Study of Performance of Several Techniques of Fault Diagnosis for Induction Motors in Steady-State with SVM Learning Algorithms

Diagnosis of faults in induction motors has been accomplished traditionally following a two stage procedure: first, the selected diagnostic quantity is measured and treated with an appropriate signal analysis tool, and, second, the diagnostic signal is evaluated, either by trained personnel or by automatic systems, to determine the presence and the severity of the fault. The techniques used in each of these two stages are selected using performances criteria related to the goal of each stage. On the contrary, this paper presents a coupled approach to the design of the diagnostic system, by performing a comparative study to determine which combination of automated expert system and diagnostic techniques is the optimal one to detect faults of induction machines. This combined system can outperform traditional designs of diagnostic systems with two independent stages. The result of this study proves than a combination of SVM classifier with a method of obtaining fault features through the AC module of the analytical signal, generated with a Hilbert transform using a Welch transform is the best combination, providing a very high ratio of success about 98%.