Multi-Sensor Abnormality Detection Without Preset Thresholds
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional abnormality detection methods face challenges in detecting sensor abnormalities without presetting a threshold value, especially when operation conditions or environments change, making it difficult to determine abnormalities at the predictive stage.
Innovation Solution
An abnormality detection device that acquires detection results from multiple sensors at predetermined intervals and uses an estimator to identify likely abnormal sensors based on mode information elements extracted through various analysis processes, including principal component analysis and trajectory visualization, without relying on preset threshold values.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If preset threshold values are used for abnormality detection, then detection simplicity is improved, but detection accuracy deteriorates when operation conditions or environments change
Solution Approach 1:
The patent applies dynamics by transitioning from static preset threshold values to dynamic adaptive thresholds. The system automatically adjusts detection thresholds based on real-time operation conditions and environmental parameters, allowing the detection criteria to evolve with changing system states. This resolves the contradiction by maintaining detection simplicity while improving accuracy under varying conditions through automated threshold adaptation.
Solution Approach 2:
The patent implements parameter changes by modifying the threshold values themselves based on detected operation conditions. Instead of using fixed thresholds, the system dynamically adjusts detection parameters (thresholds) according to actual system state, environmental temperature, load conditions, and other variables. This allows the detection mechanism to remain simple in structure while achieving high accuracy through parameter adaptation.
2Device complexity
If preset threshold values are used, then device complexity is reduced, but the ability to detect abnormalities at predictive stage deteriorates
Solution Approach 1:
The patent applies preliminary action by establishing baseline profiles and normal operation patterns before actual abnormality detection begins. The system pre-learning phase captures typical operation conditions and establishes reference thresholds, enabling early detection of deviations. This preliminary preparation enhances predictive detection capability without significantly increasing operational complexity, as the heavy lifting is done during the initial setup phase.
Solution Approach 2:
The patent implements feedback mechanisms where detection results and operation conditions continuously inform threshold adjustments. The system monitors actual system behavior, compares it against learned patterns, and automatically refines detection thresholds based on feedback from real-world operation. This closed-loop approach improves predictive capability while maintaining relatively simple device architecture through automated learning and adaptation.
3Measurement precision
If multiple analysis processes are implemented without preset thresholds, then detection accuracy is improved, but device complexity increases
Solution Approach 1:
The patent applies segmentation by dividing the complex analysis into distinct modular processes: data acquisition module, pattern recognition module, threshold determination module, and detection decision module. Each module performs a specific function and can be independently optimized or adjusted. This segmentation maintains detection accuracy through comprehensive multi-process analysis while managing complexity through clear modular separation and defined interfaces between components.
Data Source
AI summary
An abnormality detection device includes an acquirer configured to acquire a plurality of detection results of a plurality of sensors that have detected a state of a detection target at predetermined time intervals and an estimator configured to estimate a specific sensor among the plurality of sensors on the basis of a plurality of mode information acquisition process data elements for enabling various types of feature extraction according to properties of the abnormality obtained from a plurality of detection results acquired by the acquirer.