Automatic Fault Diagnosis Method of Electrical Machinery and Apparatus by Using Kohonen's Self-Organizing Map

It is objective and practical for an electric machinery and apparatus test to computerize its diagnosis process with fault location identification. It is possible to recognize a fault location from measured voltage and current information by using a Kohonen's self-organizing map (SOM) which only needs a learning process of relation between the locations and their voltage and current characteristics. Diagnosis methods for some types of apparatus are proposed by using the SOM. First a diagnosis method for a three-phase PWM inverter which may have a switch fault is proposed. Twelve types of switch fault patterns are analyzed for its DC and harmonic components which are used as input data for their diagnosis map. The method is also applied to a multi-level, for example, three-level inverter. Secondly a diagnosis method of a linear induction motor which may have a turn or a turn-to-turn fault is proposed. Diagonal impedance for all fault cases under considerations are used for the SOM learning process and they are checked for their fault detections. Finally a diagnosis method for an induction machine which may have a turn fault is proposed. An impedance, which is a small slip-dependent term and expresses the degree of motor asymmetry, is computed from two sets of applied input voltages and currents and used as SOM input data. A simple mathematical model of an induction motor with the reference frame transformation is investigated. The above application examples are all validated for its simulated results by developed Matlab programs.