Combining learning methods and time-scale analysis for defect diagnosis of a tramway guiding system

This paper presents a diagnosis system for detecting tramway rollers defects. First, the continuous wavelet transform is applied on vibration signals measured by specific accelerometers. Then, the Singular Values Decomposition (SVD) is applied on the time-scale representations to extract a set of singular values as classification features. The resulting multi-class classification problem is decomposed into several 2-class sub-problems. The predicted probabilities are coupled using a pairwise coupling method. Empirical results demonstrate the efficiency and robustness of the overall diagnosis system on measurement data.