Machine learning device, vehicle testing system, machine learning method, and vehicle testing method

Abstract

The invention provides a machine learning device, a vehicle test system, a machine learning method, and a vehicle test method, which make it easy to reproduce loads such as road slope resistance during actual road travel, and comprises: a learning data acquisition unit (51) which acquires learning data comprising load data and vehicle travel data, and outputs the acquired learning data; the load data comprises road slope resistance of a driving path and a driving resistance variation amount caused by vehicle behaviors, and the vehicle driving data comprises a vehicle speed, an acceleration operation amount and a brake operation amount when the vehicle is driven on the driving path; and a machine learning unit (52) that performs machine learning on the correlation between the load data and the vehicle travel data, and generates a simulated gradient prediction model indicating the correlation between the load data and the vehicle travel data.

Description

Technical Field

[0001] The present invention relates to a machine learning device, a vehicle test system, a machine learning method, and a vehicle test method for simplifying the reproduction of loads such as road gradient loads during actual road driving. Background Art

[0002] Conventionally, based on the vehicle driving data obtained during actual road driving, a road load simulation (Road Load Simulation) is performed using a dynamometer installed in a test laboratory to simulate the vehicle driving. In this road load simulation, rolling resistance (A+B×V), air resistance (C×V 2 ), inertial resistance (I×α) or road gradient resistance (M×g×sinθ), etc., to set the target driving resistance applied by the dynamometer to the vehicle. In addition, in actual road driving, in addition to the above-mentioned driving resistance, inertial resistance, and road gradient resistance, driving resistance caused by vehicle behavior (acceleration and deceleration, etc.) is also generated.

[0003] Specifically, the target running resistance (RL[N]) is expressed by the following equation (1). RL=A+B×V+C×V 2 +I×α+M×g×sinθ Road +F Variable Here, V: vehicle speed [m / s] A, B, C: driving resistance constants θ Road : Road slope (road slope in the direction of travel) [deg] M: test vehicle weight I: Equivalent inertial mass [kg] α: acceleration [m / s 2 ] g: acceleration due to gravity F Variable : The amount of change in driving resistance caused by vehicle behavior (acceleration, deceleration, steering, etc.) The values ​​of the items A, B, and C vary depending on the type of vehicle, road surface conditions, tire type, etc., and are different for each vehicle. The item C is the air resistance coefficient, and the items A and B are coefficients representing rolling resistance.

[0004] Here, in the road load simulation, it is important to reproduce not only the rolling resistance (A+B×V) and air resistance (C×V 2 ), inertial resistance (I×α), and reproduces the road slope resistance (M×g×sinθ Road ) and the driving resistance variation caused by vehicle behavior (F VariableHowever, the driving resistance variation (F Variable ) and only measure the road slope (θ Road ) (and the measurement accuracy of the road slope is poor), and the slope command value (torque command value) based on the road slope is input into the dynamometer.

[0005] On the other hand, in order to reproduce the road slope resistance (M×g×sinθ Road ) and the driving resistance variation caused by vehicle behavior (F Variable ), as shown in Patent Document 1, a method of obtaining a gradient curve based on the torque matching result is considered. In this method, the gradient curve of the road slope (θ Road ) plus the driving resistance change (F Variable ) is converted into a value (hereinafter referred to as the simulated slope (θ Emulation )). As a result, the simulated slope (θ Emulation ) is input into the dynamometer. In addition, as shown in Patent Document 2, a method of combining a gyro sensor, an acceleration sensor, GPS information, etc. to calculate the road slope with high accuracy is disclosed. In addition, there is a method of calculating based on values ​​measured during driving such as a gyro sensor, a wheel torque sensor, and external pressure measurement, or a method of calculating based on elevation information of a commercially available topographic map or a high-precision three-dimensional map.

[0006] However, the method shown in Patent Document 1 requires torque matching every time the vehicle is actually driven on the road, which is time-consuming and laborious. In addition, the method shown in Patent Document 2 requires time to install various sensors such as gyro sensors and wheel torque sensors, which is time-consuming and laborious. Moreover, in the external pressure measurement, it is difficult to detect slight changes in road slope, and thus the vehicle is affected by pressure changes other than elevation, such as when passing through a tunnel. In addition, the simulated slope (θ Emulation ). In addition, in the method using commercially available topographic maps, it is difficult to obtain data covering all roads with high granularity at a low cost, and it is also impossible to obtain simulated slopes (θ Emulation ). Prior art literature

[0007] Patent Document 1: Japanese Patent Application No. 2021-508799 Patent Document 2: Japanese Patent No. 5273599 Summary of the invention

[0008] Therefore, in view of the above-mentioned problems, a main object of the present invention is to simplify the reproduction of loads such as the road gradient load during actual road travel and the amount of travel resistance variation caused by vehicle behavior.

[0009] That is, the machine learning device of the present invention generates a simulated slope prediction model for vehicle testing, and the machine learning device is characterized in that it comprises: a learning data acquisition unit, which acquires learning data composed of load data and vehicle driving data, the load data including the road slope resistance of the driving path and the driving resistance change caused by the vehicle behavior, and the vehicle driving data including the vehicle speed, acceleration operation amount and braking operation amount when the vehicle is driving on the driving path; and a machine learning unit, which performs machine learning on the correlation between the load data and the vehicle driving data to generate a simulated slope prediction model representing the correlation between the load data and the vehicle driving data.

