Vehicle component response learning method, vehicle component response calculation method, vehicle component response learning system, and storage medium

Abstract

The present invention is a vehicle component response learning method that does not perform actual road travel but accurately obtains response data of a vehicle component in a desired driving environment through simulation, and generates a learning completion model related to the response of the vehicle component that is a vehicle or a part thereof, including: (1) an input step of providing an input including parameters associated with a vehicle speed, a load, and a temperature that are assumed to be actual road travel to the vehicle component; (2) an acquisition step of acquiring response data of the vehicle component, and acquiring input data indicating the input and the response data as training data; and (3) a generation step of generating a learning completion model related to the response of the vehicle component using machine learning based on the training data.

Description

Technical Field

[0001] This invention relates to a vehicle component response learning method, a vehicle component response calculation method, a vehicle component response learning system, a vehicle component response learning device, and a vehicle component response learning program. Background Technology

[0002] In recent years, research has been conducted on transient modeling of exhaust gases and other materials using Dynamic DoE (see Non-Patent Literature 1, 2).

[0003] In transient modeling of exhaust gases, a transient exhaust gas model is created using machine learning. Furthermore, to create the transient exhaust gas model, ECU parameters related to engine combustion are used as input parameters to obtain training data. Other ECU parameters include, for example, ignition timing and valve timing for controlling engine intake and exhaust. Existing technical documents Non-patent literature

[0004] Non-patent literature 1: "Dynamic Modelling for Gasoline Direct Injection Engines", Taro Shishido, 5 others, Keihin Technical Review Vol. 6 (2017). Non-patent literature 2: "Artificial neural network (ANN) assisted prediction of transient NOx emissions from a high-speed direct injection (HSDI) diesel engine", Xiao Hang Fang, 4 other authors, International Journal of Engine Research (April 27, 2021). Summary of the Invention The technical problem to be solved by the present invention

[0005] However, while methods for creating transient exhaust gas models are disclosed in the aforementioned non-patent documents 1 and 2, no methods are disclosed for creating transient exhaust gas models that take into account the actual road driving environment. Furthermore, in order to use such transient exhaust gas models for predicting actual road exhaust gases, it is necessary to correlate them with other input parameters based on the actual road environment (vehicle speed, accelerator opening, elevation, external air temperature, etc.), and further effort is required to improve the prediction accuracy in actual road driving simulations.

[0006] Furthermore, in the aforementioned non-patent documents 1 and 2, it is necessary to set ECU parameters related to engine combustion in order to create a transient exhaust gas model. Examples of ECU parameters include engine ignition timing and valve timing for controlling engine intake and exhaust. These parameters are only accessible to a limited number of engineers, such as OEMs or Tier 1 (primary contractors). Moreover, to understand the impact of these parameters on the engine during dynamic changes, repeated verification tests and multivariate analysis of large amounts of data are required. Furthermore, changes in parameter settings, combinations, or timing errors can lead to engine damage. Therefore, in reality, the creation of transient exhaust gas models using Dynamic DoE can only be carried out by a limited number of engineers with access to control devices such as ECUs.

[0007] Therefore, the present invention was made to solve the above-mentioned problems, and its main objective is to obtain the response data of vehicle components under the desired driving environment with high precision through simulation without actual road driving. Technical solutions for solving technical problems

[0008] That is, the vehicle component response learning method of the present invention generates a learning completion model related to the response of a vehicle component as a vehicle or a part of the vehicle, characterized by comprising the following steps: an input step, providing input containing parameters related to hypothetical actual road driving speed, load, and temperature to the vehicle component; an acquisition step, acquiring response data of the vehicle component, acquiring input data representing the input and the response data as training data; and a generation step, using machine learning based on the training data to generate a learning completion model related to the response of the vehicle component.

[0009] In this vehicle component response learning method, input parameters relating to hypothetical real-world driving conditions such as vehicle speed, load, and temperature are provided to the vehicle component. The resulting response data is then obtained. Based on training data consisting of the input data representing the inputs provided to the vehicle component and the response data, machine learning (including statistical methods) is used to generate a learned model related to the vehicle component's response. As a result, by using this learned model, model input data for the vehicle component under desired driving conditions and response data output from the learned model can be obtained through simulation without actual road driving. Furthermore, since control parameters such as the engine's combustion ECU parameters are not changed, response models can be easily and safely generated without the need for repeated trials to create the most suitable experimental plan and the analysis of large amounts of data as a result.

[0010] To more faithfully reproduce the driving environment of actual road travel, it is effective to consider changes in air pressure in addition to temperature variations. Therefore, in order to obtain vehicle component response data under the desired environment with high accuracy through simulation, it is preferable that, in the input step, inputs containing parameters related to air pressure under the assumption of actual road travel, in addition to those related to vehicle speed, load, and temperature, are provided to the vehicle component.

[0011] As a specific implementation method, it is preferred that, in the input step, the parameters are varied within their respective ranges of variation for the vehicle speed, the load, the temperature, and the air pressure, and the input as a combination of them is provided to the vehicle component.

