Fault detection and diagnosis method for three-phase induction motor

• 主要作者: Yousefi, Rozbeh
• 格式: Thesis
• 语言: 英语
• 出版: 2018
• 主题: TK Electrical engineering. Electronics Nuclear engineering
• 在线阅读: http://eprints.utm.my/81684/1/RozbehYousefiPSKE2018.pdf

Induction motors (IM) are critical components in many industrial processes. There is a continually increasing interest in the IMs’ fault diagnosis. The scope of this thesis involves condition monitoring and fault detection of three phase IMs. Different monitoring techniques have been used for fault detection on IMs. Vibration and stator current monitoring have gained privilege in literature and in the industry for fault diagnosis. The performance of the vibration and stator current setups was compared and evaluated. In that perspective, a number of data were captured from different faulty and healthy IMs by vibration and current sensors. The Principal Component Analysis (PCA) was utilized for feature extraction to monitor and classify collected data for finding the faults in IMs. A new method was proposed with the combined use of vibration and current setups for fault detection. It consists of two steps: firstly, the training part with the aim of giving acceleration property (nature of vibration data) to the current features, and secondly the testing part with the aim of excluding the vibration setup from the fault detection algorithm, while the output data have the property of vibration features. The 0-1 loss function was applied to show the accuracy of vibration, current and proposed fault detection method. The PCA classified results showed mixed and unseparated features for the current setup. The vibration setup and the proposed method resulted in substantial classified features. The 0-1 loss function results showed that the vibration setup and the developed method can provide a good level of accuracy. The vibration setup attained the highest accuracy of 98.2% in training and 92% in testing. The proposed method performed well with accuracies of 96.5% in training and 84% in testing. The current setup, however, attained the lowest level of accuracy (66.7% in training and 52% in testing). To assess the performance of the proposed method, the Confusion matrix of classification in NN was utilized. The Confusion matrix showed an accuracy of 95.1% of accuracy and negligible incorrect responses (4.9%), meaning that the proposed fault detection method is reliable with minimum possible errors. These vibration, current and proposed fault detection methods were also evaluated in terms of cost. The proposed method provided an affordable fault detection technique with a high accuracy applicable in various industrial fields.