A Machine Vision Anomaly Detection System to Industry 4.0 Based on Variational Fuzzy Autoencoder

Area of Science:

• Industrial Automation and Cybersecurity
• Machine Learning Applications
• Cyber-Physical Systems Security

Background:

• Industry 4.0 integrates cyber-physical systems, creating vulnerabilities for cyberattacks.
• Effective security requires continuous threat assessment and stakeholder awareness.
• Anomaly detection is crucial for identifying deviations from expected industrial processes.

Purpose of the Study:

• To identify production line defects stemming from cyberattacks using advanced machine vision.
• To propose an original variational fuzzy autoencoder (VFA) methodology for anomaly detection.
• To enhance the cybersecurity of Industry 4.0 environments.

Main Methods:

• Development of a variational fuzzy autoencoder (VFA) methodology.
• Integration of fuzzy entropy and Euclidean fuzzy similarity measurements.
• Implementation of nonlinear transformations through deterministic functions for realistic vision.

Main Results:

• The proposed VFA system accurately evaluates and categorizes anomalies.
• The system demonstrates high accuracy in complex industrial environments.
• The machine vision approach effectively identifies cyberattack-induced defects.

Conclusions:

• The VFA methodology offers a robust solution for detecting cyberattack-related anomalies in Industry 4.0.
• Advanced machine vision techniques are vital for securing smart manufacturing.
• The system provides a realistic and accurate means of identifying production line defects.