[0010] According to the machine learning device, it is possible to generate a simulated gradient prediction model indicating the correlation between load data and vehicle travel data, wherein the load data includes the road gradient resistance (M×g×sinθ Road ) and the driving resistance variation caused by vehicle behavior (F Variable ), the vehicle driving data includes the vehicle speed, acceleration operation amount and brake operation amount when the vehicle is driving on the driving path. Therefore, by inputting various vehicle driving data into the simulated slope prediction model, the road slope resistance (M×g×sinθ) corresponding to each vehicle driving data can be obtained. Road ) and the driving resistance variation caused by vehicle behavior (F Variable As a result, the road gradient resistance (M×g×sinθ) of the actual road can be Road ) and the driving resistance variation caused by vehicle behavior (F Variable ) etc. makes load reproduction easy.

[0011] Here, it is preferable that the load data include a load calculated based on a simulated gradient obtained by adding a gradient-converted value of a running resistance variation amount caused by vehicle behavior to a road gradient of the running route.

[0012] As a specific embodiment of the present invention, it can be considered that the learning data acquisition unit acquires learning data consisting of load data used in torque matching or road load simulation and vehicle driving data used at this time, and the machine learning unit performs machine learning on the correlation between the acquired load data and the vehicle driving data to generate a simulated slope prediction model representing the correlation between the load data and the vehicle driving data.

[0013] Different vehicles have different pedal characteristics. Even in the same vehicle, if the pedal setting (pedal setting for the eco mode, pedal setting for the sport mode, etc.) is changed, the pedal characteristics for each mode become different. In this case, in order to apply the generated simulated slope prediction model to vehicles with different pedal characteristics, it is preferred that the machine learning unit uses a pedal coefficient representing the pedal characteristics of the vehicle for the simulated slope prediction model to generate a simulated slope prediction model corresponding to the pedal coefficient. According to this structure, when the generated simulated gradient prediction model is used for a different vehicle, a simulated gradient prediction model corresponding to the different vehicle can be easily generated. In addition, when the pedal setting is changed in the same vehicle, a simulated gradient prediction model corresponding to the changed pedal setting can be easily generated.

[0014] Preferably, the learning data acquisition unit acquires a road model including a correct or known road slope of at least a portion of the driving path as the learning data, and the machine learning unit performs machine learning on the correlation between the data after the road slope is removed from the simulated slope and the vehicle driving data, and generates a vehicle behavior load model representing the correlation between the vehicle behavior load data after the road slope is removed from the load data and the vehicle driving data. According to this structure, when the road slope (θ Road ) is known, it is possible to distinguish the road slope resistance (M×g×sinθ Road ) and the driving resistance variation caused by vehicle behavior (F Variable ).

[0015] Preferably, the learning data acquisition unit acquires first learning data consisting of the load data in a first vehicle, the vehicle driving data and the road model on which the first vehicle drives, and second learning data consisting of the load data in a second vehicle, the vehicle driving data and the road model on which the second vehicle drives, and the machine learning unit generates the vehicle behavior load model of the first vehicle based on the first learning data of the first vehicle, and generates the vehicle behavior load model of the second vehicle based on the second learning data of the second vehicle.

[0016] In addition, the vehicle test system of the present invention has a vehicle test device such as a dynamometer for applying a load to a test object during driving, wherein the test object is a vehicle or a part of the vehicle, and the vehicle test system is characterized by comprising: a model storage unit for storing a simulated slope prediction model representing the correlation between load data and vehicle driving data, wherein the load data includes a road slope resistance (M×g×sinθ Road ) and the driving resistance variation caused by vehicle behavior (F Variable), the vehicle driving data includes the vehicle speed, acceleration operation amount and braking operation amount when the vehicle is traveling on the driving path; and a load data calculation unit, which calculates the load data input to the vehicle test device based on the vehicle driving data and the simulated slope prediction model.

[0017] Preferably, the model storage unit stores a vehicle behavior load model representing the correlation between vehicle behavior load data and the vehicle driving data, the vehicle behavior load data representing the change in driving resistance caused by the vehicle behavior, and the load data calculation unit calculates the vehicle behavior load data based on the vehicle driving data and the vehicle behavior load model.

[0018] Preferably, the data of the first vehicle is used to test the second vehicle, and the load data calculation unit uses a road model of the driving path of the first vehicle and a vehicle behavior load model representing the correlation between vehicle behavior load data and the vehicle driving data to calculate the load data input into a vehicle test device for testing the second vehicle, wherein the vehicle behavior load data represents the change in driving resistance caused by the vehicle behavior of the second vehicle.

[0019] Preferably, the second vehicle is, for example, a vehicle with an advanced driving assistance system (ADAS vehicle) or an autonomous driving vehicle (AD vehicle). In addition, the second vehicle is not limited to an ADAS vehicle or an AD vehicle.

