Data device, learning data generation device, data conversion program, learning data generation program, and control system

Abstract

This facilitates the construction of machine learning models used to control mechanical devices. [Solution] The data device comprises a reference conversion unit and a reconversion unit. The reference conversion unit receives reference operation parameters output by a reference machine device control model constructed by machine learning operations on a reference machine device, and converts them into operation parameters that are independent of individual machine devices and outputs them. The reconversion unit reconverts the operation parameters into control operation parameters, which are operation parameters of the same type as the reference operation parameters, according to the mechanical characteristics of the machine device to be controlled.

Description

[Technical Field]

[0001] This disclosure relates to a data device for controlling a mechanical device, a learning data generation device, a data conversion program, a learning data generation program, and a control system. [Background technology]

[0002] Patent Document 1 describes a neural network system for autonomously driving an autonomous vehicle. In recent years, data-driven machine control methods based on machine learning have attracted attention as a means of achieving appropriate control of machine control devices in complex situations involving various factors. While conventional rule-based and model-based methods build algorithms based on prior knowledge and mathematical models, machine learning builds algorithms based on data, hence the name "data-driven." Data-driven planners do not require prior knowledge or mathematical models, but they do require a vast amount of training data for the machine learning model to acquire functionality. A data-driven planner takes information about the surroundings of the machine and the state of the machine as input, and outputs the amount of operation of the machine. [Prior art documents] [Patent Documents]

[0003] [Patent Document 1] Special Publication No. 2019-533810 [Overview of the Initiative] [Problems that the invention aims to solve]

[0004] Detailed consideration by the inventors revealed the following problems: Model-based planners clearly distinguish between the roles of recognition, judgment, and operation, and both the parameters within each category and the variables exchanged between categories are interpretable and clear. Therefore, it is easy to modify parameters in response to design changes. On the other hand, data-driven planners do not have a clear distinction between the roles of recognition, judgment, and operation, and are composed of large-scale neural network models. Therefore, the parameters and variables within data-driven planners are difficult to interpret. As a result, it is difficult to modify parameters in response to design changes. For this reason, with data-driven planners, it is necessary to change the training data and retrain every time the specifications of the machine are changed. Furthermore, if the planner is applied to a machine of a different model than the one from which the training data was acquired, the desired behavior may not be obtained due to the different characteristics of each machine. In this case, it becomes necessary to restart the training process from scratch for each machine.

[0005] This disclosure aims to facilitate the construction of machine learning models used to control mechanical devices. [Means for solving the problem]

[0006] One aspect of the present disclosure is a data device (A1,2) that includes a reference conversion unit (A2,13) ​​and a reconversion unit (A3,14), as shown in Figure 9, and outputs control operation parameters reconverted by the reconversion unit to a controlled machine (A5).

[0007] The reference conversion unit is configured to take at least one reference operation parameter, which is an operation parameter output by a reference machine learning model (A4,11) that is a machine learning model constructed by machine learning operations on a reference machine learning device, as input, and convert it into at least one operation parameter that is independent of individual machine devices, using a machine device model that is a physical model based on the mechanical characteristics of the reference machine device, and output the converted at least one operation parameter.

[0008] The reconversion unit is configured to reconvert at least one operating parameter output by the reference conversion unit to at least one control operating parameter which is the same type as at least one reference operating parameter, according to the mechanical characteristics of the controlled machine device which is the machine device being controlled.

[0009] The data device of this disclosure, configured in this manner, outputs control operation parameters generated by converting the reference operation parameters output by the reference machine device control model to the machine device to be controlled. As a result, the control device of this disclosure does not need to retrain and construct a new machine device control model that outputs operation parameters by machine learning operations on the machine device to be controlled in order to control the machine device to be controlled. Therefore, the data device of this disclosure can eliminate the need to construct a machine device control model for each of the multiple machine devices to be controlled that are of different types, and can facilitate the construction of machine device control models used to control the machine device.

[0010] Another aspect of the present disclosure is a training data generation device (B1, 100) for generating training data for machine learning, as shown in Figure 10, comprising a collection and conversion unit (B2, 101) and a generation unit (B3, 102). The data collection and conversion unit is configured to receive at least one data collection operation parameter, which is an operation parameter for operating the data collection target machine (B4), which is a machine device for performing data collection, and to convert it into at least one operation parameter that is independent of the individual machine device, using a machine device model, which is a physical model based on the mechanical characteristics of the data collection target machine. The unit is configured to output the converted at least one operation parameter.

