Fault Detection of Induction Motors Using Just In Time Classifiers

Induction motors (IMs) are efficient and reliable but may fail like any other machinery. Failure to detect IM faults in a timely manner will result in irreparable damage. In the real world, data collected from IMs is due to a change in status (load change, speed change, etc.), which can lead to concept drift (CD). In this work, stator data is collected under dynamic conditions from a setup IM in the laboratory. We use Just In Time (JIT) classifiers to detect stator fault in the non-stationary environments, which can be adapted to CD and increase classification accuracy, and the Intersection of Confidence Interval-based change detection tests is used to detect the CD. This mechanism is extended using a preprocessing method for multi-dimensional data. To evaluate this method, several scenarios have been used to simulate CD in the forms of abrupt, gradual, recurrent concepts and the appearance of a new class.