Abnormality Detection Using Supervised Shapelet Learning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In abnormality detection for infrastructure and manufacturing devices, it is challenging to learn classification models without abnormal data, leading to inconsistent shapelet presentation and requiring hyperparameter adjustments, which affects determination basis performance.
Innovation Solution
An information processing device that uses a supervised shapelet learning method to learn a detection model from normal data only, generating shapelets during operation for abnormality detection without requiring hyperparameter adjustments, and outputs a consistent basis for abnormality detection.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If abnormal data is used for learning classification models, then classification performance is improved, but data collection difficulty increases and reliability decreases
Solution Approach 1:
The patent converts the unavailability of abnormal data (a harmful constraint) into a benefit by using only normal data for training. The detection model learns to identify abnormalities by understanding normal patterns, transforming the limitation into an advantage for data collection ease and reliability.
Solution Approach 2:
Instead of training on abnormal data to detect abnormalities, the patent inverts the approach by training on normal data only. The classification model learns normal patterns and automatically identifies deviations as abnormalities, reversing the conventional training paradigm.
2Ease of manufacture
If shapelet learning is performed without abnormal data, then data collection ease is improved, but classification accuracy deteriorates
Solution Approach 1:
The patent transforms the limitation of having only normal data into a benefit by designing a detection model that specifically looks for deviations from normal patterns. This approach maintains high classification accuracy while eliminating the need for abnormal data collection.
Solution Approach 2:
The detection model serves itself by automatically identifying abnormalities based on learned normal patterns without requiring external abnormal data for training. The system uses its own learned representations to detect deviations, eliminating the need for separate abnormal data collection.
3Reliability
If hyperparameter adjustment is performed, then classification performance is improved, but device complexity increases
Solution Approach 1:
The patent implements a self-service approach where the detection model automatically determines optimal parameters based on normal data patterns without requiring manual hyperparameter adjustment. The system adapts to data characteristics automatically, eliminating the complexity of hyperparameter tuning.
Solution Approach 2:
The patent changes the fundamental parameters of the learning approach by training on normal data only and using deviation detection, which automatically adjusts model behavior to suit the data availability constraints without requiring manual hyperparameter optimization.
4Loss of information
If shapelets are generated during operation, then determination basis presentation is improved, but processing time increases
Solution Approach 1:
The patent performs preliminary learning of normal patterns during the training phase using only normal data. Once trained, the model can quickly generate shapelets during operation by comparing current data against the pre-learned normal patterns, reducing real-time processing requirements.
Solution Approach 2:
The patent segments the processing into two phases: offline training using normal data to learn patterns, and online inference where shapelets are generated by comparing current data against learned patterns. This segmentation reduces the computational burden during operation while maintaining determination basis presentation.
Data Source
AI summary
According to one embodiment, an information processing device includes one or more processors. The one or more processors are configured to: detect whether input waveform data is in a first state by using a detection model; acquire a plurality of pieces of second state waveform data in a second state detected in advance by using the detection model when detected to be in the first state; learn a classification model for classifying whether waveform data is in the first state or the second state, by using first state waveform data detected to be in the first state and the plurality of pieces of second state waveform data as learning data, to generate one or more partial waveform patterns serving as a basis for indicating that the first state waveform data is in the first state; and output the generated partial waveform pattern.