[0011] The generation unit is configured to generate training data by converting at least one operation parameter output by the collection and conversion unit back into at least one reference operation parameter, which is the same type of operation parameter as at least one collection operation parameter, according to the mechanical characteristics of the reference machine, which is a reference machine.

[0012] The learning data generation device of the present disclosure configured as described above can generate learning data for a reference machine device control model constructed by machine learning the operations on the reference machine device, using a plurality of target collection machine devices with different models, and thus can facilitate the construction of a machine device control model used to control the machine device.

[0013] Another aspect of the present disclosure is a data conversion program for causing a computer to function as a reference conversion unit (13) and a reconversion unit (14), and outputting the control operation parameters reconverted by the reconversion unit to the controlled machine device.

[0014] The computer controlled by the data conversion program of the present disclosure can form part of the data device of the present disclosure and can obtain the same effects as the data device of the present disclosure. Another aspect of the present disclosure is a learning data generation program for causing the computer of a learning data generation device (100) that generates learning data for machine learning to function as a collection conversion unit (101) and a generation unit (102).

[0015] The computer controlled by the learning data generation program of the present disclosure can form part of the learning data generation device of the present disclosure and can obtain the same effects as the learning data generation device of the present disclosure.

[0016] Another aspect of the present disclosure is a control system (2) that includes a reference conversion unit (13) and a reconversion unit (14), and controls a controlled machine device based on the control operation parameters reconverted by the reconversion unit.

[0017] The control system of the present disclosure is a system including the data device of the present disclosure and can obtain the same effects as the data device of the present disclosure.

Brief Description of Drawings

[0018] [Figure 1]This is a block diagram showing the configuration of a vehicle control system. [Figure 2] This is a functional block diagram showing the functional configuration of the vehicle control system. [Figure 3] This is a block diagram showing the configuration of a compensator. [Figure 4] This graph shows the trajectory of a vehicle when turning left at an intersection. [Figure 5] This graph shows the change in steering input over time when a vehicle turns left at an intersection. [Figure 6] This graph shows the time change in yaw rate when a vehicle turns left at an intersection. [Figure 7] This is a block diagram showing the configuration of a training data generation device. [Figure 8] This is a functional block diagram showing the functional configuration of the training data generation device. [Figure 9] This block diagram shows the configuration of the data device in this disclosure. [Figure 10] This is a block diagram showing the configuration of the learning data generation device in this disclosure. [Modes for carrying out the invention]

[0019] [First Embodiment] A first embodiment of this disclosure is described below with reference to the drawings. The vehicle control system 1 of this embodiment is installed in a vehicle capable of autonomous driving. Autonomous driving means that the vehicle's driving operations are performed automatically on behalf of the vehicle's occupants. The vehicle control system 1 enables, for example, Level 3 or higher autonomous driving. The level of autonomous driving refers to the autonomous driving levels defined by the Society of Automotive Engineers (SAE).

[0020] A vehicle equipped with the vehicle control system 1 may have both an automatic driving function and a manual driving function. The vehicle may also be a hybrid vehicle having both an engine and an electric motor as its driving source. The vehicle is not limited to vehicles with an automatic driving function or hybrid vehicles; it may also be a vehicle having only an engine or only an electric motor as its driving source. Hereinafter, a vehicle equipped with the vehicle control system 1 will be referred to as a controlled vehicle.

[0021] As shown in Figure 1, the vehicle control system 1 comprises a vehicle control device 2 and an actuator 3. The vehicle control device 2 is an electronic control device centered around a microcomputer equipped with a CPU 2a, ROM 2b, RAM 2c, and GPU 2d, etc. The various functions of the microcomputer are realized by the CPU 2a executing a program stored in a non-transitional physical recording medium. In this example, ROM 2b corresponds to the non-transitional physical recording medium that stores the program. Furthermore, the execution of this program executes the method corresponding to the program. Note that some or all of the functions performed by the CPU 2a may be configured in hardware using one or more ICs, etc. Also, the number of microcomputers constituting the vehicle control device 2 may be one or more.

