Equipment Sound Anomaly Detection With Adaptive Normalizing Flow
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current anomaly detection methods for industrial equipment face challenges in balancing cost and accuracy, particularly in unsupervised settings where data collection is costly and learning is unstable, especially when applying domain transformation without pair data.
Innovation Solution
An anomaly detection apparatus using adaptive batch normalization in Normalizing Flow to estimate anomaly degrees based on associations between normal sound distributions from multiple equipment, enabling efficient learning of normal models and stable domain transformation without requiring 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 the cost of data collection and learning is reduced, but the accuracy of anomaly detection deteriorates due to inability to capture equipment-specific variations
Solution Approach 1:
The patent segments the learning process into two distinct phases: (1) learning a common normal model from multiple equipment to capture general characteristics, and (2) adapting this model to individual equipment using equipment-specific normal sound data. This segmentation allows the system to benefit from both shared knowledge and equipment-specific customization, resolving the contradiction between cost efficiency and detection accuracy.
Solution Approach 2:
The patent performs preliminary learning of a common normal model before adapting it to individual equipment. This preliminary action establishes a foundation that can be efficiently adapted later, reducing the overall cost and complexity while maintaining high accuracy through subsequent equipment-specific adaptation.
2Measurement precision
If individual normal models are learned for each piece of equipment, then the accuracy of anomaly detection is improved, but the cost of data collection and learning increases
Solution Approach 1:
The patent merges the learning of common characteristics and equipment-specific characteristics into a unified adaptive framework. By combining the common normal model with equipment-specific adaptations, the system achieves high accuracy without requiring completely separate learning processes for each equipment, thereby controlling costs.
Solution Approach 2:
The patent changes parameters of the common normal model to adapt it to individual equipment characteristics. Instead of learning entirely new models for each equipment, the system adjusts parameters of the existing common model using equipment-specific data, reducing the computational cost and data requirements while maintaining high detection accuracy.
3Device complexity
If domain transformation is performed without pair data, then the complexity of data collection is reduced, but the stability of learning deteriorates
Solution Approach 1:
The patent introduces an intermediary adaptive normal model that bridges the gap between source domain equipment and target domain equipment. This intermediary model is first trained on source equipment and then adapted to target equipment, providing a stable learning pathway that avoids the instability associated with direct domain transformation without pair data.
Solution Approach 2:
The patent incorporates feedback mechanisms in the adaptive learning process, where the performance of the transformed model on target equipment data is continuously evaluated and used to refine the adaptation. This feedback loop stabilizes the learning process by ensuring that transformations are optimized for the specific target domain characteristics.
Data Source
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).