[0012] To vary the entire range of environmental factors (temperature and air pressure) within a desired predicted time period or within a desired predicted range of movement, it is sometimes impossible to achieve this using only a temperature variation device or an air pressure variation device. Furthermore, depending on the user's simulation, there may be cases where a portion of the range of environmental factors (temperature and air pressure) variation is sufficient. Therefore, it is preferable that, in the input step, the range of temperature and air pressure variation is divided into multiple segments, and the segmented input obtained by dynamically varying the temperature and air pressure within these segments is provided to the vehicle component. The acquisition step acquires the response data of the vehicle component provided with the segmented input, and acquires the segmented input data representing the segmented input and the response data as segmentation training data.

[0013] In order to generate continuous data between the segmented blocks, it is preferable that the input step provides the vehicle component with input obtained by dynamically varying the temperature and air pressure at the boundaries between the segmented blocks.

[0014] In order to vary the temperature and air pressure during the input step, it is preferable to vary the temperature and air pressure by means of an environmental variation device that is connected to the vehicle component and varies the temperature and pressure, or by means of an environmental variation chamber that houses the vehicle component and varies the temperature and pressure.

[0015] To effectively obtain training data, it is preferable that the input step uses the DoE method to generate a dynamic experiment plan consisting of the combination of the parameters.

[0016] As a specific implementation of the response data, preferably, when the vehicle component is a vehicle or a part thereof that includes at least an engine, the response data is exhaust gas data; when the vehicle component is a vehicle or a part thereof that includes at least a secondary battery, the response data is power consumption data, current data, voltage data, SOC data and / or battery temperature data (e.g., cell temperature data or module temperature data, etc.); when the vehicle component is a vehicle or a part thereof that includes at least a fuel cell, the response data is hydrogen consumption data, oxygen consumption data, power generation current data, power generation voltage data and / or battery temperature data.

[0017] As a specific implementation method, it can be considered that the exhaust gas data is measured using an exhaust gas analysis device, the power consumption data, the current data, the voltage data, the SOC data, or the temperature data are measured using a power consumption meter, ammeter, voltmeter, or thermometer, and the hydrogen consumption data, oxygen consumption data, power generation current data, power generation voltage data, or battery temperature data are measured using a hydrogen meter, oxygen meter, ammeter, voltmeter, or thermometer.

[0018] As a specific implementation method, it can be considered that: when the vehicle component is a vehicle or a part thereof that includes at least a secondary battery, the parameter associated with the vehicle speed or the load is the charging current and / or discharging current, and the parameter associated with the temperature is the ambient temperature of the secondary battery; when the vehicle component is a vehicle or a part thereof that includes at least a fuel cell, the parameter associated with the vehicle speed or the load is the hydrogen supply, oxygen supply and / or water supply, and the parameter associated with the temperature is the ambient temperature of the fuel cell.

[0019] Furthermore, as a vehicle component response calculation method of the present invention, it is characterized by using a learned model generated by the above-described vehicle component response learning method to calculate the response data of the vehicle component during actual road driving. Specifically, the vehicle component response calculation method uses output parameters (vehicle speed, load, temperature, or pressure) generated based on the learned model and a model capable of simulating actual road driving (e.g., IPG CarMaker) to calculate the response data of the vehicle component during actual road driving. If this is the vehicle component response calculation method, then the response data of the vehicle components under the desired driving environment can be obtained through simulation without actual road driving.

[0020] Furthermore, the vehicle component response learning system of the present invention generates a learning completion model related to the response of a vehicle component that is a vehicle or part of the vehicle, characterized in that it comprises: an input unit that provides input including parameters related to a hypothetical actual road driving speed, load, and temperature to the vehicle component; an acquisition unit that acquires response data of the vehicle component and acquires input data representing the input and the response data as training data; and a generation unit that, based on the training data, uses machine learning to generate a learning completion model related to the response of the vehicle component.

[0021] Specifically, preferably, the input unit includes: a dynamometer for applying a load to the vehicle component; and an environmental variation device having a test chamber for housing the vehicle component and for varying the temperature of the test chamber, or an environmental variation device connected to the vehicle component and for varying the temperature.

[0022] Furthermore, the vehicle component response generation program of the present invention generates a learning completion model related to the response of a vehicle component that is a vehicle or part of the vehicle. The program is characterized in that a computer functions as an input data production unit, an acquisition unit, and a generation unit. The input data production unit produces input data representing parameters to be provided to the vehicle component, including parameters related to hypothetical actual road driving speed, load, and temperature. The acquisition unit acquires the response data of the vehicle component and uses the input data and the response data as training data. The generation unit generates a learning completion model related to the response of the vehicle component based on the training data, using statistical methods or machine learning. Invention Effects

[0023] According to the present invention described above, it is possible to obtain the response data of vehicle components under the desired driving environment with high precision through simulation without actual road driving. Attached Figure Description

