Towards robust and understandable fault detection and diagnosis using denoising sparse autoencoder and smooth...

Area of Science:

• Industrial Process Control
• Machine Learning for Engineering
• Data Analytics in Manufacturing

Background:

• Industrial fault detection and diagnosis require high accuracy and interpretability.
• Existing autoencoder methods struggle with noise and generalization.
• Understanding root causes of process disturbances is crucial for effective intervention.

Purpose of the Study:

• To develop a robust and understandable two-step framework for fault detection and diagnosis.
• To improve fault detection accuracy and generalization using denoising sparse autoencoder.
• To enhance root-cause analysis of faults with smooth integrated gradients.

Main Methods:

• Fault detection using denoising sparse autoencoder (DSAE) for noise robustness and generalization.
• Fault diagnosis and root-cause variable identification using smooth integrated gradients (SIG) for feature importance denoising.
• Evaluation on the Tennessee Eastman process dataset.

Main Results:

• The DSAE-SIG method demonstrated higher diagnosis accuracy compared to existing methods.
• Successful identification of potential root-cause variables for process disturbances.
• The framework provides a more understandable output for industrial applications.

Conclusions:

• The proposed DSAE-SIG framework effectively addresses challenges in industrial fault detection and diagnosis.
• This approach offers improved accuracy, robustness, and interpretability in identifying fault causes.
• The method is validated for practical application in industrial settings like the Tennessee Eastman process.