State classification of transformers using nonlinear dynamic analysis and Hidden Markov models

Abstract

The mechanical state of windings is a key factor affecting the reliability and safety of operating power transformers. The vibration based technique provides an alternative and non-intrusive way to diagnose transformers. In this paper, we propose a novel approach for winding condition diagnosis based on analyzing the nonlinear relationship between the electromagnetic force and the forced winding vibration. First, the basic theory of the winding vibration model is reviewed, and the influence of the nonlinear vibration is also discussed. Next, the nonlinear feature is extracted using cross recurrence plot analysis, and the feature sequence consists of the nonlinear ratios under different loads are obtained. Finally, Hidden Markov models are employed to classify the winding conditions. During the laboratory experiment, the winding structures under typical conditions were simulated, including normal, degraded and anomalous classes. The HMM-based classifiers are trained and tested, whose results are compared with those of other approaches such as artificial neural network and naive Bayes classifier. In the end, three field transformers are presented to validate the trained model, and it is proved that the proposed method is effective for winding condition diagnosis.