[0022] The vehicle control device 2 receives sensor data generated by one or more sensors (not shown) that detect the surrounding conditions of the controlled vehicle and the condition of the controlled vehicle, and outputs a target control quantity. Examples of sensor data include camera images from an on-board camera and vehicle operation quantities. It may also be sensor information for understanding the surrounding environment of the controlled vehicle and the condition of the controlled vehicle, such as Lidar or radar. In the following examples of control quantities, acceleration / deceleration ~a(t) and front wheel steering angle ~δ(t) are given, but acceleration, accelerator opening, brake opening, brake fluid pressure, steering wheel angle, etc. may also be used.

[0023] Actuator 3 controls acceleration and deceleration. ~ a(t) and front wheel steering angle ~Based on δ(t), the accelerator, brakes, and steering system of the controlled vehicle are activated. As shown in Figure 2, the vehicle control device 2 includes a data-driven planner 11 and a model-following controller 12 as functional blocks realized by the CPU 2a executing a program stored in ROM 2b.

[0024] The data-driven planner 11 includes a learning model generated by machine learning using multiple vehicle-related image data of the base vehicle's surroundings and multiple vehicle operation data for operating the base vehicle (e.g., steering operation data, accelerator operation data, brake operation data). The base vehicle is a vehicle that collects the above-mentioned vehicle-related image data and vehicle operation data for training the data-driven planner 11.

[0025] This learning model, for example, takes image data captured by an in-vehicle camera as input data and outputs vehicle operation data. The model-following controller 12 comprises a base vehicle model 13 and a compensator 14.

[0026] The method for realizing these elements constituting the vehicle control device 2 is not limited to software; some or all of these elements may be realized using one or more hardware components. For example, if the above functions are realized by an electronic circuit which is hardware, that electronic circuit may be a digital circuit containing a large number of logic circuits, an analog circuit, or a combination thereof.

[0027] The base vehicle model 13 is a model that takes acceleration / deceleration a(t) and front wheel steering angle δ(t) as inputs and outputs vehicle speed V(t) and yaw rate γ(t), and is constructed with a vehicle model having the same specifications as the vehicle from which the training data was acquired (i.e., the base vehicle). The vehicle model in this embodiment is described by a vehicle mathematical model such as the dynamic two-wheel model shown in equations (1) to (4).

[0028]

number

[0029] Table 1 shows the definitions of the variables and parameters in equations (1) to (4).

[0030] [Table 1]

[0031] Here, the vehicle speed V(t) is updated in equation (5) and is used not only as the output of the base vehicle model 13 but also in equations (1) through (4).

[0032]

number

[0033] The base vehicle model 13 can be thought of as interpreting the output of the data-driven planner 11 with a vehicle model that has a clearly defined internal structure. In this embodiment, the planner output, presented as physical quantities of actuator operation, is interpreted as physical quantities of vehicle behavior, namely vehicle speed and yaw rate.

[0034] The compensator 14 sets the vehicle speed V(t) and yaw rate γ(t), which are the outputs of the base vehicle model 13, as target values, and adjusts the acceleration and deceleration so that the vehicle to which the data-driven planner 11 is actually applied follows the target values. ~ a(t) and front wheel steering angle ~ This is a target value tracking control system that determines δ(t). Note that the relationship between acceleration / deceleration and vehicle speed does not depend on the vehicle specifications, therefore, acceleration / deceleration ~ a(t) is the same value as the output of the data-driven planner 11 (i.e., acceleration / deceleration a(t)). Therefore, in the following, the front wheel steering angle ~ Let's explain how to calculate δ(t).

[0035] The target value tracking control system is implemented by a control system based on a vehicle mathematical model, similar to the base vehicle model 13. For example, it is realized by adaptive control based on a dynamic two-wheel model. The compensator 14 uses the non-linear systems shown in equations (6) and (7).

[0036]

Equation

[0037] Here, x(t) ∈ R n , u(t) ∈ R m , y(t) ∈ R l , G(x) ∈ R n , H(x) ∈ R n×m , C ∈ R l×n where G(x) and H(x) are smooth non-linear functions with respect to x(t).

[0038] As shown in FIG. 3, the compensator 14 includes a subtractor 21, a feedback linearization controller 22, a calculation unit 23, and a multiplier 24. The subtractor 21 outputs a subtraction value obtained by subtracting y(t) output from the multiplier 24 from the target value y r (t) of y(t).

