整合即時設備監控數據之先進工程資產壽期評估模式研究發展

Abstract

工程資產管理所包含之流程與功能具備有分散環境且協同作業之特性，然目前之工程資產管理仍 高度仰賴自我維護之經驗法則以及週期性例行維修保養，如此之維修保養模式無法預防突發事件之發 生，輕則造成機台效能與生產日數之經濟性損失，重則造成人員傷亡等不可挽回的遺憾。尤其電力產 業為民生必需工業，電力之停擺同時造成大眾生活之不方便，以及產業產能之停擺。因此如何維持電 力產業之運作穩定，尤其對現有設備壽期（生命週期）之正確評估，一直是學界與產業界共同努力之 目標。本產學計畫將深入探究電力產業中最關鍵的大型變壓器設備，了解設備特性以及關鍵參數，探 討相關文獻及其分析方法與限制，與業者共同研擬出整合即時設備狀態數據之變壓器壽期評估模式與 方法。 本產學計畫將分兩方向同步進行，包含變壓器歷史數據之分析以及關鍵參數擷取，以及壽期預測 方法論之探討。於變壓器歷史數據分析上，將深入針對變壓器10 年內之歷史數據進行分析，藉此尋 找直接影響變壓器效能之關鍵參數，包含油壓、油溫、線圈溫度等。於壽期預測方法論之探究上，將 分析目前可用之方法論，包含機率分配、模糊理論、類神經網路、迴歸分析等，以找出最有效率之預 測模型。先期將以邏輯斯迴歸模型結合自迴歸移動平均模型為研究方法論。更將與業者共同進行方法 論之驗證，藉由實際數據之導入，驗證本產學計畫所提之方法論之可行性與準確性。期藉此產學合作 計畫，提供業者具高度參考價值之壽期預測方法，有利於業者發展工程資產管理（預防、維修、替換 規劃）時，可更為精準有效，確保電力持續供應之穩定與信賴，保障大眾之用電方便以及企業之產能 穩定。

The processes and functions of engineering asset management (EAM) are characterized by its distributed and collaborative nature. However, recent EAM still relies on self-maintained experiential rule-bases and periodic maintenance, which can fail to predict and prevent emergency shutdowns that do on occasion lead to disastrous events and cost greater losses. Especially for the electrical power industry, the shortage of power causes inconvenience to the public and potentially reduces the production rate of the industry. Consequently, the means to ensure steady and reliable power is a critical issue to both academia and industry. This research proposes an in-depth study of the most critical equipment in the power supply chain, i.e., the large-scale transformers. The research will identify the key parameters influencing transformer, health, condition, and lifespan. Meanwhile, lifespan forecasting and evaluation methodologies are developed for transformers. This research will be executed by conducting historical data analysis, determining key parameters, and developing forecasting models. For historical data analysis and key parameters determination, the asset condition data covering the recent 10 years will be used to identify the parameters (e.g., oil pressure, oil temperature, and coil temperature) that significantly influence the asset health status and remaining life of transformers. For deriving the forecasting models, state-of-the-art methodologies are reviewed including fuzzy theory, artificial neural networks and regression analysis. The well accepted approaches will be used to construct modified forecasting model for transformers. The Logistic Regression (LR) combined with autoregressive moving averages (ARMA) is proposed in this research as the prediction methodology. Afterward, the modified forecasting model will be verified with real data to fine-tune the accuracy and efficiency of the forecasting model for practical asset management applications. Thus, the equipment supplier (our cooperative company is Fortune Electric Co., Ltd.) and their key customers (Taiwan Power Co.) will benefit from reliable and accurate asset life prediction methodologies, which will contributes to the company’s preventive maintenance and prognostic engineering asset management. With effective and efficient asset management policies in place, the power supply companies can better serve the general public and industries which depend on their daily service operations with little risk of failure.