Steering characteristic analysis device using machine learning

Abstract

To appropriately analyze steering characteristics of vehicles.SOLUTION: A characteristic analysis device 10 includes: an AI model acquisition unit that acquires an AI model that has learned, as a data set, a test condition of a steering test, driver's measurement data including driver's myoelectric potential data, obtained in a specific section of the steering test, and a result of the driver's sensory evaluation; an input unit that acquires a test condition of a steering test to be evaluated and measurement data including the myoelectric potential data of the driver who performs the steering test of the vehicle to be evaluated; and a characteristic information generation unit that inputs the test condition of the steering test of the vehicle to be evaluated, driver's response data and electroencephalogram data into the AI model and generates sensory evaluation of the steering test of the vehicle to be evaluated.SELECTED DRAWING: Figure 1

Description

[Technical Field]

[0001] The present invention relates to a steering characteristic analysis device using machine learning. [Background technology]

[0002] One of the evaluation targets for vehicle performance is steering feeling, which evaluates the behavior of the vehicle in response to operations performed by the driver. Patent Document 1 describes a steering feeling measuring device that includes a first biological information detection means for detecting first biological information generated by steering as intended by the driver, a second biological information detection means for detecting second biological information other than the first biological information generated by steering as intended by the driver, and a collection means for collecting the second biological information detected by the second biological information detection means together with the first biological information detected by the first biological information detection means, in association with the first biological information detected by the first biological information detection means. [Prior art documents] [Non-patent literature]

[0003] [Patent Document 1] Japanese Patent Application Laid-Open No. 2003-177079 Summary of the Invention [Problem to be solved by the invention]

[0004] The steering feel is evaluated based on a subjective sensory evaluation of the driver's driving experience and various acquired information. Therefore, a sensory evaluation must be obtained for each vehicle under test. Furthermore, because the sensory evaluation is a subjective evaluation by the driver, there is a risk that the evaluation criteria may fluctuate, resulting in fluctuating results.

[0005] The present invention has been made in consideration of the above, and aims to provide a steering characteristic analysis device using machine learning that can appropriately analyze steering characteristics and evaluate vehicle performance. [Means for solving the problem]

[0006] In order to solve the above-mentioned problems and achieve the objectives, the steering characteristic analysis device of the present disclosure includes an AI model acquisition unit that acquires an AI model that has been trained using as a data set the test conditions of the steering test, the driver's measurement data including the driver's electromyography data obtained in a specific section of the steering test, and the results of the driver's sensory evaluation, an input unit that acquires the test conditions of the steering test of the vehicle to be evaluated and measurement data including the electromyography data of the driver who performs the steering test of the vehicle to be evaluated, and a characteristic information generation unit that inputs the test conditions of the steering test of the vehicle to be evaluated and the driver's response data and electroencephalogram data into the AI ​​model to generate a sensory evaluation of the steering test of the vehicle to be evaluated. [Effects of the Invention]

[0007] According to the present invention, the steering characteristics of a vehicle can be appropriately analyzed. [Brief explanation of the drawings]

[0008] [Figure 1] FIG. 1 is a schematic block diagram of a characteristic analysis device according to this embodiment. [Figure 2] FIG. 2 is a schematic diagram showing an example of a travel route for a steering test. [Figure 3] FIG. 3 is a schematic diagram showing an example of an electroencephalogram measuring device. [Figure 4] Figure 4 is a conceptual diagram of an AI model. [Figure 5] FIG. 5 is a schematic diagram showing an example of a data set. [Figure 6] FIG. 6 is a flowchart illustrating the processing flow of the characteristic analysis device. [Figure 7] FIG. 7 is a flowchart illustrating the processing flow of the characteristic analysis device. [Figure 8] FIG. 8 is a flowchart illustrating the processing flow of the characteristic analysis device. [Figure 9] FIG. 9 is a schematic diagram showing an example of measurement data. DETAILED DESCRIPTION OF THE INVENTION

[0009] Preferred embodiments of the present invention will be described in detail below with reference to the accompanying drawings. However, the present invention is not limited to the embodiments described below.

