Fault diagnosis based on counterfactual inference for the batch fermentation process.

Area of Science:

• Industrial Process Control
• Machine Learning Applications
• Biotechnology

Background:

• Effective fault diagnosis is crucial for industrial processes, but current methods struggle with fault amplitude assessment and efficiency in batch fermentation.
• Existing techniques often lack the precision needed for complex biological systems like L. plantarum fermentation.

Purpose of the Study:

• To develop an advanced fault detection and diagnosis model for batch fermentation processes.
• To address limitations in assessing fault amplitude and improve diagnostic efficiency.

Main Methods:

• Utilized mutual information (MI) for quality-related process variable selection.
• Employed a convolutional neural network based on variational autoencoder (CNN-VAE) for fault detection.
• Developed a fault diagnosis based on counterfactual inference (FDCI) for root cause analysis.
• Constructed two statistics from latent and residual domains of the CNN-VAE model.

Main Results:

• The proposed CNN-VAE model effectively detects faults in process data.
• FDCI successfully identifies the root causes of detected faults.
• The approach demonstrated high effectiveness in both simulated and real-world L. plantarum batch fermentation processes.

Conclusions:

• The integrated CNN-VAE and FDCI approach offers a robust solution for fault detection and diagnosis in batch fermentation.
• This method improves upon existing techniques by providing fault amplitude insights and enhanced efficiency.
• The study validates the model's applicability in complex biotechnological processes.