AI狀態分類模型應用於轉動機械故障預測與壽命預估

Application of AI status classification for predictive maintenance of rotating machinery

王淑麗(Shu-Li Wang) ； 王朝民(Chau-Min Wang) ； 許（馬豈）(Chi Hsu) ； 周宜德(Yi-Te Chou) ； 張淵仁(Yuan-Jen Chang)

《石油季刊》 58卷2期 (2022/06) Pp. 91-101

摘要

使用數據驅動的預測和健康管理（Prognostics and Health Management, PHM）技術於轉動機械預知保養，相較基於專家知識的物理模型方法，數據驅動方法根據歷史數據建立模型，適合複雜工況與多種參數耦合難以使用物理模型的情況。本文建立PHM的六大步驟：資料擷取、數據前處理、特徵提取、模型訓練、健康評估與圖形化，實現了壓縮機線上即時狀態變化監測。透過機器學習與歷史紀錄相集成，可以建立AI預知保養模式，達到設備自主健康診斷的效果。基於PHM的設備監測模型開發方法，在該方法中，使用統計特徵捕捉設備狀態變化，解決原始訊號對設備健康衰退的低敏感問題，透過Logistic Regression與AI狀態分類模型（SVM演算法）導入，可即時判斷壓縮機系統的狀態轉變和異常原因。證明此方法的效益，以儲運所運轉中的丙烯壓縮機進行測試，研究發現電流、溫度、壓力特徵在狀態變化的監測表現優於振動特徵，目前正推廣至其他煉廠中。結果顯示，基於Logistic Regression演算法的健康指標衰退模型判定係數為0.98，基於SVM演算法的狀態診斷模型準確率為99.8%，二者皆具極佳的預測能力，因此可以得出結論，AI狀態分類模型（SVM）及預測與健康管理技術（PHM）在實際煉廠環境中能有效監測設備操作健康情況。

關鍵字

AI狀態分類 ； 預測與健康管理技術 ； 健康曲線

並列摘要

Compressor is a key and important equipment for oil refinery and petrochemical plants. When an abnormal failure occurs, it will not only cause damage to the compressor, but also cause injure to the surrounding systems, resulting in production interruptions, industrial safety accidents, and a large number of downtime and maintenance costs. In this paper, a data-driven Prognostics and Health Management (PHM) technology is used for predictive maintenance of rotating machinery. Compared with the expert knowledge physical model method, the data-driven method builds models based on historical data, which is suitable for complex working conditions and coupling of multiple parameters which is difficult to use physical models. Through Logistic Regression and AI status classification model (SVM algorithm) import, the status change and the reason for the exception of compressor system can be judged in real time. In order to prove the effectiveness of the proposed method, a test was conducted with an online propylene compressor. The study found that the current, temperature, and pressure characteristics are better than the vibration characteristics in the monitoring of state changes. The results show that the determination coefficient of the health index decline model based on the logistic regression algorithm is 0.98, and the accuracy rate of the state diagnosis model based on the SVM algorithm is 99.8%. Both have excellent predictive capabilities. It can be seen that the AI state classification model (Support Vector Machine, SVM) and Prognostics and Health Management (PHM) can effectively monitor system health degradation in actual industrial environments.

並列關鍵字

SVM ； PHM ； Health Curve

參考文獻

R. Adhikari, & R. K. Agrawal, “An Introductory Study on Time Series Modeling and Forecasting,” LAP Lambert Academic Publishing, Germany, 2013, pp. 18 - 22.

J. Lee, F. Wu, W. Zhao, M. Ghaffari, L. Liao, & D. Siegel, “Prognostics and health management design for rotary machinery systems—Reviews, methodology and applications,” Mechanical Systems and Signal Processing, vol. 42, Issues 1–2, Jan.2014, pp. 314-334.

張淵仁、陳昱翔、許驥、賴政佑、張平昇、謝聲偉，(2020) “線性與非線性迴歸演算法於 PHM 方法預測刀具剩餘可用壽命之比較”，機械新刊，第 47 期，pp.34-44。