A Power Transformer Fault Prognosis System Applying Integrated DGA Knowledge Base

以整合式油中氣體分析為基之變壓器故障預測系統

張瑞芬(Amy J.C. Trappey) ； 張宇辰(Juice Y.C. Chang)

《中國機械工程學刊》 38卷2期 (2017/04) Pp. 189-197

摘要

A power transformer is an important facility in the power supply system. If a transformer unexpectedly fails or shuts down, it will cause a severe damage to the entire power supply network. Hence, establishing a good and real-time power transformer fault prognosis system to avoid unexpected break down is a critical issue to a power company. The Dissolved Gas Analysis (DGA) has been the most widely used measures to detect the hazardous gases and predict power transformers potential malfunctions. Currently, there are several approaches to interpret the fault types after certain gases are detected. However, the present methods sometimes provide conflict diagnostic outcomes, which confuse the equipment maintenance engineers. This phenomenon prevents the maintenance crews from repairing the transformers correctly and promptly and often leads to severe problems. Thus, this research integrates five sets of heuristic knowledge rules, including the Doernenburg ratio, Rogers ratio, Duval triangle, the dominant gases, and the phase analysis into an intelligent and integrated fault prognosis system to improve the accuracy and reliability of the prognosis. Further, the fault prognosis results are verified empirically by comparing with actual transformers abnormal data (in total of 65 sets). In summary, this research develops a web-based fault prognostic system, incorporating the integrated DGA knowledge base. The system, as the main module of an intelligent power transformer asset management platform, automatically predicts the fault types using the real-time sensor-collected DGA data.

關鍵字

engineering asset management ； transformer prognosis ； dissolved gas analysis

並列摘要

電力變壓器是電力系統中之重要設施。一旦變壓器無預警故障，將對電網造成嚴重損害。因此，對電力公司而言，建立優異且即時性的變壓器故障預測系統，避免不可預期之故障是重要議題。油中氣體分析法(Dissolved Gas Analysis, DGA)已被廣泛運用來監測變壓器內絕緣物劣化產生之危險氣體與預測變壓器之潛在失效可能。現今，有許多方法可在異常氣體被測得時，解釋其可能故障類型。然而，目前各方法之預測分析常有衝突性存在，導致該設備維修人員誤解與誤判。此現象導致維修人員無法正確且及時進行維修之決策與行動，導致更嚴重之損失。故本研究旨在整合五種啟發式規範，包含：Doernenburg 法、Rogers 法、Duval triangle法、主導氣體法及樣相分析法，建立一整合性變壓器故障預測推論平台，提升故障預測之準確性與可信度，並透過65 組資料進行其正確性之驗證。最後，本研究建構出網站式之故障預測雛型系統，本系統可藉由即時收集油中氣體感應器之資料，進行推論並自動預測相關故障類型。