Multidimensional intelligent diagnosis system based on Support Vector Machine Classifier

Heeding the diagnostic requirements of electromechanical systems applied in automotive and aeronautical sectors, a multidimensional diagnostic system based on Support Vector Machine classifier is presented in this paper. In this context, different stationary and non-stationary speed and torque conditions are taken into account over an experimental actuator, in the same way, different single and combined failures scenarios are analyzed. In order to achieve a proper reliability in the diagnosis process, a multidimensional strategy is proposed: currents and vibrations from an electro-mechanical actuator are acquired. A great deal of features is calculated using statistical parameters from the acquired signals in time and frequency domain. Additionally, advanced time-frequency domain analysis techniques, such as Wavelet Packet Transform and Empirical Mode Decomposition, are used to achieve features which provide information in non-stationary conditions. The feature space dimensionality is analyzed by a feature reduction stage based on Partial Least Squares, which optimizes and reduces the feature set to be used for diagnosis proposes. The classification core is based on Support Vector Machine. Moreover, this work provides a performance comparison between the proposed classification algorithm and others such as Neural Network, k-Nearest Neighbor and Classification Trees. Experimental results are presented to demonstrate the feasibility and diagnostic capability of the proposed system.