[0024] Figure 1 This is a schematic diagram illustrating the configuration of a vehicle component response learning system according to one embodiment of the present invention. Figure 2 The diagram (left) shows the blocks representing temperature and air pressure (elevation) in the same implementation method, and the diagram (right) shows the combination of vehicle speed (engine speed) and load (accelerator pedal position). Figure 3 This is a flowchart of a vehicle component response learning method implemented in the same way. Figure 4 This is a schematic diagram illustrating the process up to the generation and simulation of the exhaust gas model in the same implementation method. Figure 5 This is a diagram schematically illustrating the configuration of a simulation device according to the same implementation method. Figure 6 The results are simulations of CO2 emissions from actual driving using the same exhaust gas model implemented in the same way. Figure 7 This is a diagram illustrating an example of waste gas concentration prediction in a modified implementation. Figure 8 The diagram shows (a) a diagram of an experimental system for generating a learning-completed model related to the response of an actual battery (secondary battery), (b) a flowchart of the generation of the battery model, and (c) a schematic diagram of a simulation using the battery model. Figure 9 The results are simulations of the State of Charge (SOC) during or after actual driving using a battery model with a modified implementation. Figure 10 The diagram shows (a) a test system for generating a learning completion model related to the response of a fuel cell, (b) a flowchart of the generation of the fuel cell model, and (c) a schematic diagram of a simulation using the fuel cell model. Detailed Implementation

[0025] Hereinafter, a vehicle component response learning system according to one embodiment of the present invention will be described with reference to the accompanying drawings.

[0026] <Device Configuration of Vehicle Component Response Learning System 100> The vehicle component response learning system 100 of this embodiment is a system that generates a learning completion model related to the response of a vehicle component that is a part of an actual vehicle. This learning completion model is used to calculate, for example, exhaust gas data or energy consumption data of the vehicle component through simulated calculations of road driving tests that comply with regulations. Here, road driving tests that comply with regulations are, for example, the Road Driving Emission (RDE) tests introduced in countries such as Europe. Furthermore, in addition to being used for road driving tests that comply with regulations, the generated learning completion model can also be used to calculate exhaust gas data or energy consumption data of the vehicle component during road driving tests in the development phase of the vehicle component.

[0027] Specifically, such as Figure 1 As shown, the vehicle component response learning system 100 includes the following: (1) an input unit 2, (2) an acquisition unit 3, and (3) a generation unit 4. Furthermore, in the following, the vehicle component is an actual engine, and the response data is exhaust gas data from the actual engine. (1) Input unit 2 provides the actual engine with inputs including parameters related to the engine speed, load, temperature and air pressure as the vehicle speed under the assumption of actual road driving. (2) Obtaining part 3, obtaining actual engine exhaust data, obtaining input data representing input and exhaust data as training data. (3) Generation Unit 4, based on the training data, uses machine learning (including statistical methods) to generate a learning completion model (exhaust gas model) related to the exhaust gas of the actual engine.

[0028] <Input Section 2> The input unit 2 provides the actual engine E, which is a vehicle component, with input consisting of a combination of engine speed, load, temperature and air pressure. It includes: an input data generation unit 21, which generates input data representing the combination of parameters; and an input device 22, which provides the input to the actual engine E based on the combination of parameters determined by the input data generation unit 21.

[0029] The input data generation unit 21 is composed of the control device 5, which uses the Dynamic Experiment Planning (DoE) method to determine the combination of parameters (experiment plan) required to obtain the training data for the generation step. Furthermore, the control device 5 is a computer equipped with a CPU, internal memory, input / output interfaces, input devices such as a keyboard, output devices such as a display, and communication devices. Moreover, the control device 5 functions as a vehicle component response learning device by storing a vehicle component response learning program in its internal memory.

[0030] Specifically, the input data generation unit 21 uses the Dynamic Experimental Planning (DoE) method to dynamically vary each parameter within its respective range of engine speed, load, temperature, and air pressure, determining the combination of these parameters. Alternatively, vehicle speed or the output shaft speed of the transmission can be used as a parameter instead of engine speed. Load can be parameters such as accelerator pedal position (accelerator pedal position). Furthermore, the input data generation unit 21 can also use environmental factors other than temperature and air pressure as parameters.

[0031] like Figure 2 As shown in the left figure, the input data generation unit 21 of this embodiment divides the temperature and air pressure variation range into multiple segments, and determines a combination of dynamically varying temperature and air pressure in each of these segments. Furthermore, Figure 2 The left figure shows an example of dividing the range of temperature and air pressure variations into 12 segments. Figure 2 In the left graph, the horizontal axis represents temperature, and the vertical axis represents altitude (air pressure). Additionally, Figure 2 The right-hand diagram shows the combination of engine speed (engine RPM) and accelerator pedal position. Figure 2 In the right-hand diagram, the horizontal axis represents engine speed (engine RPM), and the vertical axis represents the accelerator pedal position.

[0032] Furthermore, the input data production department has 21 pairs. Figure 2 The combination of engine speed (engine RPM) and accelerator pedal position in the right-hand diagram, plus... Figure 2 The combination of dynamically varying temperature and air pressure in each segment of the left figure determines the segmented input provided to the actual engine E.

[0033] The input device 22 includes: a dynamometer 22a, connected to the actual engine E; an engine speed input unit 22b, which inputs engine speed data to the dynamometer control device 22a1 of the dynamometer 22a; an accelerator opening input unit 22c, which inputs accelerator opening data to the engine control unit (ECU) of the actual engine E; an environmental variation device 22d, which varies the temperature and pressure of the test chamber 22d1 housing the actual engine E; and an environmental factor input unit 22e, which inputs temperature data and air pressure data to the environmental variation device 22d. Alternatively, an environmental variation device connected to the actual engine E and used to vary the temperature and pressure inside the actual engine E can also be used as the input device 22.

[0034] The engine dynamometer 22a rotates the actual engine E to a specified engine speed based on the engine speed data input through the engine speed input unit 22b.

