Detecting Faults at the Edge via Sensor Data Fusion Echo State Networks.

Area of Science:

• Industrial Automation
• Artificial Intelligence
• Sensor Technology

Background:

• Industry 4.0 relies heavily on sensors and actuators, necessitating advanced diagnostic capabilities beyond simple telemetry.
• Correlating large volumes of sensor data for effective fault diagnosis in complex industrial systems is challenging.
• Edge devices have limited resources, restricting the deployment of complex AI models for real-time analysis.

Purpose of the Study:

• To propose an Echo State Network (ESN) architecture for industrial fault detection.
• To utilize sensor data fusion within the ESN to enhance diagnostic accuracy.
• To enable deployment of fault detection models on resource-constrained Edge devices.

Main Methods:

• Implemented an Echo State Network (ESN), a type of Recurrent Neural Network (RNN) with sparse weights.
• Employed sensor data fusion techniques to integrate heterogeneous data sources (vibration and current signals).
• Deployed the ESN model on a scale replica industrial plant for fault detection.

Main Results:

• The proposed ESN model successfully detected the majority of faults in the industrial plant.
• The ESN architecture demonstrated low complexity and memory footprint, suitable for Edge deployment.
• Achieved a favorable trade-off between precision, recall, F1-score, and inference time compared to other methods.

Conclusions:

• Echo State Networks combined with sensor data fusion offer a viable solution for industrial fault detection on Edge devices.
• The ESN approach provides an efficient and accurate method for diagnosing faults in Industry 4.0 environments.
• The study validates the feasibility of deploying sophisticated AI models on resource-constrained edge computing platforms.