Vehicle Component Abnormality Mapping Using Simulated Fault Data
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing abnormality determination systems for vehicle components, such as automatic transmissions, face challenges in accurately setting thresholds for abnormality detection due to the infrequent occurrence of abnormalities, leading to inappropriate threshold values.
Innovation Solution
A training method using a simulator and storage to update mapping data based on input variables and teaching data, simulating various conditions to generate characteristic variables that indicate the possibility of component abnormalities, thereby improving the accuracy of abnormality detection.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If the upper limit value of vibration acceleration is set based on previous vibration data, then the abnormality determination apparatus can detect abnormalities, but the determination accuracy deteriorates because abnormalities occur infrequently and the upper limit value becomes inappropriate
Solution Approach 1:
The patent applies preliminary action by using a simulator to pre-generate vibration data under abnormal conditions before actual deployment. The simulator creates virtual training data including abnormal vibration patterns, allowing the determination apparatus to be pre-trained with comprehensive datasets that include rare abnormal scenarios, thus improving detection accuracy without requiring extensive real-world abnormal data
Solution Approach 2:
The patent uses copying by creating virtual copies of abnormal vibration data through simulation. Instead of relying solely on scarce real abnormal data, the system generates synthetic training data that replicates abnormal vibration characteristics, enabling the determination apparatus to learn from these copied patterns and improve its ability to detect actual abnormalities
2Measurement precision
If a large number of training data are collected to improve mapping accuracy, then the output variable appropriateness increases, but the data collection time and cost increase significantly
Solution Approach 1:
The patent applies copying by generating synthetic training data through simulation rather than collecting extensive real-world data. The simulator creates virtual copies of various operating conditions and abnormal scenarios, providing a large dataset for training without the time and cost constraints of physical data collection
Solution Approach 2:
The patent replaces the mechanical data collection process with a computational simulation system. Instead of physically collecting vibration data from actual vehicles over extended periods, the system uses computer-based simulation to generate training data, substituting physical measurement with computational modeling
Data Source
AI summary
A training method includes: a simulation step outputting, by a simulator, a characteristic variable based on an input parameter set, the parameter set being input to the simulator and indicating that a specific component is presumed to have an abnormality in advance; and a training step updating, by a training device, mapping based on an input training data and an input teaching data, the input variables being input to the training device as the training data, the input variables including the characteristic variable output in the simulation step, the abnormality determination variable being input to the training device as the teaching data, the abnormality determination variable indicating that the specific component has the abnormality.