Self-Organizing Map Abnormality Warning System
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing abnormality detection methods using Self-Organizing Maps (SOM) often inaccurately identify normal states as abnormal due to large deviations in input data, leading to false warnings, and struggle to distinguish between normal and abnormal states effectively.
Innovation Solution
A method that temporarily accumulates detection data with large distances from nodes, issues warnings only upon repeated acquisition, generates the self-organizing map using normal state data, sets distinct abnormality threshold values, and allows operator evaluation to refine warnings and threshold adjustments, enhancing accuracy and reducing false positives.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If deviation between input data and neuron is used to determine abnormality, then abnormality detection can be performed, but false warnings occur when normal state data shows large deviation
Solution Approach 1:
The system performs preliminary classification of detection data into normal and abnormal categories before issuing warnings. By pre-establishing a normal state model and comparing incoming data against it, the system can identify and filter out normal variations that would otherwise trigger false alarms, while still detecting true abnormalities.
Solution Approach 2:
The system uses feedback mechanisms where detection results are continuously compared with the normal state model, and the model is refined based on accumulated detection data. This feedback loop allows the system to learn from past observations and improve its ability to distinguish between normal variations and true abnormalities, reducing false warnings over time.
2Difficulty of detecting and measuring
If self-organizing map is formed to detect abnormalities, then multi-dimensional detection data can be analyzed, but normal states may be incorrectly identified as abnormal
Solution Approach 1:
The system segments the detection data into distinct categories (normal and abnormal) by mapping multi-dimensional data onto a self-organizing map structure. This segmentation allows the system to handle complex multi-dimensional data while maintaining clear distinction between normal operational variations and true abnormalities, preventing misclassification.
3Speed
If detection data with large deviation is immediately flagged as abnormal, then rapid abnormality detection is achieved, but false alarms increase due to accidental events
Solution Approach 1:
Instead of immediately flagging data with large deviation, the system performs preliminary classification against the normal state model first. This preliminary action filters out normal variations before issuing warnings, maintaining rapid detection capability while reducing false alarms from accidental events.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
A method for issuing a warning of an abnormality of a target object by using a self-organizing map prescribed by a plurality of nodes indicating a normal state of the target object includes acquiring detection data indicating a state of the target object, performing learning of the self-organizing map by using the detection data indicating a distance between a feature value and a node smaller than an abnormality threshold value predefined, storing the detection data indicating the distance larger than the abnormality threshold value in a storage unit as recorded data, and performing notification of the abnormality of the target object when the number of pieces of the recorded data stored in the storage unit reaches a predetermined number.