[0020] Moreover, preferably, the machine learning method of the present invention is a machine learning method for vehicle testing, which obtains learning data consisting of load data and vehicle driving data, wherein the load data includes the road slope resistance of the driving path and the driving resistance change caused by the vehicle behavior, and the vehicle driving data includes the vehicle speed, acceleration operation amount and braking operation amount when the vehicle is driving on the driving path, and machine learning is performed on the correlation between the load data and the vehicle driving data to generate a simulated slope prediction model representing the correlation between the load data and the vehicle driving data.

[0021] On this basis, the vehicle testing method of the present invention uses a dynamometer that applies a load to a test object during driving, and the test object is a vehicle or a part of the vehicle. The vehicle testing method is characterized in that a simulated slope prediction model that represents the correlation between load data and vehicle driving data is used to calculate the load data based on the vehicle driving data, the load data includes the road slope resistance of the driving path and the driving resistance change caused by the vehicle behavior, the vehicle driving data includes the vehicle speed, acceleration operation amount and braking operation amount when the vehicle is driving on the driving path, and the calculated load data is input into the dynamometer to test the vehicle.

[0022] According to the present invention thus constituted, it is possible to easily reproduce loads such as the gradient resistance during actual road travel. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] Figure 1 It is a schematic diagram showing a vehicle testing system according to a first embodiment of the present invention. Figure 2 It is a schematic diagram showing the machine learning and slope prediction of the first embodiment. Figure 3 It is a schematic diagram showing the procedure of the vehicle testing system according to the first embodiment. Figure 4 It is a schematic diagram showing the procedure of the vehicle testing system according to the second embodiment. Figure 5 It is a schematic diagram showing the generation of a simulated slope prediction model according to the third embodiment. Figure 6 Schematic diagram showing the procedure of the vehicle testing system according to the fourth embodiment. Figure 7 is the load data (simulated slope (θ Emulation ))'s calculation steps. Figure 8 Schematic diagram showing the procedure of the vehicle testing system according to the fifth embodiment. Fig. 9 It is a schematic diagram showing the procedure of the vehicle testing system in the modification of the fifth embodiment. Fig.10 is the load data (simulated slope (θ Emulation ))'s calculation steps. DETAILED DESCRIPTION

[0024] Hereinafter, various embodiments of the vehicle test system of the present invention will be described with reference to the accompanying drawings. In addition, for ease of understanding, any of the following figures are schematically depicted in an appropriately omitted or exaggerated manner. The same components are marked with the same reference numerals and the description thereof is appropriately omitted.

[0025] <First Embodiment> like Figure 1 As shown, the vehicle testing system 100 according to the first embodiment reproduces the travel load during actual road travel by means of road load simulation control in a test laboratory.

[0026] Specifically, the vehicle test system 100 includes a chassis dynamometer 2 for causing a vehicle as a test object to simulate driving, and a control device 3 for controlling the chassis dynamometer 2. The chassis dynamometer 2 includes a roller 21 for supporting a tire of the vehicle, and a dynamometer 22 connected to the roller 21 and applying a load to the vehicle W via the roller 21. The vehicle is not limited to a complete vehicle, but may be a part of a vehicle including an engine, a motor, a powertrain, and the like.

[0027] Before performing a vehicle test using the vehicle test system 100 , the following (1) to (3) are performed. (1) Road Test (2) Torque matching (3) Machine learning for simulated slope prediction model

[0028] (1) Road driving test The road driving test is a test in which the vehicle is driven on an actual road and vehicle driving data such as vehicle speed, acceleration operation amount, brake operation amount or steering operation amount during the actual road driving are obtained. The acceleration operation amount may also be the accelerator pedal position or the throttle opening. The brake operation amount may also be the brake pedal position or the brake pressure. The steering operation amount may also be the steering angle or the left and right wheel speed difference. In addition, in the road driving test, for example, the voltage or current of the battery, motor, etc., the warm-up state of the internal combustion engine (such as the cooling water temperature or the engine oil temperature, etc.), the charge state or deterioration state of the battery, or the internal or peripheral temperature of the battery, motor, etc. are obtained as vehicle information, the temperature, humidity or air pressure around the vehicle are obtained as environmental information, and the latitude, longitude or altitude are obtained as position information. In addition, the emission amount of the measured component contained in the exhaust gas discharged from the internal combustion engine (engine) of the vehicle may also be measured.

[0029] (2) Torque matching The torque matching is performed by causing the vehicle W to travel on the chassis dynamometer 2 to obtain a dynamometer load (torque matching load (RepForce)) applied to the vehicle W by the chassis dynamometer 2 .

[0030] Specifically, the torque matching controls the driving robot 4 mounted on the vehicle W and driving the vehicle W so as to achieve the acceleration operation amount, brake operation amount and / or shift position of the road driving test, and controls the chassis dynamometer 2 so that the vehicle speed of the vehicle W becomes the vehicle speed of the road driving test. The dynamometer load obtained from the dynamometer through these controls is obtained. In addition, the dynamometer load (torque matching load (RepForce)) thus obtained is used for the machine learning shown below. Moreover, through torque matching, it is possible to obtain the road slope (θ Road ) plus the driving resistance change (FVariable ) is converted into a value (simulated slope (θ Emulation )).