[0010] The characteristic analysis device (steering characteristic analysis device) according to this embodiment uses machine learning to analyze the characteristics of a vehicle's steering performance. A vehicle is a moving mechanism that is driven by a driver, such as a passenger car, truck, bus, or motorcycle. A vehicle is equipped with a mechanism that allows the driver to input a steering operation, such as a steering wheel, to change the direction of travel. The characteristic analysis device analyzes how the driver feels about the vehicle behavior that occurs in response to the steering operation input by the driver.

[0011] 1 is a schematic block diagram of a characteristic analysis device according to this embodiment. A test system 1 equipped with the characteristic analysis device 10 includes the characteristic analysis device 10, a test condition data acquisition unit 12, a measurement data acquisition unit 14, and a sensory evaluation data acquisition unit 16. The test system 1 acquires information on a vehicle steering test using the test condition data acquisition unit 12, the measurement data acquisition unit 14, and the sensory evaluation data acquisition unit 16. The characteristic analysis device 10 is connected to the test condition data acquisition unit 12, the measurement data acquisition unit 14, and the sensory evaluation data acquisition unit 16, and acquires data from each unit.

[0012] 2 is a schematic diagram showing an example of a driving route for a steering test. In the test system 1, a driver drives a vehicle to be evaluated on a test course 100. The test course 100 has a route that requires steering operations, and the driver inputs steering operations while driving on the test course 100 and evaluates the sensations felt while driving. A section 102 of the test course 100 is the subject of evaluation.

[0013] The test condition data acquisition unit 12 acquires data on the test conditions for the steering test. The test condition data acquisition unit 12 acquires information on the test conditions through communication or operator input. The test condition data acquisition unit 12 may be integrated with the identification analysis device 10. The test conditions include vehicle information, test course 100 information, driver information, etc. Vehicle information includes the vehicle model, the type of steering device installed, etc. Test course 100 information includes the location, shape, weather conditions, etc. of the test course. Driver information includes the driver's identification information (ID), etc.

[0014] The measurement data acquisition unit 14 acquires, as measurement data, driver data and vehicle data during the steering test. That is, the measurement data acquisition unit 14 acquires various information about the vehicle and the driver while driving on the test course 100. The vehicle data while driving includes steering information input to the vehicle, accelerator and brake information, and vehicle acceleration and speed information in each direction while driving. The driver information is information from various sensors worn by the driver. The driver's measurement data during the steering test is data detecting various reactions obtained from the driver performing the steering test, including electromyogram data and electroencephalogram data. The measurement data acquisition unit 14 acquires, as the driver's measurement data, information detected by an electromyogram sensor that detects movements of the arms, shoulders, neck, etc., and an electroencephalogram sensor that detects electroencephalograms. The measurement data acquisition unit 14 acquires movements of the driver's arms, shoulders, neck, etc. as electromyogram data using the electromyogram sensor. The measurement data acquisition unit 14 acquires reactions of various parts of the driver's brain as electroencephalogram data using the electroencephalogram sensor.

[0015] FIG. 3 is a schematic diagram showing an example of an electroencephalogram (EEG) measuring device. The measurement data acquiring unit 14 acquires data from the measuring device shown in FIG. 3. The measuring device shown in FIG. 3 includes a plurality of detection terminals 122 attached to a driver's head 120 to detect EEGs output from various parts of the driver. The detection terminals 122 detect responses at various locations in the driver's brain as EEGs. The measuring device records the detected EEGs in association with the detection terminals, and outputs the data to the measurement data acquiring unit 14. Note that the method for acquiring EEG data, which is measurement data of EEGs, is not limited to this, and various methods capable of detecting the driver's brain responses (EEG data) can be used. Furthermore, various methods capable of detecting the driver's physical responses can be used to acquire measurement data of the driver's physical responses, such as EMG data.

