Deep Probabilistic Principal Component Analysis for Process Monitoring

Area of Science:

• Industrial Process Monitoring
• Machine Learning
• Fault Detection

Background:

• Probabilistic latent variable models (PLVMs) like PPCA are crucial for industrial process monitoring.
• Existing methods may lack the feature extraction capabilities needed for complex industrial data.

Purpose of the Study:

• To propose a novel Deep Probabilistic Principal Component Analysis (DePPCA) model.
• To enhance process monitoring and fault detection using deep learning and probabilistic modeling.

Main Methods:

• DePPCA construction involves greedy layer-wise pretraining and end-to-end fine-tuning.
• Hierarchical deep structure extraction using cascaded PPCA modules.
• Theoretical validation through variational inference.

Main Results:

• DePPCA achieves superior monitoring performance even with univariate feature compression.
• The model enables fast feature extraction and online monitoring procedures.
• Effectiveness demonstrated on the Tennessee Eastman (TE) and multiphase flow (MPF) processes.

Conclusions:

• DePPCA offers an accurate and efficient approach to industrial process monitoring.
• The model integrates deep learning and probabilistic modeling for advanced fault detection.
• The proposed method allows for rapid and effective monitoring using minimal extracted features.