Intelligent on-line transformer monitoring, diagnostics, and decision making

The failure of a large power transformer is an area of significant concern since it can result in millions of dollars in costs as well as a possible interruption of power. Therefore, it is desirable to detect the existence of abnormal or anomalous changes in the transformer's internal condition which could indicate an incipient failure. This thesis proposes and tests an anomaly detection scheme that is based on both spatial and temporal information and ultimately integrates intelligence into the detection process. It was developed from experience gained from field-deployed monitoring systems that currently monitor several 345/115 kV autotransforiners in a large utility transmission system. The scheme is extensively tested with data from the field-deployed systems. Results indicate that nuisance or false alarms are virtually eliminated while the sensitivity to anomalous changes is preserved. Failure data from a 450 MVA shell-form autotransformer are analyzed and results indicate that the proposed detection scheme would have raised an alarm, indicating an abnormal rate of change of the gas residual, 8 days prior to the actual failure. The interest in transformer monitoring has accelerated over the last few years due to structural changes in the electricity industry. This thesis examines the existing and potential incentives for the acquisition, utilization, and commercial development of transformer monitoring systems. Potential benefits are calculated based on capital cost avoidance, environmental cost avoidance, and operational benefits. The cost of a monitoring system is estimated using three scenarios. From a commercial standpoint, both a transformer monitoring system as a product and a fee-based transformer monitoring service appear to be viable business opportunities. Thesis Committee: James L. Kirtley, Jr., Committee Chairman Professor, Electrical Engineering and Computer Science Richard D. Tabors Senior Lecturer, Technology, Management and Policy Senior Research Engineer, Laboratory for Electromagnetic and Electronic Systems Bernard C. Lesieutre Associate Professor, Electrical Engineering and Computer Science