[0016] The sensory evaluation data acquisition unit 16 acquires the evaluation results of the vehicle driven by the driver in the steering test. The sensory evaluation data acquisition unit 16 is a terminal into which the driver or an operator who acquires the result information from the driver inputs the evaluation results. The sensory evaluation data acquisition unit 16 may be integrated with the characteristic analysis device 10. The evaluation results are, for example, a result of rating the information felt about the vehicle on a 10-point scale. Examples of evaluation items include gain, connection, and phase delay as vehicle behavior, and neutral clarity, built-up feeling, friction feeling, viscosity feeling, and springiness as steering feeling. In addition, the sensory evaluation may include evaluations such as likes / dislikes, comfort / anxiety, light / heavy, and whether or not the steering is possible as intended. The characteristic analysis device 10 can create a data set for each sensory evaluation by acquiring evaluation results for the same sensory evaluation items for multiple drivers.

[0017] The characteristic analysis device 10 creates a trained AI model that predicts the results of the sensory evaluation based on the results of the steering test, and estimates the sensory evaluation (characteristics) based on the trained AI model and the results of the steering test of the evaluation target.

[0018] The characteristic analysis device 10 can be said to be a computer, and includes an input unit 70, an output unit 72, a communication unit 74, a storage unit 76, and a control unit 78. The input unit 70 is a user interface that accepts user operations (inputs), and may be, for example, a mouse or a keyboard. The output unit 72 is a device that outputs information, and may be, for example, a display. The communication unit 74 is a communication module that communicates with an external device, and may be, for example, an antenna. The characteristic analysis device 10 communicates with an external device via wireless communication, but wired communication may also be used, and any communication method may be used.

[0019] The storage unit 76 is a memory that stores various information such as the calculation contents and programs of the control unit 78, and includes, for example, at least one of a main storage device such as a random access memory (RAM), a read only memory (ROM), and an external storage device such as a hard disk drive (HDD). The program for the control unit 78 stored in the storage unit 76 may be stored in a recording medium readable by the characteristic analysis device 10. The storage unit 76 also stores an AI model N, which is a learning model in AI (artificial interference). The storage unit 76 also stores a dataset 77 used for learning the AI ​​model N.

[0020] FIG. 4 is a conceptual schematic diagram of the AI ​​model N. The AI ​​model N can be said to be a model in which weighting coefficients and bias values ​​have been learned. The AI ​​model N is a learning model trained by deep learning, and is composed of a model (neural network) that reconstructs a sprung spectrogram from an unsprung spectrogram, trained by deep learning, and variables. Deep learning is a type of machine learning technique, and in a narrow sense, is composed of a neural network with, for example, four or more layers. In this embodiment, the AI ​​model N is a CNN (Conventional Neural Network) model, and as shown in FIG. 4, includes, for example, an encoder layer MN1 and a decoder layer MN2 that include multiple convolution layers and multiple pooling layers. When input data IN is input, the AI ​​model N performs calculations in the encoder layer MN1 and the decoder layer MN2 and outputs output data ON.

[0021] The AI ​​model shown in Figure 4 is an example, and models other than neural network models may be used. The AI ​​model is a supervised learning model, and various models and analysis methods capable of regression analysis can be used. For example, a linear multiple regression model, a decision tree model, a random forest model, or a support vector machine model can be used.

[0022] FIG. 5 is a schematic diagram showing an example of a data set. The data set 220 shown in FIG. 5 is created by the control unit 78 and stored in the storage unit 76. The data set 220 stores, for each evaluation test, a data ID (data No.), test subject (driver) information 212, ESP specification information 214, sensory evaluation score information 216, and measurement data feature data 218 in association with each other. The ESP specification information 214 is vehicle information that indicates steering performance. The sensory evaluation score information 216 is data resulting from an evaluation of the vehicle's driving results by the test subject. The measurement data feature data 218 is data resulting from processing data acquired from the test subject and the vehicle during the steering test.

[0023] The control unit 78 is a computing device and includes a computing circuit such as a CPU (Central Processing Unit). The control unit 78 includes a data set creation unit 79, a learning unit 80, an AI model acquisition unit 82, a measurement data acquisition unit 84, a data processing unit 86, and a characteristic data processing unit 88. The control unit 78 reads and executes a program (software) from the storage unit 76, thereby realizing the data set creation unit 79, the learning unit 80, the AI ​​model acquisition unit 82, the measurement value acquisition unit 84, the data processing unit 86, and the characteristic information generation unit 88 and performing their processing. The control unit 78 may perform these processes using a single CPU, or may be provided with multiple CPUs and perform the processes using the multiple CPUs. At least a portion of the processing performed by the data set creation unit 79, the learning unit 80, the AI ​​model acquisition unit 82, the measurement data acquisition unit 84, the measurement value processing unit 86, and the characteristic information generation unit 88 may be realized by a hardware circuit.

