Evaluation of machine learning techniques for electro-mechanical system diagnosis

摘要: The application of intelligent algorithms, in electro-mechanical diagnosis systems, is increasing order to reach high reliability and performance ratios critical complex scenarios. In this context, different multidimensional based on machine learning techniques, are presented evaluated an actuator scheme. used methodology includes the acquisition physical magnitudes from system, such as vibrations stator currents, enhance monitoring capabilities. features calculation process statistical time frequency domains features, well time-frequency fault indicators. A reduction stage is, additionally, included compress descriptive information a reduced feature set. After, classification algorithms Support Vector Machines, Neural Network, k-Nearest Neighbors Classification Trees implemented. over inputs corresponding previously learnt classes, generalization capabilities with classes slightly modified experimental test bench analyze suitability each algorithm for kind application.