[0031] If the formula θ is expressed Emulation The export method is as follows. RepForce=A+BV+CV 2 +Iα+Mgsinθ Road +F Variable Mgsinθ Emulation =Mgsinθ Road +F Variable θ Emulation =sin -1 ((RepForce-(A+BV+CV 2 +Iα)) / Mg)

[0032] (3) Machine learning for simulated slope prediction model The machine learning Figure 1 The machine learning device 5 shown in the figure is performed. The machine learning device 5 includes a learning data acquisition unit 51 that acquires learning data for generating a simulated slope prediction model, and a machine learning unit 52 that generates a simulated slope prediction model using the learning data.

[0033] like Figure 2 As shown in “at the time of learning”, the learning data is composed of load data indicating the dynamometer load obtained by the above-mentioned torque matching and vehicle travel data for torque matching.

[0034] The load data used as learning data is, for example, the road slope (θ Road ) plus (here, plus) the running resistance change (F Variable ) is converted into a value (simulated slope (θ Emulation )). Alternatively, it can be “slope resistance (M×g×sinθ Road )” and “the change in driving resistance caused by vehicle behavior (F Variable ) converted into slope resistance".

[0035] The vehicle travel data used as learning data is the vehicle speed, accelerator operation amount, brake operation amount, steering operation amount, etc. during travel along the travel route, or calculated values ​​such as square values ​​or differential values ​​obtained from them.

[0036] The machine learning unit 52 performs machine learning on the correlation between the load data obtained by the torque matching and the vehicle travel data used for the torque matching, and generates a load data (simulated slope (θEmulation )) A simulated slope prediction model (machine learning model) based on the correlation between the vehicle driving data. Figure 2 As shown in the “Prediction” of FIG. 1 , by inputting vehicle driving data into the simulated slope prediction model, the simulated slope (θ) of the driving path obtained by the vehicle driving data can be calculated. Emulation ). In addition, the machine learning algorithm in the machine learning unit 52 is a machine learning method characterized by inputting past information and utilizing it in current and future predictions (for example, a recurrent neural network (RNN)).

[0037] <Testing Procedure of First Embodiment> Figure 3 The test procedure of the first embodiment is shown. A first road driving test (Road Test No. 1) is performed on a vehicle as a test object to obtain vehicle driving data serving as learning data.

[0038] Here, the vehicle driving data used as learning data preferably includes a wider driving area. Therefore, there is a tendency that the longer the number of trips or the driving time, the wider the driving area, the higher the learning accuracy or prediction accuracy of machine learning. For example, the calculated values ​​such as vehicle speed, acceleration and deceleration, steering operation amount, road slope, or their square values ​​or differential values ​​involve many aspects, and a driving test that includes their combination is preferred. In detail, for example, a 30-minute driving test is considered to include a combination of urban areas, mountain roads, and highways. In addition, it is considered that multiple drivers conduct driving tests on multiple driving routes to obtain vehicle driving data.

[0039] Then, the vehicle driving data obtained in the first road driving test is used to perform torque matching, and the measured RepForce is used to generate load data (simulated slope (θ Emulation )).

[0040] The dynamometer control unit 33 uses the load data (simulated slope (θ Emulation )) The dynamometer 22 is subjected to road load simulation (RLS) control, thereby performing a simulated driving test of the vehicle W on the chassis dynamometer 2. In addition, the driving of the vehicle in the road load simulation (RLS) control is considered to be performed by using a driving robot 4, but it can also be driven by a person.

[0041] On the other hand, the vehicle driving data obtained in the first road driving test and the load data (simulated slope (θ Emulation )) as learning data, and perform machine learning on their correlation to generate a representation of vehicle driving data and load data (simulated slope (θ EmulationThe simulated slope prediction model is stored in the model storage unit 31 of the control device 3 (refer to Figure 1 ). In addition, the simulated slope prediction model may also be stored in an information processing device different from the control device 3.

[0042] When the second to Nth road driving tests (Road Test No. 2 to N Road Test No. 2 to N) are performed, the vehicle driving data obtained in these road driving tests are input to the control device 3 without torque matching. As a result, the load data calculation unit 32 of the control device 3 inputs the vehicle driving data to the simulated slope prediction model to calculate the load data (simulated slope (θ)) for performing the simulated driving test on each road driving test. Emulation )). The dynamometer control unit 33 uses the calculated load data (simulated slope (θ Emulation )) The dynamometer 22 is subjected to road load simulation (RLS) control, thereby performing a simulated driving test of the vehicle W on the chassis dynamometer 2. In addition, the driving of the vehicle in the road load simulation (RLS) control is considered to be performed by using a driving robot 4, but it can also be driven by a person.