[0024] (Dataset Creation Department) The dataset creation unit 79 extracts and associates data of the same steering test from the data acquired by the test condition data acquisition unit 12, the measurement data acquisition unit 14, and the sensory evaluation data acquisition unit 16, extracts necessary information from each piece of data, and creates data in which the test conditions, measurement data, and sensory evaluation data of one piece of driving data are associated with each other. The dataset creation unit 79 executes the above process for each test and creates the dataset shown in FIG. 5. The dataset creation unit 79 also identifies the section to be evaluated from the measurement data and extracts necessary data. The extracted data may then be processed to create feature data. It is preferable to set standards for data processing in advance.

[0025] (Study Department) The learning unit 80 performs machine learning on the unlearned AI model to generate a trained AI model. The learning unit 80 reads the data in the dataset 77, and performs machine learning on the AI ​​model N using the test conditions and measurement data as input and the sensory evaluation as output. Here, the sensory evaluation to be output may be set by an operator as the evaluation target. The learning unit 80 trains an unlearned AI model to generate a trained AI model.

[0026] Next, the AI ​​model acquisition unit 82, the measurement data acquisition unit 84, the data processing unit 86, and the characteristic information generation unit 88 execute a process to output a sensory evaluation using the trained AI model generated by the learning unit 80 and the test conditions and measurement data of the test to be evaluated.

[0027] (AI Model Acquisition Department) The AI ​​model acquisition unit 82 acquires the trained AI model generated by the training unit 80. The AI ​​model acquisition unit 82 outputs the acquired trained AI model to the characteristic information generation unit 88.

[0028] (Measurement data acquisition section) The measurement data acquisition unit 84 acquires data on the test to be evaluated from the test condition data acquisition unit 12 and the measurement data acquisition unit 14 .

[0029] (Data Processing Department) The data processing unit 86 processes the data acquired by the measurement data acquisition unit 84 into data that can be input to an AI model. The data processing unit 86 performs the same processing as that performed on the test condition data and measurement data when the data set creation unit 79 creates data. Examples include processing to extract some data from the measurement data and converting the measurement data into feature quantities. The data processing unit 86 also performs similar processing when the learning unit 80 processes the input data.

[0030] (Characteristics information generation unit) The characteristic information generation unit 88 inputs the data processed by the data processing unit 86 into the trained AI model to generate an estimated value of the sensory evaluation. The characteristic information generation unit 88 may output the generated characteristic information to the output unit 72, or may output (transmit) it to another device.

[0031] (Processing flow) The processing flow of the characteristic analysis device 10 explained above will be described. Fig. 6 is a flowchart illustrating the processing flow of the characteristic analysis device. A method for estimating sensory evaluation using a trained AI model will be described with reference to Fig. 6.

[0032] As shown in FIG. 6, the characteristic analysis device 10 acquires a trained AI model using the AI ​​model acquisition unit 82 (step S12). Next, test conditions are determined (step S14). Next, a test is performed based on the determined test conditions, and measurement data is acquired (step S16). The test is performed by a driver driving on a test course. While the driver is driving, the measurement data acquisition unit 14 acquires measurement data of the driver, including information on the driver's operations, electromyography data, and electroencephalogram data. The measurement data acquisition unit 14 can also acquire data from sensors attached to the vehicle. The measurement data acquisition unit 14 supplies the acquired measurement data to the characteristic analysis device 10. The characteristic analysis device 10 processes the measurement data acquired from the measurement data acquisition unit 14 into a state that can be processed by the trained AI model. Specifically, the characteristic analysis device 10 processes the measurement data into the same format as the input data used when creating the trained AI model.

