Autoencoder for Non-Stationary Vibration Detection in Noisy Environments
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current artificial intelligence systems for detecting non-stationarity in objects based on vibration require noise canceling processing to exclude environmental sounds, which is time-consuming and costly, and necessitate large datasets labeled as normal and abnormal sounds, increasing effort and cost.
Innovation Solution
A trained autoencoder is developed to learn stationary sound data, including environmental noise, allowing it to differentiate between stationary and non-stationary states without the need for noise canceling, using pre-training with stationary vibration feature data to minimize differences between input and output, enabling accurate detection of non-stationarity even in the presence of environmental sounds.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If noise canceling processing is performed to exclude environmental sounds, then detection accuracy is improved, but training time and cost increase
Solution Approach 1:
The patent applies preliminary action by pre-training the autoencoder with stationary sound data (including environmental noise) before actual detection. This pre-training enables the model to automatically adapt to and filter out environmental sounds during the detection phase, eliminating the need for separate noise canceling processing and significantly reducing training time while maintaining detection accuracy.
2Measurement precision
If noise canceling processing is performed to exclude environmental sounds, then detection accuracy is improved, but cost increases
Solution Approach 1:
The patent performs preliminary training with stationary sound data that includes environmental noise, enabling the autoencoder to learn and adapt to background sounds. This preliminary adaptation eliminates the need for expensive and time-consuming noise canceling processing in the detection phase, thereby reducing overall training costs while maintaining high detection accuracy.
3Measurement precision
If large datasets with labeled normal and abnormal sounds are used, then detection accuracy is improved, but effort and cost increase
Solution Approach 1:
The patent applies preliminary action by using only stationary sound data (without requiring labeled abnormal sound data) for pre-training the autoencoder. This approach eliminates the time-consuming process of collecting, labeling, and preparing large datasets with both normal and abnormal sound labels, while still achieving high detection accuracy through the autoencoder's ability to detect deviations from the learned stationary pattern.
4Measurement precision
If large datasets with labeled normal and abnormal sounds are used, then detection accuracy is improved, but effort and cost increase
Solution Approach 1:
The patent performs preliminary training using only stationary sound data, which significantly reduces the effort and cost of data preparation. By eliminating the need to collect and label abnormal sound data, the approach maintains high detection accuracy while greatly simplifying the data preparation process and reducing associated costs.
5Ease of operation
If the autoencoder learns only stationary sound data, then training simplicity is improved, but ability to detect non-stationarity in noisy environments deteriorates
Solution Approach 1:
The patent applies preliminary action by pre-training the autoencoder with stationary sound data that includes environmental noise. This pre-training enables the model to learn the characteristics of both the stationary sound and the environmental noise simultaneously, maintaining training simplicity while ensuring reliable detection of non-stationarity even in noisy environments.
Data Source
AI summary
Provided is a trained artificial intelligence for detecting non-stationarity of an object, which accurately functions even in the presence of an environmental sound. Stationary vibration feature data generated from stationary vibration data that is data about stationary vibration including vibration generated in a stationary state from an object for which detection of non-stationarity is performed based on sound, the stationary vibration feature data being data about a feature of stationary vibration identified by the stationary vibration data, is input to an autoencoder to cause the autoencoder to output estimated stationary vibration feature data. A loss function between the stationary vibration feature data and the estimated stationary vibration feature data is generated, and the autoencoder is trained to minimize a difference therebetween. By repeating the above-mentioned processing, the trained autoencoder is obtained from the autoencoder.