Vehicle evaluation method
The integration of LTI and LSTM models within a ROM system simplifies vehicle evaluation, reducing time and expertise requirements while maintaining accuracy.
Patent Information
- Application Number
- JP2023074240
- Authority / Receiving Office
- JP · JP
- Patent Type
- Applications
- Current Assignee / Owner
- Filing Date
- 2023-04-28
- Publication Date
- 2024-11-08
AI Technical Summary
Existing vehicle evaluation methods require specialized knowledge and time-consuming data collection and calculation processes, leading to inefficiencies in prediction and model creation.
A method involving the creation of an LTI model based on time-series learning data, followed by an LSTM model, and combining these with a ROM system to simplify the process and reduce calculation time while maintaining accuracy.
Reduces prediction and calculation time while maintaining model accuracy, enabling efficient vehicle evaluation even for those without specialized knowledge.
Smart Images

Figure 2024158745000001_ABST
Abstract
Description
[Technical field]
[0001] The present disclosure relates to a method for evaluating a vehicle. [Background technology]
[0002] Patent Document 1 describes a method for reducing the amount of calculation while maintaining the accuracy of the simulation of a vehicle model. Specifically, it describes a method for selecting and reducing unit models and calculation items to be reduced from a vehicle model composed of multiple unit models according to the vehicle performance to be tested. [Prior art documents] [Patent documents]
[0003] [Patent Document 1] JP 2016-042280 A Summary of the Invention [Problem to be solved by the invention]
[0004] In the invention described in Patent Document 1, specialized knowledge and experience are required to select a degeneration target from a plurality of vehicle models according to the tested vehicle performance. Therefore, it takes time before the model can be utilized. Furthermore, in order to improve the prediction accuracy of the model, it is necessary to use a plurality of types of input data and a plurality of output data as learning data, and data of intermediate characteristics is also required. Therefore, the amount of measurements and calculations required to collect learning data increases, and the time required to create the model increases.
[0005] An object of the present disclosure is to reduce prediction and calculation time while maintaining the accuracy of a model for evaluating a vehicle, and to create a model by a simple method. [Means for solving the problem]
[0006] One aspect of the present disclosure is a vehicle evaluation method including the steps of acquiring time-series learning data for evaluating a vehicle, creating an LTI model based on the time-series learning data, creating an LSTM model based on the time-series learning data and prediction data output from the LTI model, and creating a ROM for obtaining output data from the time-series learning data by combining the LTI model and the LSTM model. Effect of the Invention
[0007] According to the present disclosure, it is possible to reduce prediction and calculation time while maintaining the accuracy of a model for evaluating a vehicle, and also to create a model using a simple method. [Brief description of the drawings]
[0008] [Figure 1] 1 is a block diagram showing a configuration of a ROM system according to an embodiment. [Diagram 2] 1 is a flowchart showing a process for creating a ROM system according to an embodiment. [Diagram 3] 4 is a flowchart showing a process flow using the ROM system according to the embodiment. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS
[0009] 1 shows a reduced order model (ROM) system 10 according to an embodiment. The ROM system 10 includes a linear time invariant system (LTI) model 12 and a long short term memory (LSTM) model 14.
[0010] The LTI model 12 is a model created using an LTI system, which is a control field technology. For example, the LTI model 12 is created by estimating a transfer function G(s) from time-series learning data.
[0011] The LSTM model 14 is a model created using an LSTM system, which is a deep learning method, and is a model used in a neural network (for example, a Recurrent Neural Network (RNN)). For example, the LSTM model 14 is generated by learning using time-series learning data and prediction data output from an LTI model.
[0012] In the ROM system 10, input data is input to the LTI model 12 and the LSTM model 14, and data output from the LTI model 12 is input to the LSTM model 14. Output data is obtained by the LSTM model 14.
[0013] A processing device such as a computer or a server is used to create the ROM system 10. For example, the processing device includes a processor and a storage device. The processor includes a central processing unit (CPU), a graphics processing unit (GPU), an application specific integrated circuit (ASIC), a field programmable gate array (FPGA), a digital signal processing (DSP), other programmable logic devices or electronic circuits, etc. The storage device includes a hard disk drive or various types of memory, etc.
[0014] The process of creating the ROM system 10 will be described below with reference to Fig. 2. Fig. 2 shows a flowchart showing the flow of the process. Each process shown in Fig. 2 is executed by, for example, a processor of a processing device. A storage device may be used to execute the process.