[0043] <Effects of the First Embodiment> According to the vehicle testing system 100 of the present embodiment configured in this way, it is possible to generate load data (simulated slope (θ Emulation )) and the vehicle driving data, the load data (simulated slope (θ Emulation )) includes the slope resistance of the driving path (M×g×sinθ Road ) and the driving resistance variation caused by vehicle behavior (F Variable ), the vehicle driving data includes the vehicle speed, acceleration operation amount and braking operation amount when the vehicle is driving on the driving path. Therefore, by inputting various vehicle driving data into the simulated slope prediction model, the load data (simulated slope (θ Emulation As a result, the slope resistance (M×g×sinθ) of the actual road can be Road ) and the driving resistance variation caused by vehicle behavior (F Variable ) etc. makes load reproduction easy.

[0044] <Second Embodiment> Next, a vehicle testing method and a machine learning method according to a second embodiment of the present invention will be described.

[0045] In the vehicle test method and the machine learning method of the second embodiment, the machine learning method of the machine learning device 5 is different from that of the first embodiment, such as Figure 4As shown, the simulated gradient prediction model is generated without performing torque matching using vehicle travel data obtained in a road travel test.

[0046] Specifically, the learning data acquisition unit 51 of the machine learning device 5 acquires learning data composed of load data (input gradient θ) used in an arbitrary road load simulation (RLS) and vehicle driving data obtained in a road driving test. In addition, the vehicle driving data is data generated based on actual driving data obtained by an actual vehicle driving on an actual driving route, or data generated based on virtual driving data obtained by a virtual vehicle driving on a virtual driving route.

[0047] Then, the machine learning unit 52 performs machine learning on the correlation between the load data (input slope θ) used in the road load simulation and the vehicle driving data used for the road load simulation, and generates a slope prediction model (machine learning model) representing the correlation between the load data (input slope θ) and the vehicle driving data. By inputting the vehicle driving data into the slope prediction model, the simulated slope (θ) of the driving route obtained by the vehicle driving data can be calculated. Emulation ).

[0048] <Effects of the Second Embodiment> According to the second embodiment, the simulated gradient prediction model is generated without performing torque matching using vehicle travel data obtained in a road travel test, so that the gradient resistance (M×g×sinθ) of actual road travel can be made Road ) and the driving resistance variation caused by vehicle behavior (F Variable ) etc. makes load reproduction easy.

[0049] <Third Embodiment> Next, a vehicle testing method and a machine learning method according to a third embodiment of the present invention will be described.

[0050] In the vehicle testing method and the machine learning method of the third embodiment, the structure of the machine learning unit 52 of the machine learning device 5 is different from those of the above-described embodiments.

[0051] Specifically, the machine learning unit 52 generates a simulated slope prediction model for each pedal characteristic using the simulated slope prediction model generated by the first embodiment or the second embodiment and different pedal coefficients. In addition, the pedal coefficient is a coefficient representing the pedal characteristic and can be represented by a value obtained by dividing the driving force by the pedal depression amount. Here, the vehicle speed or acceleration can also be added to the calculation of this value.

[0052] For example, Figure 5 As shown, the machine learning unit 52 uses the pedal coefficient K of the first vehicle in different vehicles (the first vehicle and the second vehicle).A The pedal coefficient K of the second vehicle B The ratio (K B / K A ) and the simulated slope prediction model f generated in the first vehicle A (x), through f B (x) = f A (x)×(K B / K A ), find the simulated slope prediction model f of the second vehicle B (x).

[0053] Furthermore, even for the same vehicle, when the pedal setting is changed (pedal setting for the energy-saving mode, pedal setting for the sports mode, etc.), the machine learning unit 52 uses the generated simulated slope prediction model f A (x) and the pedal coefficient K of the pedal setting when generating the simulated slope prediction model A The pedal coefficient K of the changed pedal setting C The ratio (K C / K A ), through f C (x) = f A (x)×(K C / K A ), find the simulated slope prediction model f of the changed pedal setting C (x).

[0054] <Effects of the Third Embodiment> According to the vehicle test system 100 of the present embodiment thus configured, in addition to the effects of the first and second embodiments, when the generated simulated slope prediction model is used for a different vehicle, a simulated slope prediction model corresponding to the different vehicle can be easily generated. Furthermore, when the pedal setting is changed in the same vehicle, a simulated slope prediction model corresponding to the changed pedal setting can be easily generated.

[0055] <Fourth Embodiment> Next, a vehicle testing method and a machine learning method according to a fourth embodiment of the present invention will be described.

[0056] In the vehicle testing method and the machine learning method of the fourth embodiment, the structure of the machine learning unit 52 of the machine learning device 5 is different from those of the above-described embodiments.

[0057] Specifically, the learning data acquisition unit 51 of the machine learning device 5 is as follows: Figure 6 As shown, the road slope (θ Road) as learning data for generating a slope prediction model. In addition, the road model at least includes vehicle information (such as vehicle speed or longitude and latitude position information) and road slope (θ Road ) The slope output model can also be road map data.

[0058] Then, the machine learning unit 52 performs machine learning on the correlation between the road model, the data after removing the road gradient from the simulated gradient, and the vehicle travel data, and generates a representation of the load data (simulated gradient (θ Emulation )) removes the road slope (θ Road ) is used to analyze the correlation between the vehicle behavior load data and the vehicle driving data.

