Equipment Sound Anomaly Detection Using Adaptive Normal Models
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current anomaly detection methods for industrial equipment face a trade-off between cost and accuracy, with unsupervised approaches requiring minimal data collection but potentially compromising detection precision, and domain transformation techniques like StarGAN experiencing unstable learning without pair data.
Innovation Solution
Anomaly detection apparatus using adaptive batch normalization in Normalizing Flow to learn a common normal model from multiple equipment, allowing for efficient adaptation to individual equipment models with reduced data collection and stabilizing domain transformation without pair data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of energy
If a common normal model is learned using normal sound from multiple pieces of equipment, then cost for data collection and learning is reduced, but anomaly detection accuracy deteriorates due to inability to capture slight differences in normal sound distribution for each equipment
Solution Approach 1:
The patent segments the learning process into two distinct phases: (1) learning a common normal model from normal sound of multiple equipment, and (2) adaptively learning individual equipment characteristics by adjusting the common model using normal sound from each specific equipment. This segmentation allows reuse of the common model across equipment while capturing individual variations, thereby reducing overall data collection costs while maintaining detection accuracy.
Solution Approach 2:
The patent performs preliminary learning of a common normal model from multiple equipment before adapting it to individual equipment. This preliminary action establishes a baseline model that captures general normal sound patterns, which then serves as a foundation for subsequent adaptive learning with minimal additional data from each specific equipment, reducing total data collection requirements while preserving accuracy.
2Measurement precision
If individual normal models are learned for each piece of equipment using only normal sound from that equipment, then anomaly detection accuracy is improved, but cost for data collection and learning increases with the number of equipment
Solution Approach 1:
The patent merges the learning processes by having individual equipment models share a common normal model as their foundation. Instead of independently learning separate models for each equipment, the system combines the common knowledge from multiple equipment with equipment-specific adaptations, thereby reducing redundant data collection and learning costs while maintaining the ability to detect equipment-specific anomalies.
Solution Approach 2:
The common normal model serves as a universal foundation that can be applied across multiple equipment of the same type. This universal model captures general normal operating patterns that are common to all equipment, allowing the system to function effectively across multiple devices without requiring complete independent learning for each one, thereby reducing overall costs while maintaining accuracy.
3Loss of energy
If domain transformation techniques like StarGAN are used without pair data, then data collection requirements are reduced, but learning stability deteriorates
Solution Approach 1:
The patent introduces adaptive batch normalization as an intermediary mechanism that stabilizes the domain transformation process when learning without pair data. This intermediary component provides a stable reference framework that guides the transformation from common normal sound distribution to individual equipment normal sound distribution, ensuring learning stability even in the absence of paired training data.
Solution Approach 2:
The patent utilizes parameter changes in the batch normalization layers during adaptive learning to stabilize domain transformation. By dynamically adjusting normalization parameters (mean and variance) based on the specific equipment's normal sound characteristics, the system achieves stable learning without requiring paired training data, effectively bridging the common model and individual equipment models.
Data Source
Figure 1~2
Figure 3~4
Figure 5~6
AI summary
An anomaly detection technique which realizes high accuracy while reducing cost required for normal model learning is provided. An anomaly detection apparatus includes an anomaly degree estimating unit configured to estimate an anomaly degree indicating a degree of anomaly of anomaly detection target equipment from sound emitted from the anomaly detection target equipment (hereinafter, referred to as anomaly detection target sound) based on association between a first probability distribution indicating distribution of normal sound emitted from one or more pieces of equipment different from the anomaly detection target equipment and normal sound emitted from the anomaly detection target equipment (hereinafter, referred to as normal sound for adaptive learning).