Machine learning based fault classification for improved induction motor performance.

Area of Science:

• Electrical Engineering
• Machine Learning
• Industrial Automation

Background:

• Three-phase induction motors are critical in industrial systems.
• Existing fault detection methods face challenges with computational cost and accuracy.
• Need for efficient and accurate fault classification algorithms is paramount.

Purpose of the Study:

• To design an effective fault classification algorithm for three-phase induction motors.
• To evaluate machine learning algorithms for accurate and efficient fault detection.
• To address limitations of existing computationally exhaustive or low-accuracy methods.

Main Methods:

• Utilized triaxial vibrational current data from motors with bearing and rotor faults.
• Preprocessed data using interpolation for missing values and Synthetic Minority Over-sampling Technique (SMOTE) for imbalance.
• Extracted frequency domain information via Fast Fourier Transform (FFT).
• Applied Principal Component Analysis (PCA) for dimensionality reduction and SelectKBest for feature selection.
• Trained and optimized hyperparameters for Random Forest (RF), Decision Tree (DT), k-nearest neighbors (KNN), and eXtreme Gradient Boosting (XGBoost).

Main Results:

• Decision Tree and Random Forest models achieved 99.95% accuracy.
• k-nearest neighbors showed good performance but with higher computational testing cost.
• eXtreme Gradient Boosting achieved 87.13% accuracy, indicating a need for further optimization.

Conclusions:

• Decision Tree and Random Forest models are highly effective for induction motor fault classification.
• The developed framework provides a robust foundation for future enhancements.
• Further research can incorporate more fault types and feature engineering for improved detection.