Autoencoder-Ensemble-Based Unsupervised Selection of Production-Relevant Variables for Context-Aware Fault Diagnosis

Area of Science:

• Industrial Engineering
• Computer Science
• Data Science

Background:

• Smart factories face challenges with complex cyber-physical systems generating vast, difficult-to-analyze data.
• Limited network capabilities hinder real-time data capture and introduce faults during diagnosis, risking production slowdowns or outages.

Purpose of the Study:

• To develop a novel approach for automatically selecting essential shop floor parameters in smart factories.
• To reduce the number of surveyed variables for efficient data analysis and fault diagnosis without network overload.

Main Methods:

• Utilized an autoencoder ensemble with minority voting to differentiate between normal and high-entropy production variables.
• Implemented automatic selection of production-relevant shop floor parameters.

Main Results:

• Successfully decreased the number of surveyed variables while maintaining fault diagnosis quality.
• Achieved higher production throughput and mitigated communication losses.
• Prevented disruptions to factory instructions.

Conclusions:

• The proposed method effectively manages data complexity in smart factories.
• It enhances fault diagnosis accuracy and network efficiency, ensuring uninterrupted production.