[0033] Next, the characteristic analysis device 10 analyzes the measurement data using the trained AI model and outputs the analysis result (step S18). That is, the characteristic analysis device 10 inputs the test conditions and the processed measurement data into the trained AI model and outputs a sensory evaluation as the analysis result.

[0034] Next, a method for creating a trained AI model will be described. FIGS. 7 and 8 are flowcharts illustrating the processing flow of the characteristic analysis device. First, a method for creating a dataset will be described using FIG. 7. Note that for the test that forms the basis of the dataset, an operator processes the test conditions, etc., and a driver drives and performs a sensory evaluation. The operator determines the test conditions (step S22). The test conditions may be set by inputting configurable conditions, and the characteristic analysis device 10 may determine them randomly. Next, a test is performed based on the test conditions determined by the test system 1 (step S24). The driver determined by the test conditions drives a vehicle with the equipment configuration of the test conditions, and drives the vehicle on a test course under the test conditions. The test system 1 acquires measurement data using the measurement data acquisition unit 14 (step S26). The measurement data acquisition unit 14 acquires data obtained by sensors attached to the vehicle and the driver of the vehicle traveling on the test course.

[0035] Next, the test system 1 acquires information on the sensory evaluation of the driver using the sensory evaluation acquisition unit 16 (step S28).

[0036] Next, the data processing unit 86 of the characteristic analysis device 10 identifies an extraction region of the measurement data (step S32). The data processing unit 86 identifies a driving section to be evaluated from the measurement data acquired in the steering test. The data processing unit 86 extracts data corresponding to the driving section. Furthermore, the data processing unit 86 may process the data of the extraction region based on settings, for example, extracting feature amounts or selecting parameters to be used.

[0037] The characteristic analysis device 10 associates the measurement data of the extracted region with the test conditions and the sensory evaluation, and stores them in a data set (step S34). The test system 1 repeats the process of Fig. 7 for each steering test, and creates a data set in which the test conditions, measurement data, and sensory evaluation are associated with each other for multiple tests under different conditions.

[0038] Next, an example of a learning process using a dataset will be described with reference to FIG. 8. The characteristic analysis device 10 learns the dataset created in the process of FIG. 7 to generate a trained AI model. The characteristic analysis device 10 acquires the dataset (step S42). Next, the characteristic analysis device 10 selects an AI model to use (step S44). Next, the characteristic analysis device 10 executes learning in the learning unit 80 (step S46). That is, the learning unit 80 performs machine learning on the AI ​​model using the test conditions of the dataset and processed measurement data as input and the results of the sensory evaluation as output, to create a trained AI model.

[0039] The characteristic analysis device 10 creates multiple trained AI models while changing the AI ​​model and data set, and selects the trained AI model to be used in FIG. 6 from the created trained AI models, thereby creating a trained AI model that can output a sensory evaluation with high accuracy. The characteristic analysis device 10 may also separate data sets for each test condition, for example, whether the driver is an expert or a person with general skills, and create a trained AI model for each. A trained AI model may also be created for each evaluation item of the sensory evaluation.

[0040] FIG. 9 is a schematic diagram illustrating an example of measurement data. FIGS. 10 to 12 are each a schematic diagram illustrating an example of analysis results. As shown in FIG. 9, the characteristic analysis device 10 acquires information such as a steering angle 312, an electroencephalogram 314, and an electromyogram 316 as measurement data. Note that FIG. 9 is an example of data, and information on the electroencephalogram 314 and the electromyogram 316 is acquired from multiple locations. Data other than the electroencephalogram and the electromyogram is also acquired. The characteristic analysis device 10 extracts data of specific regions 322, 324, and 326 from the steering angle 312, the electroencephalogram 314, and the electromyogram 316. The specific regions 322, 324, and 326 are data from the same time period, that is, data from the time period when the driver is operating the steering device during the steering test. The characteristic analysis device 10 analyzes the extracted data, calculates feature amounts, and creates a dataset.

[0041] The characteristic analysis device 10 performs machine learning on the AI ​​model using the data set, thereby evaluating the influence of each parameter of the measurement data on the sensory evaluation, and creating a trained AI model in which the influence of the parameter of the measurement data on the sensory evaluation has been adjusted.

