Fault Diagnosis Classification Using Partial System Models
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current fault diagnosis algorithms either require a full system model, which is not always available, or disregard system information, leading to complex models and inefficient classification processes.
Innovation Solution
A classification-based diagnosis approach that uses partial system model information, such as system topology and component behavior, to simplify the classifier complexity by learning parameters of unknown components using a Bayesian approach and state estimation, reducing the number of parameters needed and speeding up the training process.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If machine-learning based approaches are used to perform fault diagnosis, then the system can operate without a full system model, but the classifier complexity increases significantly and training time is extended
Solution Approach 1:
The patent segments the system model into known components and unknown components. The classifier is designed to work with this segmented structure, only needing to learn parameters for the unknown components while utilizing the known component models directly. This segmentation reduces the overall classifier complexity while maintaining the ability to operate without a complete system model.
Solution Approach 2:
The patent applies partial action by using only the necessary portion of the system model (known components) rather than requiring the complete model. The classifier learns only the missing parameters for unknown components, which is a partial learning approach that reduces complexity while still achieving effective fault diagnosis.
2Adaptability or versatility
If machine-learning based approaches are used to perform fault diagnosis, then the system can operate without a full system model, but the training time is significantly extended
Solution Approach 1:
By segmenting the system into known and unknown components, the patent reduces the scope of learning required during training. Only the unknown components need parameter learning, which significantly reduces training time compared to learning all system parameters from scratch, while still maintaining adaptability to systems with incomplete models.
Solution Approach 2:
The patent performs preliminary action by incorporating known system model information before the training process begins. This pre-existing knowledge about known components eliminates the need to train those portions of the system, reducing the overall training time while maintaining the ability to handle systems where not all models are available.
3Device complexity
If model-based approaches are used to perform fault diagnosis, then the classifier complexity is reduced, but a full system model is required which is not always available
Solution Approach 1:
The patent uses partial action by implementing a model-based approach that only requires partial system model information (known components) rather than a complete model. The classifier is designed to work with this partial information, maintaining low complexity while achieving adaptability to operate with incomplete model availability.
Solution Approach 2:
The patent introduces an intermediary structure that bridges the gap between partial model information and complete fault diagnosis capability. The classifier acts as an intermediary that combines the known component models with learned parameters for unknown components, enabling the system to function adaptively without requiring a full system model while maintaining reasonable complexity.
Data Source
AI summary
A classification-based diagnosis for detecting and predicting faults in physical system (e.g. an electronic circuit or rail switch) is disclosed. Some embodiments make use of partial system model information (e.g., system topology, components behavior) to simplify the classifier complexity (e.g., reduce the number of parameters). Some embodiments of the method use a Bayesian approach to derive a classifier structure.