[0035] The engine speed input unit 22b performs its function through the dynamometer control device, inputting engine speed data contained in the combination determined by the input data generation unit 21 to the engine dynamometer 22a.

[0036] The accelerator opening input unit 22c performs its function through the control device 5, inputting load data (accelerator opening data) contained in the combination determined by the input data generation unit 21 to the ECU of the actual engine E.

[0037] The environmental change device 22d includes: a test chamber 22d1 housing the actual engine E; and a temperature and pressure change device 22d2 connected to the actual engine E housed in the test chamber 22d1, which causes the temperature and pressure of the actual engine E to change. Furthermore, the temperature and pressure change device 22d2 changes the temperature and pressure of the test chamber 22d1 based on temperature and pressure data input from the environmental factor input unit 22e.

[0038] The environmental factor input unit 22e performs its function through the control device 5, inputting temperature data and air pressure data included in the combination determined by the input data production unit 21 into the temperature and pressure variation device 22d2.

[0039] In this embodiment, the input data generation unit 21 divides the temperature and air pressure variation range into multiple parts, and determines the combination (segmented input) obtained by dynamically changing the temperature and air pressure in the segmented block. The input device 22 provides the input to the actual engine E for each segmented input.

[0040] <Part 3> The acquisition unit 3 includes: an exhaust gas analysis device 31, which acquires exhaust gas data of the actual engine E provided with input by the input device 22; and a training data acquisition unit 32, which acquires input data representing the input provided to the actual engine E and exhaust gas data obtained by the exhaust gas analysis device 31 as training data (learning dataset).

[0041] The exhaust gas analysis device 31 analyzes at least one of the following in the exhaust gas emitted from the actual engine E: carbon dioxide (CO2), carbon monoxide (CO), particulate matter (PM), particulate matter number (PN), ammonia (NH3), nitrous oxide (N2O), nitrogen oxides (NOx), formaldehyde (HCHO), and / or total hydrocarbons (THC). The exhaust gas data measured by the exhaust gas analysis device 33 is sent to the training data acquisition unit 32.

[0042] The training data acquisition unit 32 functions via the control device 5, acquiring exhaust gas data sent from the exhaust gas analysis device 31, and also acquiring input data representing the input provided to the actual engine E when acquiring the exhaust gas data. Furthermore, the training data acquisition unit 32 correlates this exhaust gas data and the input data as training data. Additionally, the training data acquisition unit 32 sends the training data to the generation unit 4.

[0043] In this embodiment, the input data generation unit 21 divides the temperature and air pressure variation range into multiple segments and determines the combination (segmented input) obtained by dynamically changing the temperature and air pressure in the segmented blocks. The training data acquisition unit 32 acquires training data (segmented training data) for each segmented block.

[0044] <Generation Section 4> The generation unit 4 functions through the control device 5, using machine learning methods such as neural networks (NNs) to generate a learning model (exhaust gas model) related to the exhaust gas of the actual engine, based on training data (in this case, segmented training data). Furthermore, the machine learning in this embodiment includes statistical methods such as multiple regression analysis, deep learning, and artificial intelligence concepts.

[0045] The generation unit 4 in this embodiment can also use the training data of each segmented block to generate a learned model (exhaust gas model) for each segmented block, or it can merge the training data of each segmented block to generate a learned model (exhaust gas model) of the overall range of temperature and air pressure variation.

[0046] <Vehicle Component Response Learning Method> Next, the vehicle component response learning method using the above-mentioned vehicle component response learning system 100 will be described.

[0047] like Figure 3 As shown, the vehicle component response learning method includes: input step S1, providing the actual engine E with inputs that assume the actual road driving speed, load, temperature and air pressure as parameters; acquisition step S2, acquiring the exhaust gas data of the actual engine E, and acquiring the input data representing the input and the exhaust gas data as training data; and generation step S3, using machine learning based on the training data to generate a learning completed model (exhaust gas model) related to the exhaust gas of the actual engine E.

[0048] In input step S1, as described above, the input data generation unit 21 uses the Dynamic Experiment Planning (DoE) method to determine the segmented input that dynamically varies the temperature and pressure for each of the multiple blocks obtained by dividing the temperature and pressure variation range. Furthermore, in Figure 4In the "Dynamic Test Design," an example of a transient test sequence generated by the input data generation unit 21 is shown. Furthermore, the segmented input determined by the input data generation unit 21 is input to the actual engine via the input device 22 (see [reference]). Figure 4 The “Dynamic Test”.

[0049] In step S2, exhaust gas data of the actual engine E is acquired whenever each combination is input to the actual engine E. Then, the training data acquisition unit 32 acquires the exhaust gas data together with the input data when acquiring the exhaust gas data as segmentation training data.

[0050] In step S3, based on the training data or segmented training data, machine learning is used to generate an exhaust gas model of the actual engine E. Additionally, in... Figure 4 An example of exhaust gas modeling is shown in "Dynamic Modelling". The generated learned model (exhaust gas model) is stored in the memory of control device 5.

[0051] <Simulation using a learned completion model (exhaust gas model)> Next, refer to Figure 5 The simulation device 200 (information processing device) for simulating exhaust gas data of an actual engine E during road driving tests using the generated learning completion model (exhaust gas model) is described.

