained by the present embodiment ends up being 40 (graph in the center of fig. 8). In this case, the target speed in the brain of the operator can be estimated by analyzing the database.

Further, as shown in the lower graph of fig. 8, when the load is large, the control input v is large accordingly_c(t)(torque) is automatically calculated.

Comparative example

Hereinafter, a comparative example of the inductive feedback control device according to the present embodiment will be described. The comparative example realizes the inductive feedback control only by the outer loop. That is, this comparative example omits a control loop regarding the inside of the subject apparatus. Hereinafter, the effectiveness of the cascade control system (outer loop + inner loop) in the inductive feedback control according to the present embodiment will be described with reference to this comparative example.

Fig. 10 is a configuration diagram of an inductive feedback control apparatus according to a comparative example. As shown in fig. 10, the sensitivity feedback control apparatus 200 according to the comparative example includes a target device 10, a sensitivity detector 20, a control unit 30A, and a database 60A. The target equipment 10 is a hydraulic excavator operated by an operator in the same manner as the above-described embodiment. The inductive detector 20 is the same as the inductive detector 20 of the inductive feedback control apparatus 200 of the present embodiment shown in fig. 1. The perceptual detector 20 detects vital sign information x (t) of the operator corresponding to the output u (t) of the subject device 10, and decides the comfort y (t) of the operator based on the vital sign information x (t). The control unit 30A determines the control input v to the target device 10 based on the difference e (t) between the target value (target comfort level) r (t) and the comfort level y (t) with respect to the comfort level y (t)_c(t)。

The control part 30A uses K, for example_P"、K_I"、K_D"PID control represented by the following equation (36) is executed as PID gains, respectively.

Formula 36

The control unit 30A performs database drive control using the database 60A. The database 60A sequentially stores data necessary for adjusting the PID gain of the control unit 30A, such as the target comfort level r (t), the comfort level y (t), and the control input v_c(t), etc.

In the present comparative example, the output u (t) of the target equipment 10 is also the response speed of the bucket of the hydraulic excavator, and the control input v (t) to the target equipment 10_c(t) is also torque.

Fig. 11 is a graph showing the result of the inductive feedback control performed by the database drive type control while adjusting the control parameters of the control unit 30A in the comparative example. In fig. 11, the upper graph shows comfort level y (t) at each step (time). The central graph shows the output u (t) (speed) of each step (time). The lower graph shows the control input v for each step (time)_c(t) (torque). For obtaining initial data u₀、y₀K represents the PID gain of the control unit 30A in the same manner as in the foregoing embodiment_P=3.5，K_I=0.5，K_D= 3.5. The other parameters are r =0.8 and w are the same as in the above-described embodiment_h=40，σ=10，α=0，η=［80、60、80］。

As shown in the upper graph of fig. 11, in the comparative example, the PID gain is adjusted by the database-driven type control, but the comfort level y does not follow the target comfort level r. That is, the perceptual feedback control according to the comparative example cannot achieve sufficient control performance.

From this result, in the inductive feedback control according to the present embodiment, the compliance of the inner loop (the 2 nd control unit 40) is improved by the cascade control system. This embodiment facilitates the design of the control system, and improves the accuracy of the perceptual feedback control.

While the embodiments (including examples) of the present invention have been described above, the present invention is not limited to the above-described embodiments, and various modifications can be made within the scope of the present invention. That is, the above description of the embodiments is merely an example of the nature, and the present invention is not intended to limit the application objects or the uses thereof.

For example, although the inductive feedback control device has been described in the present embodiment by taking a hydraulic excavator (construction machine) as an example, the device can be applied to other devices operated by an operator.