Integrating Real-Time Monitoring and Asset Health Prediction for Power Transformer Intelligent Maintenance and Decision Support

Abstract

Large sized transformers are an important part of global power systems and industrial infrastructures. An unexpected failure of a power transformer can cause severe production damage and significant loss throughput the power grid. In order to prevent power facilities from malfunctions and breakdowns, the development of real-time monitoring and health prediction tools are of great interests to both researchers and practitioners. An advanced monitoring tool performs real-time monitoring of key parameters to detect signals of potential failure through data mining techniques and prediction models. Asset managers use the result to develop a suitable maintenance and repair strategy for failure prevention. Principal component analysis (PCA) and back-propagation artificial neural network (BP-ANN) are the algorithms adopted in the research. This chapter utilizes industrial power transformers’ historical data from Taiwan and Australia to train and test the failure prediction models and to verify the proposed methodology. First, PCA detects the conditions of transformers by identifying the state of dissolved gasses. Then, the BP-ANN health prediction model is trained using the key factor values. The integrated engineering asset management database includes nine gases in oil as input factors (N₂, O₂, CO₂, CO, H₂, CH₄, C₂H₄, C₂H₆, and C₂H₂). After applying the principal components algorithm, the research identifies five factors from the Taiwan operational transformer data and six factors from the Australia data. The integrated PCA and BP-ANN fault diagnosis system yields effective and accurate predictions when tested using Taiwan and Australia data. The accuracy rates are much higher (i.e., 92 and 96 % respectively) when compared to previous result of 69 and 75 %. This research is benchmarked against the DGA heuristic approaches including IEEE’s Doernenburg and Rogers and IEC’s Duval Triangle for the experimental fault diagnoses.