[0052] The simulation device 200 is a computer that uses a driving environment model that models the driving environment, a driving mode model that models the driving mode, a vehicle model that models the actual vehicle, and a generated learning model (exhaust gas model) to simulate driving tests on the road. It is equipped with a CPU, internal memory, input / output interfaces, input devices such as a keyboard, output devices such as a display, and communication devices.

[0053] The simulation device 200 includes: a relational data storage unit 201 for storing various models; and a simulation unit 202 for simulating the driving according to the RDE test.

[0054] The driving environment model stored in the relational data storage unit 201 is a model obtained by quantifying the driving road, the road surface resistance, markings, temperature and humidity, the elevation of the driving road, and the degree of congestion (traffic volume) of the driving road.

[0055] In addition, for example, when using the software "CarMaker" from IPG Automotive Co., Ltd., the driving mode model stored in the relational data storage unit 201 is a model obtained by quantifying Dynamics, Energy Efficiency, and Nervousness.

[0056] Specifically, the driving style model includes, for example, the following parameters listed in "CarMaker". Dynamics parameters: Cruising speed [km / h], Corner cutting coefficient [-], dt Change of pedals [s], Max. Long Acceleration [m / s²] 2 Max. Long Disceleration (maximum sustained negative acceleration) [m / s²] 2 Max. Lat Acceleration (maximum lateral acceleration) [m / s²] 2 Exponent of GG Diagram [-], Time of Shifting [s], Engine Speeds min / mas [rpm], Engine Speeds idle up / acceleration down [rpm], Long Smooth Throttle Limit [-] Parameters of Energy Efficiency: Min. dt Accel / Decel (time derivative of minimum acceleration / negative acceleration) [s], Long Drag Torque Braking [-], Long Drive Cycle Tol [-], Long Drive Cycle Coef [-] The parameters for Nervousness are: Long SDV Random[-], Long SDV Random f[-].

[0057] Furthermore, the vehicle model stored in the relational data storage unit 201 includes at least an engine model obtained by numerating the engine. This model is a representation of vehicle information such as the type of vehicle (truck, passenger car, etc.), weight, transmission type (MT, AT, CVT, etc.), tire diameter, gear ratio, engine characteristics (throttle opening and the relationship between speed and output torque, etc.), ECU control characteristics (the relationship between accelerator opening and throttle opening, etc.), TCU control characteristics (conditions for changing gear ratios and their timing, etc.), or BCU control characteristics (the distribution of braking force to each wheel, etc.). These models are pre-stored in the relational data storage unit 201.

[0058] The simulation unit 202 simulates the emission test results of the MAW (Moving Averaging Window) method or the Power Bining method within the specified range, based on the various models stored in the relational data storage unit 201, for example, a road driving test according to the RDE test.

[0059] Furthermore, the simulation unit 202 simulates the emission amounts of each exhaust gas component contained in the exhaust gas emitted during the aforementioned driving conditions, based on the learned completion model (exhaust gas model) stored in the relational data storage unit 21. The exhaust gas model can also be stored in the simulation unit 202. Additionally, Figure 4 The "Emission Prediction" section presents an example of the simulation results for emission amounts. Here, when simulating the emission amounts of each exhaust gas component in the simulation unit 202, the emission amounts of each exhaust gas component can be simulated simultaneously using each model (including the exhaust gas model) stored in the relational data storage unit 201. Alternatively, the emission amounts of each exhaust gas component can be simulated using each model (except the exhaust gas model) stored in the relational data storage unit 201 to simulate road driving tests, simulating the emission amounts of each exhaust gas component based on various parameters obtained through this simulation and the exhaust gas model. For example... Figure 6 As shown, for example, by using the simulation device 200, various actual driving scenarios can be simulated, thereby estimating the CO2 emissions during actual driving and determining the use case and the worst case.

[0060] <Effects of this implementation method> According to the vehicle component response learning system 100 of this embodiment, inputs such as vehicle speed, load, temperature, and air pressure (simulating actual road driving) are provided to the actual engine E to obtain exhaust gas data. Based on training data consisting of input data representing the inputs provided to the actual engine E and the exhaust gas data, machine learning is used to generate a learned model related to the exhaust gas of the actual engine E. As a result, by using this learned model, response data of vehicle components under desired driving conditions can be obtained through simulation without actual road driving. Furthermore, since control parameters such as ECU parameters for engine combustion are not changed, a response model can be easily and safely generated using a pre-suited ECU diagram.

[0061] <Other variations and implementations> Furthermore, the present invention is not limited to the embodiments described herein.

[0062] For example, the vehicle component in the described embodiment is an actual engine, but it can be any part of an actual vehicle that includes an actual engine. When generating a learning-completed model related to the response of an actual vehicle, a chassis dynamometer, a vehicle drive system testing device, etc., can be considered.

[0063] Alternatively, in the input step of the above embodiment, the input obtained by dynamically varying the temperature and air pressure at the boundaries between the segmented blocks can also be provided to the vehicle components.

[0064] The response data in the described embodiment is exhaust gas data, but it can also be temperature data of the exhaust gas purification catalyst, exhaust gas temperature data, flow rate data, or pressure data.

