A Hybrid Intelligent Technique for Induction Motor Condition Monitoring

The objective of this research is to advance the field of condition monitoring and fault diagnosis for induction motors. This involves processing the signals produced by induction motors, classifying the types and estimating the severity of induction motors faults. A typical process of condition monitoring and fault diagnosis for induction motors consists of four steps: data acquisition, signal analysis, fault detection and post-processing. A description of various kinds of faults that can occur in induction motors is presented. The features reflecting faults are usually embedded in transient motor signals. The signal analysis is a very important step in the motor fault diagnosis process, which is to extract features which are related to specific fault modes. The signal analysis methods available in feature extraction for motor signals are discussed. The wavelet packet decomposition results consist of the time-frequency representation of a signal in the same time, which is inherently suited to the transient events in the motor fault signals. The wavelet packet transform-based analysis method is proposed to extract the features of motor signals. Fault detection has to establish a relationship between the motor symptoms and the condition. Classifying motor condition and estimating the severity of faults from the motor signals have never been easy tasks and they are affected by many factors. AI techniques, such as expert system (ES), fuzzy logic system (FLS), artificial neural network (ANN) and support vector machine (SVM), have been applied in fault diagnosis of very complex system, where accurate mathematical models are difficult to be built. These techniques use association, reasoning and decision making processes as would the human brain in solving diagnostic problems. ANN is a computation and information processing method that mimics the process found in biological neurons. But when ANN-based methods are used for fault diagnosis, local minimums caused by the traditional training algorithms often result in large approximation error that may destroy their reliability.