Vibration based brake health monitoring using wavelet features: A machine learning approach

In this study, the application of wavelets has been investigated for diagnosing the faults on a hydraulic brake system of a light motor vehicle using the vibration signals acquired from a brake test setup through a piezoelectric type accelerometer. An efficient brake system should provide reliable and effective performance in order to ensure safety. If it is not properly monitored, it may lead to a serious catastrophic effect such as accidents, frequent breakdown, etc. Hence, the brake system needs to be monitored continuously. The condition of the brake components and the vibration signals are interrelated. If the failure starts progressing, the vibration magnitude will also progress. Analyzing the vibration signals under the various fault conditions is the key process in fault diagnosis. In recent decades wavelets have been focused on in many fault diagnosis studies as the wavelets decompose the complex information into simple form with high precision for further analysis. The wavelet features were extracted in order to retrieve the information from the vibration signals using discrete wavelet transform. From that discretized signal under each fault condition, the relevant features were extracted and feature selection was carried out. The selected features were then classified using a set of machine learning classifiers such as best first tree (pre-pruning, post-pruning, and unpruned), Hoeffding tree (HT), support vector machine, and neural network. The classification accuracies of all the algorithms were compared and discussed. Among the considered classifier model, the HT model produced a better classification accuracy as 99.45% for the hydraulic brake fault diagnosis.