The use of intelligent machine health monitoring schemes is increasing in critical applications as traction tasks in the transport sector. The high diagnosis capability and reliability required in these systems are being supported by intelligent classification algorithms. These classifiers use calculated features from the system to perform the diagnosis. In this context, different features calculation methods can be applied to characterize the system condition obtaining different classification results. The aim of this work is based on diagnosis capabilities evaluation of the main features calculation methods: statistical features from time, statistical features from frequency, time-frequency distributions and signal decomposition techniques. The features capabilities are quantitatively evaluated by two parameters: the classification accuracy and the discriminant coefficient. Experimental results are obtained from an electromechanical actuator under different diagnosis requirements: from single fault to combined faults detection under stationary and non-stationary speed and torque conditions.

Peer Reviewed