Power Transformers Insulation Faults Identification With DGA: A Molecular Dynamics-Assisted Method

Abstract

The accurate and effective identification of power transformer insulation fault is critical in implementing corrective actions and preventing problem reoccurrence. While the dissolved gas analysis (DGA) forms the basis for fault identification, certain challenges still remain, including the absence of clear theoretical principles, conflict results, and the oversight of multiple faults. This article addresses these issues by employing molecular dynamics (MD) simulations to investigate the decomposition of mineral oil under various insulation fault conditions. Identification is eventually achieved by a clustering-based method with MD results as initial centers. To achieve this, the molecular model of transformer mineral oil is first constructed, and its decomposition mechanism and results are studied under different insulation fault conditions. Afterward, based on the MD results, certain ratios between the decomposed gases are selected and calculated, which are utilized as the initial centers of the clustering. Finally, the fault identification can be achieved by substituting the DGA data into the established clustering classifier. The proposed method is tested with both the IEC TC 10 database and the local DGA dataset. The results show a respective 83.4% and 89% success rate in identifying single or multiple faults, verifying the effectiveness of the proposed method.