A new methodology for multiple incipient fault diagnosis in transmission lines using QTA and Naive Bayes classifier

Transmission lines' monitoring systems are hybrid-dynamical structures produce a large amount of data that renders faults diagnosis difficult. Moreover, multiple faults of unknown nature can occur simultaneously impeding their discrimination. In addition, the feature extraction step constitutes an inherent limitation to the diagnosis of multiple faults. Indeed, a feature extraction algorithm avoiding any masking effect among between different multiple faults is hard to devise. This paper proposes a methodology dedicated to the diagnosis of multiples faults in transmission lines using both and experimental and calculated Leakage Current (LC) signals' harmonics as residuals. Measured data from the real transmission line was used to modelling normal operating mode. Besides, different scenarios, including insulator chains contamination with different types and concentrations of pollutants were modeled by equivalents circuits to generate a multiple faults scenario. LCs from deteriorated insulators were inserted in the transmission line normal operating mode to implement a faulty operating conditions model. Qualitative trend analysis (QTA) was used to generate primitives and by exploiting the leakage current difference between normal and faulty operating conditions to define a features extraction algorithm for the diagnosis of specific fault modes. A Naive Bayes classifier was designed in order to identify the most prominent fault in a multiple faults scenario by means of LC data. The methodology manages to split multiple faults into single faults and reaches a classification accuracy for multiple faults of 95%.