Machine Learning-Based Experimental Fault Diagnosis in Induction Motors Using Grey Level Co-occurrence Matrix and Local Binary Patterns

Abstract

PurposeInduction machines are widely used across industrial sectors due to their robustness and efficiency. However, unexpected failures such as broken rotor bars or bearing defects can lead to severe downtime and costly repairs. To address the challenge of early and accurate fault detection, this study explores an acoustic signal-based approach using a combination of statistical texture analysis and Artificial Intelligence (AI).MethodsAcoustic signals collected from motors operating under various faulty and healthy conditions were analyzed. Feature extraction was performed using a fusion of Grey Level Co-occurrence Matrix (GLCM) with Local Binary Patterns (LBP), which together characterize both spatial dependencies and local texture variations in the signal representation. Three classifiers were subsequently trained and assessed using the collected features: multiclass Support Vector Machines (SVM), Artificial Neural Network (ANN), and K-Nearest Neighbors (KNN).ResultsAccording to experimental findings, the KNN classifier achieved high classification performance, while both ANN and SVM achieved high accuracy in distinguishing between different fault types, under controlled laboratory conditions.ConclusionThis approach demonstrates that acoustic monitoring, combined with advanced signal processing and machine learning, offers a powerful, non-invasive solution for real-time induction motor fault diagnosis, with potential benefits for predictive maintenance and improved operational reliability in industrial systems.