[0065] For example, such as Figure 7 As shown, the vehicle component response learning system 100 uses exhaust gas data as response data to generate an engine-out exhaust gas model. Additionally, it uses exhaust gas temperature data or exhaust gas purification catalyst temperature data as response data to generate an exhaust gas temperature model or catalyst temperature model. Furthermore, the simulation device 200 uses these engine-out exhaust gas models, exhaust gas temperature models, catalyst temperature models, and exhaust aftertreatment conversion efficiency models to predict exhaust gas data emitted from the tailpipe. The conversion efficiency model is, for example, a model that models the conversion efficiency based on the temperature of the exhaust gas purification catalyst.

[0066] For example, when creating an exhaust gas model, the vehicle component response learning system 100 provides input data containing parameters related to vehicle speed, load, and / or temperature to the vehicle component and outputs exhaust gas data (e.g., exhaust gas concentrations, exhaust gas flow rates, and / or exhaust gas emissions of NOx, CO, CO2, THC, and / or NH3) as response data. Then, machine learning is performed on the input data and the response data to obtain the completed learning data for the exhaust gas model (e.g., a model of NOx, CO, CO2, THC, and / or NH3).

[0067] Furthermore, when creating the exhaust gas temperature model, the vehicle component response learning system 100 provides input data, including parameters related to vehicle speed, load, and temperature, and / or exhaust gas data, to the vehicle component, obtaining exhaust gas temperature and / or exhaust catalyst temperature as response data. Then, machine learning is performed on the input data and response data to create a completed learning model of the exhaust gas temperature model. Similarly, when creating the exhaust aftertreatment conversion efficiency model, the vehicle component response learning system 100 provides input data, including exhaust gas data and parameters related to exhaust gas temperature and / or exhaust catalyst temperature, to the vehicle component, obtaining exhaust gas data as response data. Then, machine learning is performed on the input data and response data to create a completed learning model of the exhaust aftertreatment conversion model.

[0068] Furthermore, the vehicle component in the described embodiment can also be a hybrid vehicle or a part thereof, in which the engine and battery work together. Additionally, the present invention is not limited to engine vehicles or hybrid vehicles, but can also be applied to electric vehicles (EVs) or fuel cell vehicles (FCVs).

[0069] When the vehicle components include the actual battery (secondary battery) installed in the vehicle, the response data obtained by the acquisition unit 3 becomes power consumption data, current data, voltage data, SOC (State of Charge) data, or battery temperature data. Here, power consumption data or SOC data can be measured using a power consumption meter. Current data or voltage data can be measured using an ammeter or a voltmeter, respectively. Battery temperature data is the temperature data of a single cell constituting the actual battery or the temperature of a module or battery pack composed of multiple cells, and can be measured using a thermometer. Furthermore, SOH (State of Health) data can also be used as response data for the actual battery. Additionally, these data can be values ​​that can be obtained using systems already installed in the vehicle (ECU (Electric Control System)), BMS (Battery Management System) and / or VU (Vehicle Control System)).

[0070] Figure 8 (a) shows an experimental system for generating a learned completion model that correlates with the response of an actual battery (secondary battery). Additionally, Figure 8 (b) shows a flowchart of the battery model generation process. Furthermore, Figure 8 (c) shows a schematic diagram of the simulation using a battery model.

[0071] like Figure 8 As shown in (a), the test system for generating the battery model includes, for example, a charging / discharging device 22f as an input device 22, which charges or discharges an actual battery (actual battery pack, actual battery module, or actual battery cell) housed in the test chamber 22d1 of the environmental variation device 22d. Furthermore, the test system generates the battery model by charging or discharging the actual battery through this charging / discharging device 22f.

[0072] Specifically, such as Figure 8 As shown in (b), the input data generation unit 21 of the input unit 2 is the same as in the embodiment described above, and uses the Dynamic Experimental Planning (DoE) method to determine the combination of each parameter (“Dynamic Experimental Planning”). Here, each parameter is a parameter related to vehicle speed, load, and temperature during hypothetical actual road driving. The parameter related to vehicle speed and load is, for example, the charging current or discharging current, and the parameter related to temperature is, for example, the ambient temperature of the battery.

[0073] Furthermore, the input, consisting of the determined combination of parameters, is provided to the actual battery (actual battery pack, actual battery module, or actual battery cell) via the charging / discharging device 22f and the environmental variation device 22d. Here, the training data acquisition unit 32 of the acquisition unit 3 acquires the input data representing the input provided to the actual battery and the response data of the actual battery as training data (learning dataset) ("dynamic test implementation"). Here, the response data of the actual battery may include, for example, SOC data, current data, voltage data, cell temperature data, and / or module temperature data.

[0074] Next, the generation unit 4 uses the training data obtained through the aforementioned "dynamic test implementation" to generate a battery model ("modeling") of the actual battery. The battery model thus obtained is used by the simulation device 200 to simulate road driving tests.

[0075] like Figure 8As shown in (c), the simulation device 200 uses a driving mode model (driver model) modeled from driving modes, a vehicle model modeled from actual vehicles, a PCU model modeled from the power control unit (PCU) that controls the charging and discharging of actual batteries according to vehicle speed and load, and / or a generated learning-completed model (battery model) to simulate road driving tests. Through this simulation device 200, various real-world driving scenarios are simulated, thereby... Figure 9 As shown, it is possible to estimate energy consumption or changes in battery state such as SOC during or after actual driving.

