Induction Motor Fault Classification Using novel Pca, Svm, And Hybrid Ann

Open Access
Aslam, Mohamed Nadeem
Graduate Program:
Electrical Engineering
Master of Science
Document Type:
Master Thesis
Date of Defense:
July 24, 2015
Committee Members:
  • Peter Idowu, Thesis Advisor
  • Seth Wolpert, Thesis Advisor
  • Scott Van Tonningen, Thesis Advisor
  • Induction Motor
  • Fault Diagnosis
  • SVM
  • ANN
Induction motors have a long history of applications in wide ranging environments and have a proven track record in climate control applications, industrial processes, traction and in various consumer appliances. They are robust, cost-effective, efficient, and amenable to performance control with the availability of advanced drive systems. Due to the ever-expanding share of the application base, more focus are being placed on preventive and predictive maintenance, and early diagnosis of motor faults that ultimately lead to equipment or process downtime and economic repercussions. In this work, three different fault diagnosis algorithms using techniques such as Principal Component Analysis (PCA), Support Vector Machines (SVM) and Artificial Neural Networks (ANN) is presented with simulation results. Principal Component Analysis (PCA) based algorithm analyzes the three-phase current signal, which obtained from the sensors and stored in the host computer, to detect the fault. The current signal processed by the PCA based algorithm are given as input to the SVM and ANN based algorithms that analyze the data and classify the motors into five categories: Healthy, Air Gap Eccentricity (AGE), Bearing Failure (BF), Broken Rotor Bars (BRB), and Damaged Stator Slots (DSS). Through simulation in Matlab, the fault diagnosis and classification system proved to have a good ability to detect and classify faults in an induction motor. Industries have utilized customized hardware tools that tend to be very expensive to diagnose these faults. This work also attempts to build a low cost, non-invasive induction motor fault diagnostic tool using the classical Fast Fourier Transform (FFT) analysis of motor current signature. The core of this tool is a Field Programmable Gate Array (FPGA) processor. The FPGA is programmed to perform FFT of the time domain current signal that is then analyzed to detect the fault. FPGAs are an appropriate choice as they are available at cheap prices and have flexible reconfigurable properties. This property makes FPGAs more adaptable in experiments where different scenarios have to be implemented at different times. They also have higher computing speed than the general-purpose computer based software simulation tools.