Batch process fault detection for multi-stage broad learning system.

Area of Science:

• Industrial Process Monitoring
• Machine Learning for Fault Detection
• Chemical Engineering

Background:

• Traditional multivariate statistical analysis struggles with minor faults caused by equipment aging and catalyst deactivation.
• Deep neural networks offer better feature extraction but suffer from long training times, hindering real-time applications.
• The Broad Learning System (BLS) allows for network expansion without retraining, reducing training time.

Purpose of the Study:

• To develop an efficient and accurate multi-stage process monitoring framework for industrial production.
• To address the limitations of existing methods in detecting subtle faults.
• To improve the speed and effectiveness of online fault detection.

Main Methods:

• Utilized the Affinity Propagation (AP) algorithm to segment the production process into distinct stages based on their unique characteristics.
• Integrated the AP algorithm with the Broad Learning System (BLS) to create a novel AP-BLS monitoring framework.
• Applied the framework to a penicillin fermentation process for performance evaluation.

Main Results:

• The proposed AP-BLS model demonstrated superior performance in monitoring results compared to other conventional models.
• The framework effectively detected minor faults that are often missed by traditional methods.
• The integration of AP and BLS enabled efficient stage-specific monitoring and faster online response.

Conclusions:

• The AP-BLS framework offers a powerful solution for multi-stage industrial process monitoring and fault detection.
• This approach overcomes the limitations of traditional methods and standard deep learning models in terms of training time and detection accuracy.
• The model's effectiveness is validated through its successful application in the penicillin fermentation process.