Denoising Deep Autoencoder Gaussian Mixture Model and Its Application for Robust Nonlinear Industrial Process Monitoring

Process monitoring on high-dimensional nonlinear data is of great significance in the industrial process. This paper presents a denoising deep autoencoder Gaussian mixture model (DDAGMM) for anomaly detection in the industrial process. We add Gaussian white noise as preprocessing to the input data,...

Celý popis

Uloženo v:
Podrobná bibliografie
Vydáno v:2021 International Conference on Computer Information Science and Artificial Intelligence (CISAI) s. 67 - 73
Hlavní autoři: Zhou, Yin-Chang, Li, Meng-Qian, Ji, Long-Bin
Médium: Konferenční příspěvek
Jazyk:angličtina
Vydáno: IEEE 01.09.2021
Témata:
Denoising Deep Autoencoder Gaussian Mixture Model (DDAGMM)
Estimation
fault detection
Feature extraction
Gaussian noise
Information science
kernel density estimation (KDE)
Noise reduction
Process monitoring
robust process monitoring
White noise
On-line přístup:Získat plný text
Tagy: Přidat tag
Žádné tagy, Buďte první, kdo vytvoří štítek k tomuto záznamu!
Popis
Shrnutí:Process monitoring on high-dimensional nonlinear data is of great significance in the industrial process. This paper presents a denoising deep autoencoder Gaussian mixture model (DDAGMM) for anomaly detection in the industrial process. We add Gaussian white noise as preprocessing to the input data, which is further fed into a deep autoencoder to generate a low-dimensional representation and reconstruction error. Finally, we propose a monitoring strategy based on sample energy criterion to judge whether the new sample is anomaly or not. The DDAGMM can reduce the influence of outliers in the original data, has strong robustness, and can handle multi-modal data well. Compared with PCA, LDA and DAGMM based monitoring methods, the proposed counterpart shows superior performance.
DOI:10.1109/CISAI54367.2021.00021