Image-Based Approach Applied to Load Torque Estimation in Three-Phase Induction Motors.

Area of Science:

• Electrical Engineering
• Artificial Intelligence
• Machine Learning

Background:

• Accurate load torque estimation is crucial for controlling three-phase induction motors.
• Traditional methods often require detailed motor parameters, limiting their applicability.
• Non-invasive sensing techniques are desirable for real-time motor diagnostics.

Purpose of the Study:

• To develop a novel, parameter-free method for estimating load torque in induction motors.
• To leverage air gap flux measurements and convolutional neural networks (CNNs) for torque prediction.
• To validate the proposed method experimentally on a squirrel cage induction machine.

Main Methods:

• Measuring air gap flux density using Hall effect sensors within the motor.
• Converting time-domain flux signals into grayscale images.
• Utilizing an inception-type CNN for image-based load torque estimation.
• Employing three optimizers for model training and validation.

Main Results:

• The proposed method successfully estimated load torque across a wide operational range (1.5% to 93.9% of rated load).
• Four model configurations achieved a mean absolute percentage error (MAPE) of 3.7% or less.
• Specific models achieved MAPE as low as 3.0% (40x50 pixels), 3.5% (40x25 pixels), and 3.3% (50x80 pixels).
• Experimental validation was performed on a 7.5 kW squirrel cage induction machine.

Conclusions:

• The developed method offers an effective, parameter-free approach for load torque estimation in induction motors.
• Converting flux signals to images for CNN analysis is a viable strategy for motor diagnostics.
• The technique demonstrates high accuracy and broad applicability across various load conditions.