Interpretability of Causal Discovery in Tracking Deterioration in a Highly Dynamic Process

Area of Science:

• Industrial Engineering
• Manufacturing Process Monitoring
• Data Science

Background:

• Mechanical degradation significantly impacts product quality and efficiency in dynamic manufacturing.
• Viscose fiber production is a complex, dynamic process susceptible to degradation.
• Existing anomaly detection methods lack interpretability.

Purpose of the Study:

• To develop a novel, interpretable approach for monitoring mechanical degradation in viscose fiber production.
• To leverage causal discovery techniques to enhance understanding of degradation mechanisms.
• To validate the proposed method against state-of-the-art techniques.

Main Methods:

• Utilized causal discovery techniques to build interpretable causal graphs.
• Incorporated domain expertise into the causal graph construction.
• Applied the method to a case study in viscose fiber production.
• Compared results with LSTM-based autoencoder, USAD, and TranAD.

Main Results:

• The causal discovery approach provides enhanced interpretability for degradation monitoring.
• Changes in causal relations effectively identify potential problems in the production process.
• The method demonstrates alignment and validation with existing state-of-the-art techniques.
• The approach proved effective in a real-world viscose fiber production setting.

Conclusions:

• Causal discovery offers a powerful, interpretable tool for monitoring degradation in dynamic manufacturing.
• The proposed method enhances process optimization and quality control in the textile industry.
• This work contributes to advancing anomaly detection and process monitoring strategies.