Vehicle Response Learning Model for Real-World Driving Simulation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for generating transient exhaust gas models do not adequately consider real-road driving environments and require limited access to engine control unit (ECU) parameters, leading to inefficiencies and risks in data collection and prediction accuracy.
Innovation Solution
A vehicle element response learning method that uses machine learning to generate a trained model based on input parameters like vehicle speed, load, temperature, and atmospheric pressure, simulating actual road driving conditions without altering ECU parameters, thereby reducing the need for repetitive tests and improving prediction accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If transient exhaust gas model is generated using Dynamic DoE with ECU parameters, then prediction capability is improved, but access is limited to only a few engineers and repetitive confirmation tests are required
Solution Approach 1:
The patent creates a virtual copy of the engine system through a trained model that replicates the behavior of the actual engine. This virtual model allows any engineer to perform simulations and analyses without needing direct access to the physical engine's ECU parameters, thereby democratizing access while maintaining prediction accuracy.
Solution Approach 2:
The trained model serves as an intermediary between the complex ECU parameters and the end-user applications. Instead of directly accessing and manipulating ECU parameters, engineers interact with the trained model which has already encapsulated the relationships between parameters and outputs, simplifying the interface and reducing the need for specialized access.
2Measurement precision
If ECU parameters are changed for generating transient exhaust gas model, then model accuracy is improved, but engine damage risk increases
Solution Approach 1:
The patent performs all necessary parameter adjustments and model training in advance, before actual engine testing. The trained model is created offline using historical data and simulations, allowing the system to predict outcomes without requiring real-time parameter changes that could risk engine damage during actual operation.
Solution Approach 2:
The system uses a virtual copy of the engine (trained model) to test parameter changes and predict outcomes, rather than directly modifying the actual engine parameters. This allows extensive what-if analysis and optimization without exposing the physical engine to potential damage from incorrect parameter settings.
3Loss of information
If repetitive confirmation tests are performed to grasp parameter effects, then understanding of parameter impact is improved, but time consumption and data analysis burden increase
Solution Approach 1:
The patent replaces the mechanical process of repetitive physical testing with a computational approach using the trained model. Instead of physically adjusting parameters and measuring results through repeated experiments, the system uses the trained model to simulate and analyze parameter impacts computationally, dramatically reducing time and resource requirements.
Solution Approach 2:
The system performs comprehensive parameter analysis in advance during the model training phase. By pre-computing the relationships between parameters and outputs, the trained model enables rapid querying and analysis without requiring time-consuming repetitive tests, as all necessary information is already captured in the trained relationships.
4Device complexity
If transient exhaust gas model is generated without considering real-road driving environment, then model generation is simplified, but prediction accuracy in actual road driving deteriorates
Solution Approach 1:
The patent incorporates dynamic real-road driving conditions into the model training process. The trained model is trained on data that includes varying operating conditions such as different speeds, loads, and environmental factors, enabling it to adapt to and accurately predict performance across diverse real-world scenarios rather than static test conditions.
Data Source
AI summary
The present invention is to accurately obtain a vehicle element response data under a desired driving environment by simulation without performing actual road driving and is a vehicle element response learning method for generating a trained model related to a response of a vehicle element that is a vehicle or a part of the vehicle, and the method includes: (1) an input step of giving an input including parameters related to a vehicle speed, a load, and a temperature assuming actual road driving, to the vehicle element; (2) an acquisition step of acquiring response data of the vehicle element and acquiring, as training data, input data representing the input and the response data; and (3) a generation step of generating the trained model related to the response of the vehicle element, from the training data by using machine learning.