[0059] exist Figure 6 An example of the procedure for generating a vehicle behavior load model and a road model in each road driving test is described in FIG.

[0060] The road slope (θ Road ) is defined as Road Test No. 1 and a road model is obtained. Then, the vehicle driving data and the load data (simulated slope (θ) obtained by torque matching) are obtained together with the road model. Emulation )) as learning data.

[0061] Then, the load data (simulated slope (θ Emulation )) removes the road slope (θ Road ) after the vehicle behavior load data (sin -1 (F Variable / Mg)) and the correlation between the vehicle driving data and machine learning to generate a representation from the load data (simulated slope (θ Emulation )) removes the road slope (θ Road ) after the vehicle behavior load data (sin -1 (F Variable / Mg)) and the vehicle behavior load model based on the correlation with the vehicle driving data.

[0062] When the second to Nth road driving tests (Road Test No. 2 to N) are performed, the slope resistance (M×g×sinθ) of the driving path in each road driving test can be obtained by removing the “vehicle behavior load data obtained based on the vehicle driving data of each road driving test and the above-mentioned vehicle behavior load model” from the “load data obtained based on the vehicle driving data of each road driving test and the above-mentioned simulated slope prediction model”. Road ). According to the road slope (θRoad ) and latitude and longitude information can generate road models for each road driving test.

[0063] Reference Figure 7 , indicating that the load data (simulated slope (θ Emulation )) calculation. The load data calculation unit 32 of the control device 3 inputs the vehicle driving data (vehicle speed, acceleration operation amount, brake operation amount, etc.) to the vehicle behavior load model, and calculates the driving resistance change amount (F Variable In addition, the load data calculation unit 32 inputs vehicle travel data (vehicle speed, longitude, latitude, etc.) into the slope output model included in the road model to calculate the road slope (θ Road ), and the road slope (θ Road ) is input into the model for calculating the load caused by the road slope, and the slope load (M×g×sinθ Road ). The load data calculation unit 32 calculates the running resistance variation (F Variable ) and slope load (M×g×sinθ Road ) are added together. The load data (simulated slope (θ Emulation )). Then, the dynamometer control unit 33 uses the load data (simulated slope (θ Emulation )), the dynamometer 22 is subjected to road load simulation (RLS) control, thereby performing a simulated driving test of the vehicle W on the chassis dynamometer 2. In addition, the driving of the vehicle in the road load simulation (RLS) control is considered to be performed by using a driving robot 4, but it can also be driven by a person.

[0064] <Effects of the Fourth Embodiment> According to the fourth embodiment, in addition to the effects of the above-described embodiments, the road gradient resistance (M×g×sinθ Road ) load model and the driving resistance variation caused by vehicle behavior (F Variable As a result, the road gradient resistance (M×g×sinθ) can be distinguished and calculated based on any vehicle driving data. Road ) and the driving resistance variation caused by vehicle behavior (F Variable ), so the obtained road model can be flexibly applied to tests and simulations in other vehicles.

[0065] <Fifth Embodiment> Next, a vehicle testing method and a machine learning method according to a fifth embodiment of the present invention will be described.

[0066] In the vehicle testing method and machine learning method of the fifth embodiment, a vehicle behavior load model of a first vehicle and / or a road model of a path along which the first vehicle travels is generated, and a vehicle behavior load model of a second vehicle and / or a road model of a path along which the second vehicle travels is generated.

[0067] Specifically, the learning data acquisition unit 51 of the machine learning device 5 acquires the load data (simulated slope (θ Emulation )), first learning data consisting of vehicle travel data and a road model, and load data in the second vehicle (simulated slope (θ Emulation )), the second learning data consists of vehicle driving data and road model.

[0068] Then, the machine learning unit 52 generates a vehicle behavior load model of the first vehicle and / or a road model of the path along which the first vehicle is traveling based on the first learning data of the first vehicle, and generates a vehicle behavior load model of the second vehicle and / or a road model of the path along which the second vehicle is traveling based on the second learning data of the second vehicle.

[0069] <Vehicle Evaluation in Fifth Embodiment> like Figure 8 and Fig. 9 As shown, the machine learning device 5 can construct a database of road models, a database of vehicle behavior load models, or a database of driver models based on the test results of various vehicles (first vehicle) in the past. In addition, the driver model is a model that represents the driving characteristics of the driver. It can be a model obtained based on the data of past vehicle tests (vehicle speed, engine status, accelerator pedal usage, brake pedal usage, steering angle, etc.), a model obtained based on time series vehicle speed and shift data, a model obtained through a prescribed calculation, or a simple vehicle speed curve can be used directly.

[0070] When the vehicle to be developed (second vehicle) is an AD vehicle (refer to Figure 8 ) Evaluation is performed by combining the vehicle behavior load model obtained from the results of the road driving test of the vehicle to be developed, the driving method (AD system) based on the AD control of the vehicle to be developed, and the road model in the database. As a result, AD vehicles can be evaluated in various scenarios (driving curves) such as various driving routes. For example, it is possible to evaluate whether the AD system is operating correctly or whether the dynamic behavior of the vehicle is as expected.

