Development of a Real-Time Fault Detection Model for Hydraulic Brake Systems Using Vibration Analysis and Machine Learning with Wavelet Features

Abstract

Advancements in automotive technology have led to a steady increase in vehicle usage, making it crucial to monitor various control systems, particularly the brake system. This study focuses on using feature-based analysis and various algorithmic approaches to diagnose faults in the hydraulic braking system of light motor vehicles (LMVs) through vibration signals. Vibration signals were captured during different braking scenarios, including both normal and defective conditions, using a vibration transducer and a LabVIEW graphical program. These signals were then analyzed using computerized methods to extract histogram, statistical, and wavelet features. The extracted features were further analyzed using machine learning classifiers, including Continuous High-resolution Image Reconstruction using Patch priors (CHIRP), Forest by Penalizing Attributes (ForestPA), Systematically Developed Forest (SysFor), and HT. The accuracy of each algorithm was compared, and the wavelet-ForestPA model demonstrated the highest classification accuracy, achieving 99.82% for the hydraulic brake system. Based on this model, a novel online prediction system was developed using a LabVIEW graphical interface to display the brake system's condition in real-Time. This machine learning-based model is vital for providing instant notifications about the brake system's status, enhancing safety and reliability.