Non-Invasive Induction Motor Fault Diagnosis Using PCA Optimized FFT Features and SVM Classification

Three-phase induction motors are widely utilized in numerous industrial applications due to their reliability and efficiency. However, inadequate maintenance can lead to costly operational failures and downtime. Early detection of faults, especially bearing abrasion faults, is essential to maintain motor performance and reduce expenses. This study proposes a novel abrasion fault detection method utilizing load current measurements, an accessible and cost-effective diagnostic parameter. Fast Fourier Transform (FFT) analysis is applied to extract critical fault indicators from the spectral features of the motor’s load current. To address the challenge of overlapping and distinguishing fault features from healthy operational conditions, Principal Component Analysis (PCA) is employed as a preprocessing step, significantly enhancing diagnostic accuracy. Subsequently, Support Vector Machines (SVM) classify the PCA-extracted features using a Support Vector Machine (SVM) model, further improving accuracy. The proposed PCA and SVM framework introduces two key innovations having automatic PCA based selection of the most discriminative minimal FFT features, and automatic optimization of SVM hyperparameters (C and γ), enabling robust classification even under highly overlapping spectral conditions. The method improves single fault (abrasion fault) diagnostic accuracy from 79.38% (SVM only) to 92.50% by enhancing separability between healthy and faulty spectra and achieves 90.12% accuracy for more challenging multiple-fault (hole and scratch) combinations. This approach provides significant advantages supporting proactive maintenance strategies, enhancing motor reliability, and promoting cost-effectiveness in diverse industrial environments. The results underline the effectiveness and practicality of the proposed methodology, marking it as a valuable advancement in induction motor fault diagnostics.