Use this URL to cite or link to this record in EThOS:
Title: The design and implementation of a statistical pattern recognition system for induction machine condition monitoring
Author: Hatzipantelis, Eleftherios
ISNI:       0000 0001 3443 9722
Awarding Body: University of Aberdeen
Current Institution: University of Aberdeen
Date of Award: 1995
Availability of Full Text:
Access through EThOS:
Full text unavailable from EThOS. Please try the link below.
Access through Institution:
Automated fault diagnosis in induction machines is a difficult task and normally requires background information of electrical machines. Here a different methodology to the condition monitoring problem is devised. The approach is based entirely on Digital Signal Processing (DSP) and Statistical Pattern Recognition (PR). Description of machine conditions is extracted from empirical data. The main tasks that must be carried out by a PR-based condition monitoring system are: condition identification, knowledge reinforcement and knowledge creation for previously unseen conditions. The DSP operations are employed to quickly isolate sensor faults and to remove noise using data acquired from a single channel. DSP transformations may seem promising in making the monitoring system portable. Most importantly, they can compensate for operational changes in the machine. These changes affect the supply line currents and the primary signal quantities to be measured, i.e. the current and the axial leakage flux. The data which is input to the statistical monitoring system may be transformed, in the form of features, or remain unaltered. The system exploits the statistical properties of the feature vectors. The particular features, namely the LAR coefficients, convey short-term, high-resolution spectral information. For a long record, the feature vector sequence may provide information about changes in the record spectral characteristics, with time. Many induction machine processes are stationary and they can be properly be dealt with by a simple statistical classifier, e.g. a Gaussian model. For nonstationary processes, the system may employ a more comprehensive tool, namely the Hidden Markov Model. which may track the changing behaviour of the process in question. Initially a limited number of machine conditions are available to the process engineer. By identifying their boundaries, new faulty conditions could be signalled for and adopted into the database.
Supervisor: Not available Sponsor: Not available
Qualification Name: Thesis (Ph.D.) Qualification Level: Doctoral
EThOS ID:  DOI: Not available
Keywords: Automatic fault detection Pattern recognition systems Pattern perception Image processing Computer integrated manufacturing systems