[0071] When the vehicle to be developed (second vehicle) is an ADAS vehicle (refer to Fig. 9 ) The evaluation is performed by combining the vehicle behavior load model obtained from the results of the road driving test of the vehicle to be developed, the road model in the database, and the driver model. As a result, ADAS vehicles can be evaluated in various scenarios (driving curves) such as various driving routes. For example, it is possible to evaluate whether the ADAS system is operating correctly, or to evaluate whether the dynamic behavior of the vehicle is as expected. The driving of the vehicle in this evaluation is considered to be performed by using a driving robot 4, but it can also be driven by a person.

[0072] The vehicle test of the AD / ADAS vehicle described above is assumed to be performed when a new AD / ADAS system is developed, when the same AD / ADAS system is applied to other vehicles, when an AD / ADAS system is retrofitted to an existing vehicle, and the like.

[0073] <Effects of the Fifth Embodiment> According to the vehicle test method of the fifth embodiment, various scenarios can be generated using the gradient load model of the first vehicle obtained through past vehicle tests, the driver model, etc. As a result, a plurality of experience-based evaluation scenarios can be prepared. In particular, by executing the plurality of scenarios on a bench test device or in a simulated manner, a high-quality (experience-based scenario, reproducible test method) evaluation can be performed in a short time compared to tests on roads or test routes.

[0074] <Other Implementation Methods> For example, the machine learning unit 52 may also generate a road gradient (θ Road ) road model, load data (simulated slope (θ Emulation )) and the correlation of vehicle driving data is machine learned to generate a slope resistance (M×g×sinθ Road ) and the vehicle driving data, and the slope load model representing the correlation between the load data (simulated slope (θ Emulation )) removes the road slope (θ Road ) is used to analyze the correlation between the vehicle behavior load data and the vehicle driving data.

[0075] In this case, if Fig.10 As shown, the load data calculation unit 32 of the control device 3 inputs the vehicle driving data (vehicle speed, acceleration operation amount, brake operation amount, etc.) into the vehicle behavior load model, and calculates the driving resistance change amount (F Variable In addition, the load data calculation unit 32 inputs the vehicle driving data (vehicle speed, acceleration operation amount, brake operation amount, etc.) into the gradient load model to calculate the gradient load (M×g×sinθ Road ). The load data calculation unit 32 calculates the running resistance variation (FVariable ) and slope load (M×g×sinθ Road ) are added together. The added value becomes the load data (simulated slope (θ Emulation )). Then, the dynamometer control unit 33 uses the load data (simulated slope (θ Emulation )), the dynamometer 22 is subjected to road load simulation (RLS) control, thereby performing a simulated driving test of the vehicle W on the chassis dynamometer 2.

[0076] Furthermore, the vehicle test system of each of the above embodiments uses a chassis dynamometer that applies a load to the vehicle, but a dynamometer that applies a load to a part of the vehicle (e.g., a powertrain, an engine, etc.) may also be used. That is, the machine learning model obtained in each of the above embodiments can be used for a test using a part of the vehicle as a test object or a full simulation test.

[0077] Furthermore, when the measured object is an ADAS vehicle or an AD vehicle, the acceleration operation amount and the brake operation amount included in the vehicle travel data may use vehicle signals instead of physical operation amounts.

[0078] Furthermore, the road surface conditions of the travel route (eg, road surface friction coefficient, puddles, unevenness, etc.) may be included in the learning data, so that a slope prediction model can be generated that takes the road surface conditions of the travel route into consideration.

[0079] On this basis, the learning data may also include vehicle information such as vehicle weight, temperature of various parts of the vehicle, battery charge state, etc. Thus, a simulated slope prediction model, vehicle behavior load model, or road slope load model that takes into account differences in vehicle conditions can be generated.

[0080] Furthermore, the learning data may include, for example, environmental information such as ambient temperature, humidity, altitude, etc. Thus, a simulated slope prediction model taking environmental information into consideration can be generated.

[0081] In addition, the load data used for learning data is the road slope (θ Road ) plus the driving resistance change (F Variable ) is converted into a value (simulated slope (θ Emulation )), but can also be regarded as the overall driving load (rolling resistance (A+B×V), air resistance (C×V 2 ), inertial resistance (I×α), slope resistance (M×g×sinθ Road ) and the driving resistance variation caused by vehicle behavior (F Variable )).

[0082] In each of the above embodiments, the calculated load data (simulated slope (θ Emulation )) is input to the dynamometer 22 to perform a simulated driving test of the vehicle as the test object, but the calculated load data (simulated slope (θ Emulation )), conduct simulation on computer (experiments in virtual environment).

[0083] Furthermore, in each of the above-described embodiments, load data obtained by torque matching is used. However, for example, loads measured during actual road travel, such as loads calculated from torque applied to an axle, may be used for machine learning.

