Broken bars detection in an induction motor by pattern recognition

Abstract

The necessity to insure a continuous and safety operation for induction motors, involves preventive maintenance programs with fault detection techniques. This paper presents the application of a pattern recognition approach in order to detect broken bars in an induction motor. The aim is to identify the operating conditions according to the level of load. For this purpose, only electrical measurements are used. Some time or frequency-dependent parameters, which are relevant for fault detection, are described. They are used to build up a pattern vector. This vector is represented by a point and the operating condition of the induction motor are represented by classes in the space. As the dimension of the space is greater than three, the classes and their evolutions can be visualized after a principal component analysis. Then a decision system, based on this signature and the k-nearest neighbors rule, is proposed.