[0015] First, the time series training data X t is prepared (S01). For example, the training data X t is time series data representing engine torque fluctuations.
[0016] Next, the processor uses the 1D vibration analysis model to generate time-series learning data X t to the time series training data Y t (S02). For example, the training data Y t is time-series data representing the vehicle longitudinal G (G (acceleration) in the longitudinal direction of the vehicle).
[0017] Next, the processor performs training using the LTI system and generates the time series training data X t ,Y t For example, the processor creates an LTI model 12 based on the time series training data X t ,Y t Based on this, the input data, training data X t The training data Y, which is the output data from t Estimate the transfer function G(s) that creates the LTI model 12.
[0018] Next, the processor processes the time-series training data X t is input to LTI Model 12, and the time series forecast data Y is obtained from LTI Model 12. t LTI (S04) The predicted data Y t LTI is time series data representing the estimated vehicle longitudinal G.
[0019] Next, the processor uses the LSTM system to train the time series data X t ,Y t LTI The processor creates the ROM system 10 by combining the LTI model 12 and the LSTM model 14 (S05).
[0020] By using the LSTM system, it is possible to reduce the model without specialized knowledge or experience. In addition, by creating the LTI model 12 using the LTI system as a preprocessing for creating the LSTM model 14, the accuracy of the model can be maintained. In addition, since only one type of data is required for the input data and the output data, a simple ROM system 10 can be created. Since multiple types of learning data are not required, the time required to create the model can be reduced. In addition, the ROM system 10 can be created by a simple method.
[0021] Hereinafter, the process using the ROM system 10 will be described with reference to Fig. 3. Fig. 3 shows a flowchart showing the flow of the process. For example, the ROM system 10 is incorporated in a processing device such as a computer or a server, and the process by the ROM system 10 is executed. As an example, it is assumed here that the exhaust purification performance and NV (Noise Vibration) performance are evaluated by an engine bench.
[0022] First, when evaluating the exhaust purification performance using an engine bench, data X that shows the fluctuation of engine torque in time series is test is measured (S10).
[0023] Time series data X test is input to the ROM system 10, which converts the time series data X test Output data Y from time series pred (S11) The output data Y pred is prediction data that represents the vehicle longitudinal G in time series.
[0024] Time series output data Y pred is used to evaluate the NV performance, and the exhaust purification performance is evaluated using data actually measured on an engine bench (S12). For example, the NV performance and the exhaust purification performance are evaluated simultaneously in real time.
[0025] For example, the catalyst bed temperature data for evaluating the exhaust purification performance and the output data Ypred (data on vehicle longitudinal G) are displayed side by side on the display of the engine bench. Specifically, the catalyst bed temperature is actually measured by the engine bench, and the catalyst bed temperature data is displayed on the display. In addition, the engine cylinder pressure sensor measures the pressure inside each cylinder, and the combustion analysis device calculates the engine torque fluctuation. This calculation produces time series data X test is generated. Time series data X test is input to the ROM system 10, which generates output data Y pred is output, and the output data Y pred is displayed on the display. This makes it possible to simultaneously evaluate NV performance and exhaust purification performance in real time.
[0026] As described above, according to the present embodiment, it is possible to reduce prediction and calculation time while maintaining the accuracy of the model for evaluating a vehicle. In addition, since a model can be created by a simple method, even a compatibility evaluation person who has little experience in model development and vehicle dynamics analysis at an actual development site can create a model and use the model to evaluate a vehicle. In addition, in the past, since the degeneration target was selected by a person, the reliability, accuracy, etc. of the model may vary from person to person. According to the present embodiment, since it is not necessary for a person to select the degeneration target, the reliability, accuracy, etc. of the model do not vary from person to person. [Explanation of symbols]
[0027] 10 ROM, 12 LTI models, and 14 LSTM models.
Claims
[Claim 1] Obtaining time-series learning data for evaluating a vehicle; creating an LTI model based on the time-series learning data; creating an LSTM model based on the time-series learning data and prediction data output from the LTI model; creating a ROM for obtaining output data from the time-series learning data by combining the LTI model and the LSTM model; A vehicle evaluation method including:
Citation Information
Patent Citations
Vehicle simulation system
JP2016042280A