[0084] Furthermore, various modifications and combinations of the embodiments are possible as long as they do not depart from the spirit of the present invention. Industrial Applicability

[0085] According to the present invention, it is possible to easily reproduce loads such as road gradient resistance during actual road travel. Description of Reference Numerals

[0086] 100…Vehicle test system 2…Chassis dynamometer 3…Control device 31…Model storage unit 32…Load data calculation unit 33…Dynamometer control unit 4… Driving a robot 5…Machine Learning Devices 51…Learning data acquisition unit 52…Machine Learning Department

Claims

1. A machine learning device for generating a simulated slope prediction model for vehicle testing, the machine learning device being characterized by: a learning data acquisition unit that acquires learning data consisting of load data and vehicle travel data, wherein the load data includes a road gradient resistance of a travel route and a travel resistance variation caused by vehicle behavior, and the vehicle travel data includes a vehicle speed, an acceleration operation amount, and a brake operation amount when the vehicle travels on the travel route; and The machine learning unit performs machine learning on the correlation between the load data and the vehicle travel data, and generates a simulated slope prediction model representing the correlation between the load data and the vehicle travel data.

2. The machine learning device according to claim 1, characterized in that The load data includes a load calculated based on a simulated gradient obtained by adding a gradient-converted value of a running resistance variation amount caused by vehicle behavior to a road gradient of the running route.

3. The machine learning device according to claim 1 or 2, characterized in that: The learning data acquisition unit acquires learning data consisting of load data used in the road load simulation and vehicle travel data used at this time, The machine learning unit performs machine learning on the correlation between the acquired load data and the vehicle travel data, and generates a simulated slope prediction model indicating the correlation between the load data and the vehicle travel data.

4. The machine learning device according to any one of claims 1 to 3, characterized in that The machine learning unit uses a pedal coefficient representing a pedal characteristic of a vehicle for the simulated gradient prediction model, and generates a simulated gradient prediction model corresponding to the pedal coefficient.

5. The machine learning device according to any one of claims 2 to 4, characterized in that: The learning data acquisition unit acquires a road model including a correct or known road slope of at least a portion of the driving route as the learning data, The machine learning unit performs machine learning on the correlation between the data obtained by removing the road gradient from the simulated gradient and the vehicle driving data, and generates a vehicle behavior load model representing the correlation between the vehicle behavior load data obtained by removing the road gradient from the load data and the vehicle driving data.

6. The machine learning device according to claim 5, characterized in that The learning data acquisition unit acquires first learning data consisting of the load data in the first vehicle, the vehicle driving data, and the road model on which the first vehicle drives, and second learning data consisting of the load data in the second vehicle, the vehicle driving data, and the road model on which the second vehicle drives, The machine learning unit generates the vehicle behavior load model of the first vehicle based on first learning data of the first vehicle, and generates the vehicle behavior load model of the second vehicle based on second learning data of the second vehicle.

7. A vehicle testing system, comprising a dynamometer for applying a load to a test object during driving, wherein the test object is a vehicle or a part of the vehicle, wherein the vehicle testing system comprises: a model storage unit storing a simulated gradient prediction model representing a correlation between load data including a road gradient resistance of a driving route and a driving resistance variation caused by vehicle behavior and vehicle driving data including a vehicle speed, an acceleration operation amount, and a braking operation amount when the vehicle is driving on the driving route; and The load data calculation unit calculates load data to be input to the dynamometer based on the vehicle travel data and the simulated gradient prediction model.

8. The vehicle testing system according to claim 7, characterized in that: The model storage unit stores a vehicle behavior load model indicating a correlation between vehicle behavior load data and the vehicle travel data, wherein the vehicle behavior load data indicates a travel resistance variation amount caused by the vehicle behavior. The load data calculation unit calculates the vehicle behavior load data based on the vehicle travel data and the vehicle behavior load model.

9. The vehicle testing system according to claim 8, characterized in that: Using the data from the first vehicle to conduct a test on the second vehicle, The load data calculation unit calculates load data to be input into a vehicle test device for testing the second vehicle using a road model of the driving route of the first vehicle and a vehicle behavior load model representing the correlation between vehicle behavior load data and the vehicle driving data, wherein the vehicle behavior load data represents a change in driving resistance caused by the vehicle behavior of the second vehicle.

10. The vehicle testing system according to claim 9, characterized in that: The second vehicle is a vehicle with an advanced driver assistance system or an autonomous driving vehicle.

11. A machine learning method for vehicle testing, characterized in that: Acquire learning data consisting of load data and vehicle driving data, wherein the load data includes a road gradient resistance of a driving route and a driving resistance variation caused by vehicle behavior, and the vehicle driving data includes a vehicle speed, an acceleration operation amount, and a braking operation amount when the vehicle is driving on the driving route, Machine learning is performed on the correlation between the load data and the vehicle driving data to generate a simulated slope prediction model representing the correlation between the load data and the vehicle driving data.

12. A vehicle testing method, using a dynamometer for applying a load to a test object during driving, wherein the test object is a vehicle or a part of the vehicle, wherein: The load data is calculated based on the vehicle driving data using a simulated slope prediction model representing the correlation between the load data and the vehicle driving data, wherein the load data includes the road slope resistance of the driving path and the driving resistance variation caused by the vehicle behavior, and the vehicle driving data includes the vehicle speed, acceleration operation amount and braking operation amount when the vehicle is driving on the driving path, The calculated load data is input into the dynamometer to perform a test on the vehicle.

Images