Monitoring and Diagnosing Unit Transformers using Advanced Pattern Recognition and Case-based Reasoning

摘要

This paper describes a new approach that is field-ready for monitoring and diagnosing faults on large electric transformers based on a case study at a large Southeastern utility electric power generating station. Unit transformers (“UT”) are amongst the most critical components in electric power generating stations and lead-time for repair and/or replacement can range from several months to a year or more. Monitoring and diagnosing faults related to transformer degradation mechanisms has become more important to the power generation industry as many UT’s are approaching end-of-life. Combinations of indications have traditionally been processed through expert systems in order to classify the likely faults using fuzzy logic and/or rules. Advanced pattern recognition (APR) extracts features in available online monitoring parameters to provide earlier warning indications. Online monitoring alone, however, cannot provide the complete set of information that a transformer subject matter expert needs for diagnosing transformer health. Periodic tests, laboratory analysis, and predictive maintenance data supplement online monitoring parameters to assist in component health assessment. Diagnostic tools, such as case-based reasoning, can assimilate features from different sources of information (online and offline) in order to provide a holistic approach to transformer health assessment. This approach works best with an APR solution, such as the one used to develop this paper – SureSense by Expert Microsystems, that can simultaneously run multiple predictive models, can integrate online and offline diagnostic models, plus has built-in prognostic or remaining useful life capabilities.