On Fault Detection, Diagnosis and Monitoring for Induction Motors

Abstract: Abstract With progress in process of control systems, the benefits to various industrial segments such as chemical, petrochemical, cement, steel, power and desalination industries have been enormous. Recently, Fault Detection and Isolation (FDI) became an important problem in many engineering applications and continues to be an active area of research in the industrial and the control community. Moreover, robust fault detection and diagnosis are also used to monitor and assess the systems safety. On the other hand, the main objective of the on-line FDI processes is to detect and isolate the fault in its early stages, enhance of reconfiguration control, reduce maintenance costs, increase the availability and reliability, prevent unscheduled downtimes, and improved operational efficiency of the systems. Therefore, uncontrollable faults in the processes may cause considerable economic losses, degrade the quality of the process performance, or even cause serious damage to human life or health, as well as the environment. In this thesis, multiple methods and different approaches have been established and evaluated successfully, in order to detect and diagnose the faults of induction motors (IMs). The aim of this thesis is to present novel fault detection and isolation methods for the case of induction machines that would have the merit to be implemented online and being characterized by specific novel capabilities, when compared with the existing techniques. More specifically both the cases of model based and modeless (model free) fault detection and isolation methods will be considered. The proposed methods have been based on: a) Set Membership Identification, Uncertainty Bounds Violation and a minimum uncertainty boundary violation detection schemes, for multiple cases of broken bars under different load conditions and short circuits in stator windings detection having the merit of exact and fast fault detection an easy straight forward fault isolation and capabilities, b) model based Support Vector Classification for the detection of broken bars under full load conditions, using features based on the spectral analysis of the steady state stator's current, without the need of training steps (an expensive, time consuming and often practically infeasible task) and existing of a priori data sets of healthy and faulty cases, c) fault classification based on robust linear discrimination scheme in the model free case and based on novel extracted features for both short circuit and broken bar, and d) fault detection based on Principal Component Analysis (PCA) fault/anomaly detector in time domain for detecting broken rotor bars under full load conditions, e) fault classification technique for bearings based on a novel Minimum Volume Ellipsoid method for feature extraction. One of additional major contributions of this thesis is the fact that especially for the cases of broken bars, and short circuit in stator windings. All the proposed methodologies have been extensively evaluated in multiple experiments and in multiple payloads and thus it has been realistically demonstrated the merits of all the proposed fault detection and isolation schemes. Furthermore, the obtained results suggest that these novel representations can be used within condition monitoring systems.

  This dissertation MIGHT be available in PDF-format. Check this page to see if it is available for download.