Operational State Recognition of a DC Motor Using Edge Artificial Intelligence.

Area of Science:

• Artificial Intelligence
• Machine Learning
• Signal Processing
• Industrial IoT

Background:

• Predictive maintenance in industrial applications requires real-time operational state recognition.
• Edge artificial intelligence (EDGE-AI) offers on-device processing without cloud reliance.
• Sound data analysis is a viable, non-invasive method for machine health monitoring.

Purpose of the Study:

• To develop and validate EDGE-AI methodologies for DC motor operational state detection using acoustic data.
• To assess the feasibility of deploying trained AI models on microcontroller units (MCUs).
• To compare different EDGE-AI implementations based on accuracy, latency, and resource usage.

Main Methods:

• Feature extraction from audio datasets to characterize DC motor operational states.
• Training two distinct Convolutional Neural Network (CNN) models for sound-based classification.
• Post-training quantization and model compression for deployment on MCUs.
• Real-time validation experiments, including simulated stress tests, to evaluate performance.

Main Results:

• Successfully deployed and validated two CNN models on MCUs for real-time DC motor state recognition.
• Demonstrated effective operational state identification and transition time detection.
• Achieved promising results in classification accuracy, low latency, and efficient resource utilization.

Conclusions:

• EDGE-AI, utilizing sound data and CNNs, is effective for real-time predictive maintenance of industrial machinery.
• Model optimization techniques enable the deployment of sophisticated AI on resource-constrained MCUs.
• The presented approaches offer a practical solution for enhancing industrial machine monitoring and diagnostics.