Neural Network Based Detection Technique for Eccentricity Fault in LSPMS Motors

Early detection of different faults will assist motor operation and prevent it from complete damage. Condition monitoring is extremely important to monitor the motor status and isolate it under failure conditions. This paper will propose a MATLAB (R) mathematical based neural network model for early-detection of static eccentricity fault in line start permanent magnet synchronous (LSPMS) Motors. Under different combinations of fault-load conditions, the motor will be simulated to specify the characteristic of this fault. The line current will be utilized to extract the distinct principal components. The efficient selected components will be used as the input of neural network to recognize the percentage of occurrence as well as the fault's severity. The network will be trained over a specified range of static eccentricity degrees. Besides, it will be tested for unseen cases to qualify the effectiveness of the trained neural network. The testing results show a detection accuracy in a range between 95-98%.