A Noise Invariant Method for Bearing Fault Detection and Diagnosis Using Adapted Local Binary Pattern (ALBP) and Short-Time Fourier Transform (STFT)

Abstract

This study proposes a new method for bearing Fault Detection and Diagnosis (FDD) in Belt Starter Generators (BSGs) using vibration signals. The Adapted Local Binary Pattern (ALBP) method is introduced, and its performance is compared to the conventional Local Binary Pattern (LBP) technique and a Convolutional Neural Network with Support Vector Machine (CNN-SVM) model. Significantly, ALBP demonstrates superior accuracy without significantly increasing computational complexity, outperforming both LBP and the CNN-SVM model. The experimental setup involves a Specialty Motor Testing system, with vibration data collected using a single accelerometer under specific speed and torque conditions. The focus is on detecting and diagnosing bearing faults, such as lubrication and contamination, under various test conditions for Original Equipment Manufacturer (OEM) and After-Market (AM) bearings. ALBP achieves diagnostic accuracy ranging from 99% to 99.8%, representing a significant advancement in bearing FDD. Another novel aspect of this study is the training and testing of the model on separate days. This approach ensures the model’s robustness against data variability and domain shifts, unlike the traditional random data splitting method, which can yield misleadingly high accuracy on a portion of data but fails to generalize. Results show that ALBP achieves an average diagnosis accuracy of 99.4%, compared to 80% for the CNN-SVM model, further highlighting the superior performance of ALBP.

Keywords

• Bearing faults
• fault detection and diagnosis
• local binary patterns
• machine learning
• noise
• signal processing