Feature Selection Scheme Based on Pareto Method for Gearbox Fault Diagnosis

Fault diagnosis based on pattern recognition approach has three main steps viz. feature extraction, sensitive features selection, and classification. The vibration signals acquired from the system under study are processed for feature extraction using different signal processing methods. Followed by feature selection process, classification is performed. The challenge is to find good features that discriminate the different fault conditions of the system, and increase the classification accuracy. This paper proposes the use of Pareto method for optimal feature subset selection from the pool of features. To demonstrate the efficiency and effectiveness of the proposed fault diagnosis scheme, numerical analyses have been performed using the Westland data set. The Westland data set consists of vibration data collected from a US Navy CH-46E helicopter gearbox in healthy and faulty conditions. First, features are extracted from vibration signals in time, spectral, and time-scale domain, then ranked according to three different criterions namely: Fisher score, correlation, and Signal to Noise Ratio (SNR). Afterword, data formed by only the selected features is used as input for the classification problem. The classification task is achieved using Support Vector Machines (SVM) method. The proposed fault diagnosis scheme has shown promising results. Using only the feature subset selected by Pareto method with Fisher criterion, SVMs achieved 100% correct classification.