Factor Analysis of Multi-System Anomalies Under Normal State Variation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing anomaly detection methods in equipment, such as rotating equipment, struggle to accurately diagnose anomalies due to variations in normal states caused by factors like rotation speed, aging, and environmental conditions.
Innovation Solution
A factor analysis device that acquires a normal model from memory, calculates the distance between the normal model and monitoring target data for each system, and identifies systems contributing to anomalies based on these distances.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If a competitive neural network is used to learn sensor observation values, then the anomaly detection can handle variations in normal state, but the measurement precision of anomaly factors deteriorates
Solution Approach 1:
The patent segments the anomaly analysis into two distinct parts: (1) a competitive neural network that learns normal state variations across multiple systems, and (2) a factor analysis unit that calculates distances to identify specific anomaly sources. This segmentation allows the system to adapt to normal variations while maintaining precision in identifying actual anomaly factors by separating the learning function from the diagnostic function.
Solution Approach 2:
The patent introduces a factor analysis unit as an intermediary between the competitive neural network and the anomaly diagnosis output. This intermediary calculates the distance between learned normal models and actual observation values, translating the neural network's adaptive learning into precise anomaly factor identification. The factor analysis unit acts as a mediator that converts adaptive pattern recognition into measurable anomaly diagnostics.
2Reliability
If monitoring data from multiple systems is analyzed, then the comprehensiveness of anomaly detection improves, but the device complexity increases
Solution Approach 1:
The patent segments the multi-system monitoring into independent normal model learning processes for each system, where the competitive neural network learns normal states separately for each system. This segmentation allows comprehensive multi-system monitoring while maintaining manageable complexity by treating each system's normal state learning as an independent task that can be parallelized.
Solution Approach 2:
The patent creates a universal competitive neural network architecture that can learn and adapt to normal states across multiple different systems simultaneously. This multi-functional approach allows the same neural network structure to handle various systems (e.g., different sensors, equipment) without requiring separate complex analysis mechanisms for each, thereby improving comprehensiveness while controlling complexity through architectural universality.
Data Source
AI summary
A normal model is acquired from a memory. A monitoring target data for each of a plurality of systems is acquired. A distance between the normal model and a value based on the monitoring target data for each of the plurality of systems is calculated. A system that provides an anomaly factor is identified from among the plurality of systems based on the distance.