[0039] The feedback linearization controller 22 calculates u(t) from v(t) output by the subtractor 21 and outputs u(t). The calculation unit 23 calculates x(t) from equation (6) based on u(t) output by the feedback linearization controller 22 and outputs x(t).

[0040] The multiplier 24 outputs the multiplication value obtained by multiplying the preset constant C by x(t) output by the calculation unit 23 as y(t). In this embodiment, the longitudinal motion of the vehicle is described using a point mass model, and the lateral motion and rotational motion around the center of gravity are described using a dynamic two-wheel model. x(t), u(t), y(t) and G(x), H(x), C in equations (6) and (7) are defined as shown in equations (8), (9), (10), (11), (12), and (13), respectively. Here, F(t) represents the total driving force. In addition, the vehicle parameters of each model use the specifications of the vehicle to which the planner is applied (i.e., the controlled vehicle).

[0041]

number

[0042]

number

[0043]

number

[0044]

number

[0045] acceleration / deceleration ~ a(t) is calculated using equation (14).

[0046]

number

[0047] Figure 4 shows the results of a vehicle simulation illustrating the trajectories of the base vehicle and control vehicle when turning left at an intersection. Curve L1 in Figure 4 shows the trajectory of the base vehicle. Curve L2 shows the trajectory of the controlled vehicle when the vehicle control device 2 is equipped with a data-driven planner 11 and a model-following controller 12. Curve L3 shows the trajectory of the controlled vehicle when the vehicle control device 2 is equipped with a data-driven planner 11 but not a model-following controller 12.

[0048] As shown in Figure 4, when only the data-driven planner 11 is applied to a control vehicle that is different from the base vehicle, the control vehicle's trajectory deviates significantly from the base vehicle's trajectory. On the other hand, when the data-driven planner 11 and the model-following controller 12 are applied to the control vehicle, the control vehicle's trajectory becomes similar to that of the base vehicle, with the target position G coinciding with the base vehicle's trajectory.

[0049] Figure 5 shows the results of a vehicle simulation illustrating the time change in steering input when the base vehicle and control vehicle turn left at an intersection. Line L11 in Figure 5 shows the time variation of the steering amount of the base vehicle. Line L12 shows the time variation of the steering amount of the controlled vehicle when the vehicle control device 2 is equipped with a data-driven planner 11 and a model-following controller 12. Line L13 shows the time variation of the steering amount of the controlled vehicle when the vehicle control device 2 is equipped with a data-driven planner 11 but is not equipped with a model-following controller 12.

[0050] As shown in Figure 5, when the data-driven planner 11 and the model-following controller 12 are applied to the control vehicle, the steering amount of the control vehicle is corrected relative to the steering amount of the base vehicle.

[0051] Figure 6 shows the results of a vehicle simulation illustrating the time evolution of the yaw rate when the base vehicle and the control vehicle turn left at an intersection. Line L21 in Figure 6 shows the time variation of the yaw rate of the base vehicle. Line L22 shows the time variation of the yaw rate of the controlled vehicle when the vehicle control device 2 is equipped with a data-driven planner 11 and a model-following controller 12. Line L23 shows the target yaw rate output by the data-driven planner 11.

[0052] As shown in Figure 6, the steering input is corrected by model following control, causing the yaw rate of the controlled vehicle to nearly match the target yaw rate. The vehicle control device 2 configured in this way includes a base vehicle model 13 and a compensator 14.

[0053] The base vehicle model 13 is configured to take the acceleration / deceleration a(t) and front wheel steering angle δ(t) output by the data-driven planner 11, which is a machine learning model built by machine learning operations on the base vehicle, as input, and convert them into vehicle speed V(t) and yaw rate γ(t) that are independent of individual vehicles using a dynamic two-wheel model, which is a physical model based on the mechanical characteristics of the base vehicle, and output the converted vehicle speed V(t) and yaw rate γ(t).

[0054] The compensator 14 takes the vehicle speed V(t) and yaw rate γ(t) output by the base vehicle model 13 and adjusts the acceleration / deceleration a(t) and front wheel steering angle δ(t), which are the same type of operating parameters, according to the mechanical characteristics of the controlled vehicle. ~ a(t) and front wheel steering angle ~ It is configured to be converted back to δ(t).