[0042] (effect) As described above, the characteristic analysis device 10 can calculate a sensory evaluation by inputting measurement data including electromyography data and test condition data into a trained AI model. The characteristic analysis device 10 can estimate a sensory evaluation from measurement data, thereby reducing the effort required for the driver to perform the sensory evaluation. Furthermore, by performing the sensory evaluation based on information acquired from a dataset, it is possible to prevent the results of the sensory evaluation from changing due to the driver's condition during the test. In other words, it is possible to prevent the results of the sensory evaluation from changing due to the driver's physical condition or mood on that day, and to perform a sensory evaluation based on the physical quantity of electromyography data. Furthermore, even when testing different vehicles and steering systems, it is possible to estimate a sensory evaluation based on measurement data.

[0043] Furthermore, by including electroencephalogram data in the measurement data of the driver, the characteristic analysis device 10 can obtain information felt by the driver during the steering test as electroencephalogram data. This makes it possible to output a sensory evaluation based on the electroencephalogram data, which is information felt by the driver, in addition to the measurement data of the driver's myoelectric potential, and thus makes it possible to further increase the accuracy of the estimation of the sensory evaluation.

[0044] In addition, the characteristic analysis device 10 creates a trained AI model using machine learning, which allows it to efficiently perform learning based on a large number of parameters from the detection results (electroencephalogram data, electromyogram data) of multiple sensors attached to the driver and vehicle, and to estimate sensory evaluations from the measurement data with high accuracy.

[0045] Furthermore, the characteristic analysis device 10 processes the measurement data and calculates the feature amount of the extracted region, thereby making it possible to estimate the sensory evaluation of the steering test with high accuracy.

[0046] Furthermore, it is preferable to use a support vector machine model for learning, which allows the degree of influence of the parameters of the measurement data to be evaluated efficiently and improves the estimation accuracy of the sensory evaluation.

[0047] Although the embodiments and examples of the present invention have been described above, the embodiments are not limited to the contents of these embodiments. Furthermore, the above-described components include those that can be easily imagined by a person skilled in the art, those that are substantially the same, and those that are within the so-called equivalent range. Furthermore, the above-described components can be combined as appropriate. Furthermore, various omissions, substitutions, or modifications of the components can be made without departing from the spirit of the above-described embodiments. [Explanation of symbols]

[0048] 1 Test System 10 Characteristic analyzer 12 Test condition data acquisition section 14 Measurement data acquisition unit 16 Sensory Evaluation Acquisition Department 70 Input section 72 Output section 74 Communications Department 76 Memory section 77 datasets 78 Control Unit 79 Dataset Creation Department 80 Learning Department 82 AI Model Acquisition Department 84 Measurement data acquisition unit 86 Data Processing Department 88 Characteristic information generation section N AI model

Claims

1. A characteristic analysis device that evaluates a vehicle behavior in response to a steering operation input by a driver of a vehicle, an AI model acquisition unit that acquires an AI model that has been trained using, as a data set, test conditions for the steering test, measurement data of the driver including electromyography data of the driver acquired in a specific section of the steering test, and the results of a sensory evaluation of the driver; an input unit that acquires test conditions for a steering test of a vehicle to be evaluated and measurement data including myoelectric potential data of the driver who performs the steering test of the vehicle to be evaluated; and a characteristic information generation unit that inputs test conditions for a steering test of the vehicle to be evaluated and measurement data of the driver into the AI ​​model to generate a sensory evaluation of the steering test of the vehicle to be evaluated.

2. The steering characteristic analysis device using machine learning according to claim 1 , wherein the measurement data includes electroencephalogram data of the driver.

3. a data set creation unit that acquires test conditions for a steering test, measurement data of the driver, and results of a sensory evaluation of the driver, sets specific sections for the steering test, and creates data for each steering test; 3. The steering characteristic analysis device using machine learning according to claim 1 or claim 2, further comprising: a learning unit that determines an AI model to be used, trains the determined AI model using the dataset created by the dataset creation unit, and creates a trained AI model.

4. A steering characteristic analysis device using machine learning as described in claim 1 or claim 2, wherein the AI ​​model is a support vector machine model.

Images