Data-Driven ARR Generation for Fault Isolation With Limited Data
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing fault detection and isolation methods face challenges in complex systems where developing accurate system models is expensive and infeasible, especially when systems are not fully observable and lack vast historical fault data, limiting their ability to detect and isolate faults effectively.
Innovation Solution
The proposed solution involves generating data-driven analytical redundancy relationships (ARRs) using historical data during normal operation, which do not rely on system models, allowing for fault detection and isolation even when the system is not fully observable and without extensive fault data. This is achieved through methods like exhaustive search and forward feature selection to identify minimal sets of variables that can predict target variables, using machine learning models such as linear regression or recurrent neural networks to capture delays, and setting normal thresholds for alarm generation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If model-based fault detection and isolation methods are used, then fault detection capability is improved, but system complexity and cost increase due to the need for accurate system models
Solution Approach 1:
The patent creates a virtual copy of the system's normal behavior through data-driven ARRs learned from historical normal operation data. Instead of requiring complex physical models, the system learns redundancy relationships that replicate normal system behavior, which can then be compared against actual behavior to detect faults. This copying approach maintains fault detection capability while eliminating the need for complex system models.
Solution Approach 2:
The patent replaces the traditional mechanical/model-based approach with a data-driven machine learning approach. Instead of using physical system models and mathematical equations to represent system behavior, the system uses historical data and machine learning algorithms to learn redundancy relationships, substituting the mechanical modeling process with statistical learning.
2Ease of manufacture
If data-driven solutions are used for fault detection, then system model requirements are reduced, but performance deteriorates when historical fault data is limited or unavailable
Solution Approach 1:
The patent segments the fault detection problem into two independent parts: (1) learning normal system behavior from historical normal data to establish data-driven ARRs, and (2) detecting faults by comparing current behavior against the learned normal patterns. This segmentation allows the system to achieve reliable fault detection using only normal operation data, without requiring fault data for training.
Solution Approach 2:
The patent performs preliminary learning of normal system behavior and redundancy relationships during the normal operation phase. By pre-establishing the baseline of normal behavior through data-driven ARR generation, the system is prepared to detect faults immediately when they occur, without needing to have seen fault data during training.
3Measurement precision
If analytical redundancy relationships are derived from complete system models, then fault isolation capability is improved, but computational cost and data requirements increase
Solution Approach 1:
The patent extracts only the essential redundancy relationships needed for fault detection and isolation from the historical data, rather than requiring complete system models. By using feature selection and variable importance analysis, the system identifies and extracts the minimal set of variables and relationships that are sufficient for effective FDI, reducing data requirements while maintaining precision.
Data Source
AI summary
Example implementations described herein involve a new data-driven analytical redundancy relationship (ARR) generation for fault detection and isolation. The proposed solution uses historical data during normal operation to extract the data-driven ARRs among sensor measurements, and then uses them for fault detection and isolation. The proposed solution thereby does not need to rely on the system model, can detect and isolate more faults than traditional data-driven methods, can work when the system is not fully observable, and does not rely on a vast amount of historical fault data, which can save on memory storage or database storage. The proposed solution can thereby be practical in many real cases where there are data limitations.