[0055] The vehicle control device 2 converts the acceleration and deceleration ~ a(t) and front wheel steering angle ~ δ(t) is output to the control vehicle. That is, the vehicle control device 2 outputs the reconverted acceleration / deceleration. ~ a(t) and front wheel steering angle ~ The control vehicle is controlled based on δ(t).

[0056] The base vehicle model 13 is constructed using a vehicle model with the same specifications as the base vehicle, taking acceleration / deceleration a(t) and front wheel steering angle δ(t) as inputs and vehicle speed V(t) and yaw rate γ(t) as outputs.

[0057] The compensator 14 sets the movement indicated by the yaw rate γ(t) as the target value and adjusts the front wheel steering angle so that the controlled vehicle moves in accordance with the target value. ~ A target value tracking control is performed to determine δ(t). Such a vehicle control device 2 generates acceleration by converting the acceleration / deceleration a(t) and front wheel steering angle δ(t) output by the data-driven planner 11. ~ a(t) and front wheel steering angle ~ δ(t) is output to the control vehicle. As a result, the vehicle control device 2 controls the control vehicle by learning the operations on the control vehicle to determine acceleration and deceleration. ~ a(t) and front wheel steering angle ~ There is no need to re-train and rebuild the data-driven planner 11 that outputs δ(t). Therefore, the vehicle control device 2 can avoid the need to build a data-driven planner 11 for each of the multiple control vehicles, which are of different types, and can easily build the data-driven planner 11 used to control the control vehicles.

[0058] In the embodiments described above, the vehicle control device 2 corresponds to a data device and control system, the base vehicle model 13 corresponds to a reference conversion unit, the compensator 14 corresponds to a reconversion unit, and the data-driven planner 11 corresponds to a reference machine device control model.

[0059] Furthermore, the base vehicle corresponds to the standard mechanical device, the acceleration / deceleration a(t) and front wheel steering angle δ(t) correspond to the standard operating parameters, the dynamic two-wheel model corresponds to the mechanical device model, and the vehicle speed V(t) and yaw rate γ(t) correspond to the operating parameters.

[0060] Furthermore, the control vehicle corresponds to the controlled mechanical device, and acceleration / deceleration. ~ a(t) and front wheel steering angle ~δ(t) corresponds to the control operation parameter.

[0061] [Second Embodiment] A second embodiment of this disclosure is described below with reference to the drawings. The learning data generation device 100 in this embodiment is a device that generates learning data for performing machine learning on the data-driven planner 11, and as shown in Figure 7, it is mainly composed of a microcomputer equipped with a CPU 100a, ROM 100b, RAM 100c, and GPU 100d.

[0062] The various functions of the microcomputer are realized by the CPU 100a executing a program stored in a non-transitional physical recording medium. In this example, the ROM 100b corresponds to the non-transitional physical recording medium that stores the program. Furthermore, the execution of this program executes the method corresponding to the program. Note that some or all of the functions executed by the CPU 100a may be configured in hardware using one or more ICs, etc. Also, the number of microcomputers constituting the learning data generation device 100 may be one or more.

[0063] As shown in Figure 8, the movement of the data collection vehicle generates multiple control data OD1 (e.g., steering operation data, accelerator operation data, brake operation data) for controlling the data collection vehicle.

[0064] The learning data generation device 100 includes a collection vehicle model 101 and a base vehicle compensator 102 as functional blocks realized by the CPU 100a executing a program stored in ROM 100b.

[0065] The data collection vehicle model 101, like the base vehicle model 13, is a dynamic two-wheeled model that takes acceleration / deceleration and front wheel steering angle as inputs and outputs vehicle speed and yaw rate. It is constructed using a vehicle model with the same specifications as the vehicle from which the training data was acquired (i.e., the data collection vehicle). The acceleration / deceleration and front wheel steering angle input to the data collection vehicle model 101 are calculated using the control data OD1.

[0066] The compensator 102 for the base vehicle has the same configuration as the compensator 14 and is a target value following control system that determines the acceleration / deceleration and front wheel steering angle so that the base vehicle follows the target values, using the vehicle speed V and yaw rate γ, which are the outputs of the collection vehicle model 101.

