Data fusion for multiple mechanical fault diagnosis in induction motors at variable operating conditions Conference Paper uri icon

abstract

  • In this paper, data fusion based on multi-class support vector machines (SVM) is presented to detect and isolate three mechanical faults in induction motors. First, we construct the feature vector by using signatures created from frequency-domain characteristics. These signatures are obtained from mechanical vibration and line currents measurements. Then, the feature vector is used to feed SVM%27s to classify different motor conditions (normal, misalignment, unbalanced and bearing fault). Different experiments using a three phase induction motor were performed under variable operational conditions (motor speeds and load torque scenarios) in order to acquire training and validation data. The identified optimal parameters of the SVM%27s are reported. The SVM%27s are studied with two types of kernel functions, the radial basis and the polynomial functions. Data acquisition, feature extraction and SVM%27s computation were implemented by using LabView programming language. The experimental results show the effectiveness of the proposed approach in diagnosing the studied mechanical faults at different speeds and load conditions. In these experimental tests, the worst-case accuracy of the proposed method was 97.1%25. © 2010 IEEE.
  • In this paper, data fusion based on multi-class support vector machines (SVM) is presented to detect and isolate three mechanical faults in induction motors. First, we construct the feature vector by using signatures created from frequency-domain characteristics. These signatures are obtained from mechanical vibration and line currents measurements. Then, the feature vector is used to feed SVM's to classify different motor conditions (normal, misalignment, unbalanced and bearing fault). Different experiments using a three phase induction motor were performed under variable operational conditions (motor speeds and load torque scenarios) in order to acquire training and validation data. The identified optimal parameters of the SVM's are reported. The SVM's are studied with two types of kernel functions, the radial basis and the polynomial functions. Data acquisition, feature extraction and SVM's computation were implemented by using LabView programming language. The experimental results show the effectiveness of the proposed approach in diagnosing the studied mechanical faults at different speeds and load conditions. In these experimental tests, the worst-case accuracy of the proposed method was 97.1%25. © 2010 IEEE.

publication date

  • 2010-01-01