Abnormality Detection Model Generation with Virtual Feature Data
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing abnormality detection systems face challenges in generating accurate models for predictive maintenance in production sites, especially when insufficient abnormality/normality information is available for training, leading to suboptimal detection accuracy before actual operations.
Innovation Solution
An abnormality detection system that includes a control operation part, a first and second abnormality detection part, a state value storage part, and a model generation part, which generates feature values, selects combinations of feature values, and creates an extra learning data set using virtual feature values to evaluate detection accuracy, allowing for improved model generation and accuracy evaluation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If a model is constructed using limited abnormality/normality information available before actual operation, then the system can be deployed for predictive maintenance, but the detection accuracy is insufficient
Solution Approach 1:
The patent creates virtual copies of actual operation data through simulation. A data generation unit synthesizes virtual operation data that mimics real-world sensor readings, including both normal and abnormal states. This virtual data is then used to train the abnormality detection model, allowing the system to achieve high detection accuracy without requiring extensive actual abnormality data during the model construction phase before actual operation begins.
2Measurement precision
If more feature values are used in the model, then detection accuracy improves, but the complexity of model construction increases
Solution Approach 1:
The patent implements automatic feature value selection where the system evaluates and selects optimal feature values without requiring manual expert intervention. The data generation unit and model construction unit work together to automatically determine which feature values from the virtual operation data are most effective for detecting specific abnormality types. This self-service approach reduces model construction complexity while maintaining high detection accuracy through data-driven feature selection.
Data Source
AI summary
An abnormality detection system, support device, and model generation method are provided for generating a more highly accurate abnormality detection model before an actual operation. A model generation part of an abnormality detection system includes a section for generating plural feature values from state values provided from a state value storage part, a section for selecting a combination of one or plural feature values among the plural generated feature values, a section for generating an extra learning data set having at least part of a data series of the feature values of the selected combination and a data series of statistically generated virtual feature values, and a section for evaluating a detection accuracy of a model corresponding to the feature values of the selected combination using the extra learning data set.