[0067] In other words, the compensator 102 for the base vehicle uses the nonlinear system shown in equations (6) and (7). The longitudinal motion of the vehicle is described by a point mass model, and the lateral motion and rotational motion around the center of gravity are described by a dynamic two-wheel model. x(t), u(t), y(t) and G(x), H(x), C in equations (6) and (7) are defined as shown in equations (8), (9), (10), (11), (12), and (13), respectively. Here, the vehicle parameters for each model are those of the base vehicle.

[0068] The base vehicle compensator 102 outputs multiple control data OD2 based on the acceleration / deceleration and front wheel steering angle input from the collection vehicle model 101. The learning data generation device 100 configured in this way includes a collection vehicle model 101 and a compensator 102 for the base vehicle.

[0069] The data collection vehicle model 101 is configured to receive multiple control data OD1 for operating a data collection vehicle to perform data collection, convert them into vehicle speed and yaw rate independent of the individual vehicle using a dynamic two-wheel model based on the mechanical characteristics of the data collection vehicle, and output the converted vehicle speed and yaw rate.

[0070] The compensator 102 for the base vehicle is configured to generate training data by converting the vehicle speed and yaw rate output by the collection vehicle model 101 into multiple control data OD1 and multiple control data OD2 which are operation data of the same type, according to the mechanical characteristics of the base vehicle.

[0071] Such a learning data generation device 100 can generate learning data for a data-driven planner 11, which is built by machine learning operations on a base vehicle, using multiple data collection vehicles of different vehicle types. This makes it easier to build a data-driven planner 11 used to control a vehicle.

[0072] In the embodiments described above, the collection vehicle model 101 corresponds to the collection conversion unit, the base vehicle compensator 102 corresponds to the generation unit, the data collection vehicle corresponds to the data collection target machinery and equipment, the control data OD1 corresponds to the collection operation parameters, and the control data OD2 corresponds to the reference operation parameters.

[0073] Although one embodiment of the present disclosure has been described above, the present disclosure is not limited to the above embodiment and can be implemented in various modified forms. [Example 1] In the above embodiment, the controlled mechanical device was shown to be a vehicle, but it is not limited to a vehicle; for example, it could be a robot, aircraft, satellite, ship, etc.

[0074] [Differentiation 2] In the above embodiment, the base vehicle model 13 and compensator 14 are shown mounted on the control vehicle. However, the base vehicle model 13 and compensator 14 may also be mounted on a device installed outside the control vehicle (for example, a server capable of data communication with the vehicle).

[0075] The vehicle control device 2 and its method described in this disclosure may be implemented by a dedicated computer provided by configuring a processor and memory programmed to perform one or more functions embodied by a computer program. Alternatively, the vehicle control device 2 and its method described in this disclosure may be implemented by a dedicated computer provided by configuring a processor by one or more dedicated hardware logic circuits. Alternatively, the vehicle control device 2 and its method described in this disclosure may be implemented by one or more dedicated computers configured by a combination of a processor and memory programmed to perform one or more functions and a processor configured by one or more hardware logic circuits. Furthermore, the computer program may be stored as instructions executed by the computer on a computer-readable non-transitional tangible recording medium. The method for realizing the functions of each part included in the vehicle control device 2 does not necessarily need to include software, and all of its functions may be realized using one or more hardware components.

[0076] Multiple functions of one component in the above embodiment may be realized by multiple components, or one function of one component may be realized by multiple components. Furthermore, multiple functions of multiple components may be realized by one component, or one function realized by multiple components may be realized by one component. Also, some parts of the configuration of the above embodiment may be omitted. Furthermore, at least some parts of the configuration of the above embodiment may be added to or replaced with the configuration of other above embodiments.

[0077] In addition to the vehicle control device 2 described above, this disclosure can also be realized in various forms, such as a system that uses the vehicle control device 2 as a component, a program for causing the computer to function as the vehicle control device 2, a non-transitional physical recording medium such as a semiconductor memory on which this program is recorded, and a control method.