[0076] In the case where the vehicle component includes a fuel cell mounted in the vehicle, the response data acquired by the acquisition unit 3 becomes hydrogen consumption data, oxygen consumption data, power generation current data, power generation voltage data, and / or battery temperature data. Here, hydrogen consumption data can be measured using a hydrogen meter, and oxygen consumption data can be measured using an oxygen meter. Power generation current data or power generation voltage data can be measured using an ammeter or a voltmeter, respectively. Battery temperature data includes the temperature data of the fuel cell, the temperature data of a single fuel cell, or the temperature data of a fuel cell stack composed of multiple cells, and can be measured using a thermometer.

[0077] Figure 10 (a) shows an experimental system for generating a learned completion model related to the response of a fuel cell. Additionally, Figure 10 A flowchart illustrating the generation of the fuel cell model is shown in (b). Furthermore, Figure 10 (c) shows a schematic diagram of the simulation using a fuel cell model.

[0078] like Figure 10 As shown in (a), the test system for generating a fuel cell model includes, for example, a hydrogen supply device 22g for supplying hydrogen to a fuel cell housed in a test chamber 22d1 containing an environmental change device 22d; an oxygen supply device 22h for supplying oxygen (atmosphere); and a water supply device 22i for supplying water, as input devices 22. Hydrogen, oxygen, and water are supplied to the fuel cell through these hydrogen supply devices 22g, oxygen supply devices 22h, and water supply devices 22i, thereby generating a fuel cell model.

[0079] Specifically, such as Figure 10As shown in (b), the input data generation unit 21 of the input unit 2 is the same as in the embodiment described above, and uses the Dynamic Experimental Planning (DoE) method to determine the combination of parameters (“Dynamic Experimental Planning”). Here, each parameter is a parameter related to the hypothetical actual road driving speed, load, and temperature. The parameters related to the vehicle speed and load are, for example, the hydrogen supply, oxygen supply, and / or water supply, and the parameters related to the temperature are, for example, the ambient temperature of the fuel cell.

[0080] Then, the input, consisting of the determined combination of parameters, is supplied to the fuel cell via hydrogen supply device 22g, oxygen supply device 22h, water supply device 22i, and environmental change device 22d. Here, the training data acquisition unit 32 of the acquisition unit 3 acquires the input data representing the input supplied to the fuel cell and the fuel cell's response data as training data (learning dataset) ("dynamic test implementation"). Here, the fuel cell's response data may include, for example, power generation current data, power generation voltage data, or cell temperature data.

[0081] Next, the generation unit 4 uses the training data obtained through the aforementioned "dynamic test implementation" to generate a fuel cell model ("modeling"). The resulting fuel cell model is then used by the simulation device 200 to simulate road driving tests.

[0082] like Figure 10 As shown in (c), the simulation device 200 uses a driving mode model (driver model) modeled with the driving mode, a vehicle model modeled with the actual vehicle, a PCU model modeled with the power control unit (PCU) that controls the power generation of the fuel cell according to vehicle speed and load, and a generated learning-completed model (battery model) to simulate a road driving test. Furthermore, in Figure 10 In (c), an example is shown where a battery model is used in addition to the fuel cell model for simulation, but it is also possible not to use a battery model.

[0083] In the case of vehicle components including a motor mounted on the vehicle, the motor speed is input to the motor to simulate the actual vehicle speed on the road.

[0084] Furthermore, in the described embodiment, a combination of four parameters—vehicle speed, load, temperature, and air pressure—is used as the input to the vehicle component. However, a combination of three parameters—vehicle speed, load, and temperature—or a combination of three parameters—vehicle speed, load, and pressure—can also be used as the input to the vehicle component. Additionally, humidity, which can be varied by an environmental change device, can also be included as a parameter input to the vehicle component.

[0085] The multiple functions implemented by the control device 5 in the above embodiment can also be shared by multiple computers that are physically separated. For example, the input data production unit 21, the training data acquisition unit 32, and the generation unit 4 can be shared by one computer, while other functions can be shared by another computer.

[0086] Road driving tests that comply with regulations are not limited to RDE; they can also be various road driving tests as required by the laws or regulations of each country.

[0087] Furthermore, various modifications and combinations of embodiments are possible as long as they do not violate the spirit of this invention. Industrial applicability

[0088] According to the present invention, it is possible to obtain the response data of vehicle components under the desired driving environment with high precision through simulation without actual road driving. Explanation of reference numerals in the attached figures:

[0089] 100: Vehicle component response learning system; E: Actual engine (vehicle component); 22d: Environmental change device; 22d1: Laboratory; 31: Exhaust gas analysis device; 2: Input unit; 21: Input data production unit; 3: Acquisition unit; 4: Generation unit; 5: Control device (vehicle component response learning device).

Claims

1. A vehicle component response learning method, generating a learned model related to the response of a vehicle component that is a vehicle or part of the vehicle, characterized in that, Includes the following steps: The input step involves providing the vehicle component with inputs containing parameters related to the vehicle speed, load, temperature, and air pressure under the assumption of actual road driving. The acquisition step involves acquiring the response data of the vehicle component, and acquiring the input data representing the input and the response data as training data. as well as The generation step involves using machine learning, based on the training data, to generate a learned model related to the response of the vehicle components. In the case that the vehicle component is a vehicle that includes at least an engine or a part of such a vehicle, the response data is exhaust gas data. In the case that the vehicle component is a vehicle or part of a vehicle that includes at least a secondary battery, the response data includes power consumption data, current data, voltage data, SOC data, and / or battery temperature data. In the case that the vehicle component is a vehicle or part thereof that includes at least a fuel cell, the response data is hydrogen consumption data, oxygen consumption data, power generation current data, power generation voltage data, or battery temperature data.

