Machine Learning Based Method for Impedance Estimation and Unbalance Supply Voltage Detection in Induction Motors.

Area of Science:

• Electrical Engineering
• Machine Learning Applications
• Industrial Automation

Background:

• Induction motors (IMs) are vital in industrial settings but susceptible to performance degradation and failure due to unbalanced supply voltages (USV).
• Real-time detection and severity assessment of USV are critical for preventing breakdowns, ensuring operational reliability, and maintaining safety in industrial facilities.
• Motor impedance estimation is fundamental for understanding IM behavior and diagnosing faults, as abnormalities manifest as impedance modifications.

Purpose of the Study:

• To develop a reliable online method for precise detection of unbalanced supply voltages (USV) in induction motors (IMs).
• To propose machine learning (ML) models for estimating IM stator phase impedance and detecting USV conditions.
• To enhance the reliability and safety of industrial operations by accurately assessing USV severity.

Main Methods:

• Utilized voltage symmetrical components to calculate the negative voltage factor (NVF) as an indicator for USV detection.
• Developed two machine learning (ML) models: one for estimating IM stator phase impedance using phase currents, and another for detecting USV using phase currents and voltages.
• Employed a combination of a Regressor Decision Tree (DTR) model and the Short Time Least Squares Prony (STLSP) technique for dataset creation, feature engineering, and model input generation.

Main Results:

• The first ML model successfully estimated IM phase impedances using only phase currents, eliminating the need for additional sensors.
• The second ML model accurately estimated the negative voltage factor (NVF) using phase currents and voltages, enabling precise USV detection.
• The integrated STLSP and DTR approach proved effective in creating datasets and enabling ML models to estimate physical quantities like phase impedance and NVF.

Conclusions:

• The proposed machine learning approach provides a reliable and precise method for online detection of unbalanced supply voltages (USV) in induction motors.
• Estimating IM stator phase impedance and detecting USV using ML models based on current and voltage data enhances motor diagnostics and predictive maintenance.
• The developed technique contributes to improved reliability, safety, and efficiency in industrial applications susceptible to voltage imbalances.