Anomaly Threshold Setting for Predictive Maintenance Models
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Setting an appropriate threshold value for anomaly detection in predictive maintenance is challenging due to the low frequency of anomaly occurrences, making it difficult even for experts to set accurate values based on normal or slightly abnormal data.
Innovation Solution
An information processing device connected to a control device that includes a trained model to calculate scores from feature amounts, with display and calculation means to facilitate setting threshold values for either feature amounts or scores, allowing users to visualize and set threshold values more easily, and output these values for application to the trained model.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If a threshold value is set based only on normal or slightly abnormal data, then the initial threshold can be established, but the threshold cannot accurately detect rare anomalies
Solution Approach 1:
The patent introduces an information processing device as an intermediary that automatically calculates and sets threshold values using trained models and statistical processing. This mediator bridges the gap between limited anomaly data and accurate threshold setting by processing normal data through machine learning models to derive appropriate thresholds without requiring extensive rare anomaly samples.
Solution Approach 2:
The system performs self-service by automatically setting threshold values through trained models without requiring manual expert intervention. The trained model processes normal operation data and automatically determines appropriate threshold values, eliminating the need for experts to manually analyze rare anomaly cases and set thresholds based on limited data.
2Measurement precision
If experts manually set threshold values based on specialized knowledge, then accurate thresholds may be achieved, but the process becomes complex and time-consuming
Solution Approach 1:
The system enables self-service by automatically calculating threshold values using trained models and statistical processing of normal operation data. This eliminates the need for experts to manually analyze data and set thresholds, significantly reducing the time required while maintaining accuracy through automated machine learning-based calculations.
Solution Approach 2:
The patent replaces the manual mechanical process of expert threshold setting with an automated information processing system. The trained model and calculation unit substitute for expert human analysis, using statistical methods and machine learning to automatically determine threshold values, thereby reducing both time and human resource requirements.
3Measurement precision
If the trained model uses complex feature amounts, then detection accuracy improves, but the difficulty of setting and understanding threshold values increases
Solution Approach 1:
The information processing device acts as an intermediary that handles the complexity of trained models internally. It automatically calculates threshold values based on complex feature amounts processed by the trained model, shielding users from the underlying complexity while maintaining high detection precision through sophisticated machine learning processing.
Data Source
AI summary
A control device includes a trained model for receiving a feature amount calculated from information collected from the control target to output a score, and a threshold value for determining the score An information processing device displays a feature amount related to the trained model and a score calculated from the feature amount, receives setting of a threshold value for any of the feature amount and the score, calculates a threshold value for the score from the threshold value set for the feature amount, and outputs, as a threshold value to be applied to the trained model, any of the threshold value set for the score and the threshold value for the score calculated from the threshold value set for the feature amount.