Equipment State Classification Using Periodic Anomaly Models
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing equipment monitoring systems rely on threshold values for detecting abnormalities, which can be operator-dependent and prone to errors, and may not accurately classify measurement data when the equipment state changes or includes noise, leading to inaccurate determination results.
Innovation Solution
A system that uses multiple classification models learned from measurement data for each period to determine the state of equipment as normal or abnormal, taking a logical product of classification results and generating new models based on recent data to adapt to changing conditions, thereby eliminating the need for threshold values and improving accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If threshold values are used for detecting abnormalities, then the detection method is simple, but the determination accuracy deteriorates due to operator dependence and inability to handle changing equipment states
Solution Approach 1:
The patent divides the monitoring system into multiple classification models, each trained on measurement data from different periods. This segmentation allows the system to capture changing equipment states without requiring a complete system redesign, resolving the contradiction between accuracy and complexity by breaking down the problem into manageable periodic segments.
Solution Approach 2:
The patent implements dynamic adaptation by periodically retraining classification models with new measurement data. This dynamic approach allows the system to automatically adjust to changing equipment states and noise patterns, improving determination accuracy without relying on static threshold values that require manual adjustment.
2Measurement precision
If multiple classification models are used for each period, then the determination accuracy improves, but the device complexity increases
Solution Approach 1:
The patent employs periodic action by training new classification models at regular intervals using measurement data from each period. This periodic model generation and evaluation approach systematically manages multiple models, improving accuracy through diverse temporal perspectives while maintaining organized model lifecycle management.
Solution Approach 2:
The patent implements feedback mechanisms where classification results from multiple models are evaluated, and the best-performing models are selected for deployment. This feedback loop allows the system to automatically refine model selection based on performance metrics, improving accuracy while managing complexity through data-driven model optimization.
3Reliability
If threshold values are used for abnormality detection, then the system is easy to operate, but the reliability deteriorates due to operator dependence and error-prone threshold setting
Solution Approach 1:
The patent enables self-service by automatically training classification models on historical measurement data without requiring manual threshold setting. The system autonomously learns normal and abnormal patterns from the data, improving reliability by eliminating operator-dependent threshold selection while maintaining ease of operation through automated model generation and deployment.
Solution Approach 2:
The patent transforms the approach from fixed threshold parameters to dynamic classification parameters learned from data. By changing from static threshold values to adaptive classification models that evolve with equipment states, the system improves reliability while maintaining operational simplicity through automated parameter optimization.
4Adaptability or versatility
If classification models are regenerated for each period, then the adaptability to changing conditions improves, but the processing time increases
Solution Approach 1:
The patent applies preliminary action by pre-training classification models on historical measurement data during off-peak periods. This advance preparation allows the system to have ready-to-use models that can be quickly deployed when needed, improving adaptability to changing conditions while minimizing the time loss during actual monitoring operations.
Solution Approach 2:
The patent implements partial action by selectively training and evaluating only the most relevant classification models based on current equipment states and data quality. This selective approach reduces the overall processing time while maintaining high adaptability by focusing computational resources on the most impactful models rather than processing all possible models equally.
Data Source
AI summary
Provided is an apparatus including: an acquisition unit which acquires measurement data indicating a state of a target; a supply unit which supplies the measurement data acquired by the acquisition unit to a plurality of classification models respectively learned by learning data, which includes measurement data in a period in which the state of the target is normal, for periods different from each other, the plurality of classification models classifying measurement data as either normal or abnormal in response to the measurement data being input; and a determination unit which determines the state of the target as either normal or abnormal based on a plurality of classification results output from the plurality of classification models.