Vehicle Failure Diagnosis Using Normal Trend Heatmap Models
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for diagnosing vehicle failures rely on supervised learning, which requires labeled data and is not effective for detecting new or unknown failures.
Innovation Solution
The method employs unsupervised learning to create a normal trend figure model using only data representing normal vehicle states, allowing for the diagnosis of failures by analyzing deviations from this model.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If supervised learning is used for failure diagnosis, then classification accuracy for known failures is improved, but the system cannot detect new or unknown failures and requires large amounts of labeled data
Solution Approach 1:
Instead of training the model to recognize failure patterns (supervised learning), the patent inverts the approach by training the model to recognize only normal operation patterns. Any deviation from the learned normal pattern is automatically flagged as a potential failure, enabling detection of both known and unknown failure types without requiring labeled failure data.
Solution Approach 2:
The system performs self-diagnosis by comparing actual sensor data against the self-learned normal operation model. The model automatically identifies anomalies without external intervention or pre-programmed failure signatures, enabling autonomous detection of diverse failure modes including previously unseen failure types.
2Measurement precision
If complex diagnostic models are used to improve detection accuracy, then failure detection capability is improved, but memory requirements and computational resources increase
Solution Approach 1:
The patent extracts only the essential characteristics of normal operation into a simplified model representation. By focusing solely on learning normal patterns rather than cataloging all possible failure modes, the system achieves high detection accuracy with minimal memory requirements, storing only the baseline normal operation signature for comparison.
Solution Approach 2:
The system transforms the diagnostic approach from storing multiple failure signatures to storing a single normal operation model. This parameter change in the data representation strategy dramatically reduces memory footprint while maintaining or improving detection capability through anomaly comparison against the normal baseline.
3Measurement precision
If additional hardware is added to improve diagnostic capability, then detection accuracy is improved, but manufacturing cost and system complexity increase
Solution Approach 1:
The patent replaces physical diagnostic hardware or complex sensor arrays with a software-based unsupervised learning model that processes existing sensor data. This substitution achieves enhanced diagnostic capability through algorithmic analysis rather than additional physical components, simplifying manufacturing and enabling mass production deployment.
Solution Approach 2:
The unsupervised learning model serves multiple diagnostic functions simultaneously - detecting various failure types, identifying anomaly patterns, and adapting to different operating conditions - all using the same core algorithm and existing sensors. This multi-functionality eliminates the need for specialized hardware for each diagnostic task, reducing overall system complexity and manufacturing costs.
Data Source
AI summary
A method for diagnosing a failure may include collecting sensor signals related to a vehicle in a form of time series data; obtaining a feature classified into a predetermined number of frequency indices by performing a Fourier transform on a predetermined time interval among the time series data; scaling the feature to a value of a predetermined range; generating a graph showing the value depending on a frequency index; generating a heatmap from the graph; training a normal trend figure model that represents trend of data indicating a normal state in the heatmap as a figure; and determining whether there is a failure by calculating a loss for a sensor data input based on the normal trend figure model.