[0078] In addition to the learning data generation device 100 described above, this disclosure can also be realized in various forms, such as a system that uses the learning data generation device 100 as a component, a program for causing a computer to function as the learning data generation device 100, a non-transitional physical recording medium such as a semiconductor memory on which this program is recorded, and a learning data generation method. [Explanation of Symbols]

[0079] 2...Vehicle control device, 11...Data-driven planner, 13...Base vehicle model, 14...Compensator, 100...Learning data generation device, 101...Collection vehicle model, 102...Compensator

Claims

1. A reference conversion unit (13) is configured to take at least one reference operation parameter, which is an operation parameter output by a reference machine learning model (11), which is a machine learning model constructed by machine learning operations on a reference machine learning model of a reference machine learning model, as input, convert it into at least one operation parameter that is independent of individual machine learning models, using a machine learning model, which is a physical model based on the mechanical characteristics of the reference machine learning model, and output the converted at least one operation parameter, A reconversion unit (14) is configured to reconvert at least one of the operation parameters output by the reference conversion unit to at least one control operation parameter which is the same type of operation parameter as at least one of the reference operation parameters, according to the mechanical characteristics of the controlled machine device which is the machine device to be controlled. Equipped with, A data device (2) that outputs the control operation parameters, which have been reconverted by the reconversion unit, to the controlled machine device.

2. A data device according to claim 1, The reference conversion unit is a data device that takes at least one of the reference operation parameters as input and at least one of the operation parameters as output, and is constructed using a machine device model having the same specifications as the reference machine device.

3. A data device according to claim 1 or claim 2, The re-conversion unit is a data device that performs target value tracking control, where the operation indicated by the operation parameter is the target value, and the control operation parameter is determined so that the controlled machine or device performs an operation that follows the target value.

4. A training data generation device (100) that generates training data for machine learning, A collection conversion unit (101) is configured to take at least one collection operation parameter, which is an operation parameter for operating a collection target machine, which is a machine for performing data collection, as input, convert it into at least one operation parameter that is independent of individual machine, using a machine model, which is a physical model based on the mechanical characteristics of the collection target machine, and output the converted at least one operation parameter, The generation unit (102) is configured to generate the learning data by converting at least one of the operation parameters output by the collection conversion unit back into at least one reference operation parameter which is the same type of operation parameter as at least one of the collection operation parameters, according to the mechanical characteristics of a reference machine which is a reference machine. A learning data generation device equipped with the following features.

5. Computers, A reference conversion unit (13) is configured to take at least one reference operation parameter, which is an operation parameter output by a reference machine learning model (11), which is a machine learning model constructed by machine learning operations on a reference machine learning model of a reference machine learning model, as input, convert it into at least one operation parameter that is independent of individual machine learning models, using a machine learning model, which is a physical model based on the mechanical characteristics of the reference machine learning model, and output the converted at least one operation parameter, and A reconversion unit (14) is configured to reconvert at least one of the operation parameters output by the reference conversion unit to at least one control operation parameter which is the same type of operation parameter as the at least one reference operation parameter, according to the mechanical characteristics of the controlled machine device which is the machine device to be controlled. To make it function as, A data conversion program for outputting the control operation parameters, which have been reconverted by the reconversion unit, to the controlled machine or device.

6. The computer of the training data generation device (100) that generates training data for machine learning, A collection conversion unit (101) is configured to take at least one collection operation parameter, which is an operation parameter for operating a collection target machine, which is a machine for performing data collection, as input, convert it into at least one operation parameter that is independent of individual machine, using a machine model, which is a physical model based on the mechanical characteristics of the collection target machine, and output the converted at least one operation parameter, and The generation unit (102) is configured to generate the learning data by converting at least one of the operation parameters output by the collection and conversion unit back into at least one reference operation parameter, which is the same type of operation parameter as the at least one of the collection operation parameters, according to the mechanical characteristics of a reference machine, which is a reference machine. A training data generation program to enable it to function as such.

7. A reference conversion unit (13) is configured to take at least one reference operation parameter, which is an operation parameter output by a reference machine learning model (11), which is a machine learning model constructed by machine learning operations on a reference machine learning model of a reference machine learning model, as input, convert it into at least one operation parameter that is independent of individual machine learning models, using a machine learning model, which is a physical model based on the mechanical characteristics of the reference machine learning model, and output the converted at least one operation parameter, A reconversion unit (14) is configured to reconvert at least one of the operation parameters output by the reference conversion unit to at least one control operation parameter which is the same type of operation parameter as at least one of the reference operation parameters, according to the mechanical characteristics of the controlled machine device which is the machine device to be controlled. Equipped with, A control system (2) that controls the controlled mechanical device based on the control operation parameters reconverted by the reconversion unit.

Images