Vibration-Based Anomaly Detection for Induction Motors Using Machine Learning.

Area of Science:

• Engineering
• Computer Science
• Data Science

Background:

• Predictive maintenance of induction motors is crucial for industrial reliability.
• Fault diagnosis in these systems presents significant challenges, leading to downtime.
• Machine learning offers potential solutions for enhancing diagnostic capabilities.

Purpose of the Study:

• To apply machine learning techniques for improved fault diagnosis in induction motors.
• To evaluate the performance of different algorithms using the Machinery Fault Database (MAFAULDA).
• To investigate feature extraction and optimization strategies for enhanced predictive maintenance.

Main Methods:

• Statistical feature extraction from multivariate time-series data.
• Application of deep neural networks (DNNs), support vector machines (SVMs), and K-nearest neighbors (KNNs).
• Implementation of optimization and oversampling techniques to address data imbalance and improve performance.

Main Results:

• SVM achieved 95.4% accuracy, and KNN achieved 92.8% accuracy.
• Deep neural networks combined with FFT-based autocorrelation features yielded the highest accuracy at 99.7%.
• The models demonstrated high promise for effective fault diagnosis and predictive maintenance.

Conclusions:

• Machine learning, particularly DNNs with specific features, significantly enhances induction motor fault diagnosis.
• The study presents a novel approach for improving the operational health and predictive maintenance of induction motor systems.
• Accurate fault prediction leads to reduced downtime and increased industrial reliability.