Application of multiple decision trees for condition monitoring in induction motors

Induction machines (IMs) play a pivotal role in industry and there is a strong demand for their reliable and safe operation. IMs are susceptible to problems such as stator current imbalance and broken bars, usually detected when the equipment is already broken, and sometimes after irreversible damage has occurred. Condition monitoring can significantly reduce maintenance costs and the risk of unexpected failures through the early detection of potential risks. Several techniques are used to classify the condition of machines. This paper presents a new case study on the application of multiple decision trees in the on-line condition monitoring of induction motors. Some advantages can be seen, such as the improved performance of classification systems, in addition to the capacity to explain examples. The database was developed through a simplified mathematical model of the machine, considering the effects caused by asymmetries in the phase impedances of motors. A comparative analysis is performed for individual running (based on the neural networks, k-Nearest neighbor and Naive Bayes) and a multi-classifier (based on the Bagging and Adaboost) approaches. Results demonstrate that the multi-classifier systems obtain better results than those of the individual experiments.