Hybrid deep learning framework for real-time fault detection in squirrel-cage induction motors

Area of Science:

• Industrial IoT and AI
• Machine Learning for Predictive Maintenance
• Electrical Engineering and Motor Diagnostics

Background:

• The Fourth Industrial Revolution drives the need for intelligent predictive maintenance systems.
• Squirrel-Cage Induction Motors (SCIMs) are critical industrial components requiring reliable fault detection.
• Existing methods may lack the real-time accuracy and efficiency needed for modern industrial environments.

Purpose of the Study:

• To propose and evaluate a hybrid deep learning framework for real-time fault detection in SCIMs.
• To compare the performance of various deep learning architectures, focusing on hybrid models.
• To assess the computational efficiency and suitability of the framework for industrial deployment.

Main Methods:

• Development of a hybrid deep learning framework utilizing eight architectures: CNN-GRU, CNN-LSTM, LSTM, BiLSTM, Stacked LSTM, GRU, CNN, and ANN.
• Training and testing the framework on a dataset of one million samples (healthy and faulty SCIM conditions).
• Analysis of real-time sensor data (torque, speed, power, currents) for fault identification.

Main Results:

• Hybrid models CNN-GRU and CNN-LSTM achieved high classification accuracies (92.57% and 92.27%, respectively).
• These hybrid models outperformed baseline models in precision, recall, and F1-score, effectively capturing spatio-temporal features.
• The framework demonstrated computational efficiency in inference time, latency, and throughput, suitable for real-time applications.

Conclusions:

• The proposed hybrid deep learning framework offers a promising solution for accurate and efficient real-time fault detection in SCIMs.
• The study validates the effectiveness of hybrid models for capturing complex motor dynamics.
• The framework represents a significant step towards intelligent, self-aware industrial systems, with future work focusing on optimization and real-world validation.