2. The vehicle component response learning method according to claim 1, characterized in that, In the input step, the parameters are varied within their respective ranges of vehicle speed, load, temperature, and air pressure, and the combined input is provided to the vehicle component.

3. The vehicle component response learning method according to claim 2, characterized in that, In the input step, the temperature and air pressure variation ranges are divided into multiple segments, and the segmented input obtained by varying the temperature and air pressure within these segments is provided to the vehicle component. The acquisition step acquires the response data of the vehicle component provided with the segmentation input, and acquires the segmentation input data representing the segmentation input and the response data as segmentation training data.

4. The vehicle component response learning method according to claim 3, characterized in that, The input step provides the vehicle component with inputs obtained from the temperature and air pressure variations at the boundaries between the segmented blocks.

5. The vehicle component response learning method according to any one of claims 1 to 4, characterized in that, The temperature and pressure are varied by means of an environmental variation device that has a test chamber for housing the vehicle component and varies the temperature and pressure of the test chamber, or an environmental variation device that is connected to the vehicle component and varies the temperature and pressure.

6. The vehicle component response learning method according to any one of claims 1 to 4, characterized in that, The input step uses the Design of Experiments (DoE) method to generate the combination of parameters.

7. The vehicle component response learning method according to claim 1, characterized in that, The exhaust gas data were measured using an exhaust gas analysis device. The power consumption data, the current data, the voltage data, the SOC data, or the temperature data are measured using a power consumption meter, ammeter, voltmeter, or thermometer. The hydrogen consumption data, oxygen consumption data, power generation current data, power generation voltage data, or battery temperature data are measured using a hydrogen meter, oxygen meter, ammeter, voltmeter, or thermometer.

8. The vehicle component response learning method according to any one of claims 1 to 4, characterized in that, In the case that the vehicle component is a vehicle or part of a vehicle that includes at least a secondary battery, the parameter associated with the vehicle speed or the load is the charging current and / or discharging current, and the parameter associated with the temperature is the ambient temperature of the secondary battery. In the case that the vehicle component is a vehicle or part thereof that includes at least a fuel cell, the parameters associated with the vehicle speed or the load are the hydrogen supply, oxygen supply and / or water supply, and the parameter associated with the temperature is the ambient temperature of the fuel cell.

9. A method for calculating the response of a vehicle component, characterized in that, The response data of the vehicle components during actual road driving is calculated using the learning completion model generated by the vehicle component response learning method according to any one of claims 1 to 8.

10. A vehicle component response learning system, generating a completed learning model related to the response of a vehicle component that is a vehicle or part of the vehicle, characterized in that, have: The input unit provides the vehicle components with inputs containing parameters related to the vehicle speed, load, temperature, and air pressure as if driving on a hypothetical actual road. The acquisition unit acquires response data from the vehicle component, and acquires input data representing the input and the response data as training data; as well as The generation unit, based on the training data, uses machine learning to generate a learned model related to the response of the vehicle component. In the case that the vehicle component is a vehicle that includes at least an engine or a part of such a vehicle, the response data is exhaust gas data. In the case that the vehicle component is a vehicle or part of a vehicle that includes at least a secondary battery, the response data includes power consumption data, current data, voltage data, SOC data, and / or battery temperature data. In the case that the vehicle component is a vehicle or part thereof that includes at least a fuel cell, the response data is hydrogen consumption data, oxygen consumption data, power generation current data, power generation voltage data, or battery temperature data.

11. The vehicle component response learning system according to claim 10, characterized in that, The input section includes: The dynamometer provides a load to the vehicle components; and An environmental variation device having a test chamber for housing the vehicle component and at least varying the temperature of the test chamber, or an environmental variation device connected to the vehicle component and at least varying the temperature.

12. A storage medium storing a vehicle component response learning program, characterized in that, The vehicle component response learning program generates a completed learning model relating to the responses of vehicle components that are part of or related to the vehicle. The vehicle component response generation program enables the computer to function as an input data creation unit, acquisition unit, and generation unit. The input data generation unit generates input data, which represents the input to be provided to the vehicle components, including parameters related to hypothetical actual road driving speed, load, temperature, and air pressure. The acquisition unit acquires the response data of the vehicle component, and uses the input data and the response data as training data. The generation unit uses statistical methods or machine learning to generate a learned model related to the response of the vehicle component based on the training data. In the case that the vehicle component is a vehicle that includes at least an engine or a part of such a vehicle, the response data is exhaust gas data. In the case that the vehicle component is a vehicle or part of a vehicle that includes at least a secondary battery, the response data includes power consumption data, current data, voltage data, SOC data, and / or battery temperature data. In the case that the vehicle component is a vehicle or part thereof that includes at least a fuel cell, the response data is hydrogen consumption data, oxygen consumption data, power generation current data, power generation voltage data, or battery temperature data.

Images