Anomalous Sound Detection Using Variational Autoencoder Feature Extraction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Unsupervised anomaly detection in industrial settings faces challenges in designing a feature amount extraction function, especially when training data for anomalous signals is unavailable, leading to low accuracy in detecting anomalous sounds.
Innovation Solution
The development of an anomalous sound detection apparatus that uses a feature amount extraction function based on probability distribution modeling, incorporating both normal and anomalous sound models, and a threshold set using degrees of anomaly from normal sound data, optimized using a Neyman-Pearson-type optimization index and variational autoencoder, to generate acoustic features and determine anomaly levels.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of manufacture
If unsupervised anomalous sound detection is used to avoid collecting anomalous sound training data, then implementation cost is reduced, but detection accuracy deteriorates due to difficulty in designing feature amount extraction function
Solution Approach 1:
The patent introduces an intermediary component - the feature amount extraction function trained with both normal and anomalous sound data. This intermediary transforms raw sound data into optimized feature representations that enable accurate anomaly detection. By pre-training the extraction function with diverse sound data including anomalies, the system bridges the gap between unsupervised detection and accurate anomaly identification, allowing the use of simpler unsupervised models while maintaining high detection accuracy.
2Reliability
If manual monitoring of industrial devices is implemented, then detection reliability is improved, but operational cost increases due to personnel and traveling expenses
Solution Approach 1:
The patent implements self-service by enabling the anomaly detection system to automatically monitor industrial devices without requiring human intervention. The sound collection apparatus autonomously captures operational sounds, the feature extraction function automatically processes the data, and the system independently identifies anomalies. This automated self-monitoring replaces manual inspection, eliminating personnel and traveling costs while maintaining reliable detection through continuous automated surveillance.
3Measurement precision
If automated anomaly detection rules are designed for each machine type, then detection precision is improved, but device complexity and implementation cost increase
Solution Approach 1:
The patent achieves universality by developing a general-purpose feature amount extraction function that can be applied across different machine types. Instead of creating separate detection rules for each device, the system uses a unified extraction function trained on diverse sound data from various industrial devices. This single multi-functional component adapts to different machine types, providing accurate anomaly detection without requiring complex customizations for each device, thereby reducing overall system complexity.
Data Source
AI summary
To provide an anomalous sound detection training technique by which a feature amount extraction function for detecting anomalous sound can be generated irrespective of whether training data for anomalous signals is available or not. An anomalous sound detection training apparatus includes: a first function updating unit 3 that updates a feature amount extraction function and an feature amount inverse transformation function, which are input, based on an optimization index of a variational autoencoder; an acoustic feature extraction unit 4 that extracts an acoustic feature of normal sound based on training data for normal sound; a normal sound model updating unit 5 that updates a normal sound model by using the acoustic feature that is extracted; a threshold updating unit 6 that obtains a threshold φρ corresponding to a false positive rate ρ, which has a predetermined value, by using the training data for normal sound and the feature amount extraction function that is input; and a second function updating unit 8 that updates the feature amount extraction function that is updated, based on a Neyman-Pearson-type optimization index defined by the threshold φρ that is obtained, and repeatedly performs processing of each of the above-mentioned units.