電動馬達是將電能轉為動能使機器運作的重要物件，馬達的使用相當普遍，生活中許多器具都需要馬達驅動，因此馬達運轉穩定與否便至關重要。許多研究指出，影響馬達運轉損壞的重要元件為軸承，軸承是否損壞是影響馬達運轉的關鍵因素。
本研究建立資料科學的架構，探討馬達內軸承在特定的實驗條件下，分析自運轉初始狀態開始至損壞停止間的振動情形，於振動訊號中找出影響軸承壽命的因子，並預測軸承的剩餘可用壽命，探討使軸承損壞的重要因子，以及後續可行的保養方針。
由於原始資料的數量龐大，無論硬體或分析方法上都較難處理，因此本研究先運用滑動時窗(Sliding Window, SW)的演算法，在保留主要資訊量的前提下能夠大幅降低原始資料的數量，整理成可合理進行後續分析的資料集。接著分為兩大部份，分別於第三章及第四章探討：(i)以運轉前期之特徵預測損壞時間區間及(ii)隨機區段預測剩餘可用壽命。
首先，第一部份運用本研究所提出之資料科學分析架構於前期資料，使用最小平方法(Ordinary Least Squares, OLS)以及分段線性分割方法(Piecewise Linear Segmentation)兩種線性方法萃取前期資料的統計特徵，再以逐步迴歸法(Stepwise Regression)篩選重要特徵，最後以這些重要特徵建構預測模型，使用最小平方法(OLS)以及偏最小平方法(Partial Least Squares Regression, PLS)來預測軸承損壞時間。此外，再提出一隨機過程方法，幾何布朗運動模擬(Geometric Brownian Motion Simulation, GBM)，利用網格搜尋法(Grid Search)找出最佳參數組合，最後比較三種方法之優劣及適用時機。第二部份考量現實情形，測量人員無法保證每次皆是在馬達運轉初始即開始收集數據，可能是任意時段開始收集，因此分析上將多次隨機抽取一小段資料作為訓練樣本資料集，後續以第一部份相同之資料科學分析流程篩選重要特徵，使用最小平方法(OLS)、偏最小平方法(PLS)、支持向量迴歸(Support Vector Regression, SVR)建構預測模型，其中發現隨機區段萃取出之特徵存在異質性(Heteroscedasticity)，若特徵變異之權重已知，以加權最小平方法(Weighted Least Squares, WLS)修正原OLS模型；若權重未知，則使用可行廣義最小平方法(Feasible Generalized Least Squares Regression, FGLS)先估計最佳權重後，再代回迴歸模型修正異質性。最後比較不同迴歸方法之優劣及適用時機，結合第一部份之結果，提供一份綜合性的評比，以避免因軸承損壞或太早更換所帶來之損失。
實證研究是與國立台灣科技大學機械所合作，由台科大機械系機電系統整合實驗室提供之實驗資料進行分析，結果顯示，本研究提出之資料科學架構能有效處理大量振動訊號資料，且分別於前期、隨機時間區段，皆能有不同之對應方法來預測剩餘可用壽命，提供資訊給決策者決定維修時間點。

It is more than twenty-five percent of the global electricity consumed by electric motors. Electric motor is a common equipment of converting the electrical energy into mechanical energy and used in a variety of applications. Thus, the failure of the electric motors will cost enormous capacity loss. In literature, one of the rotating component inside the electric motors called bearings is the main factor of the failure.
In this study, we propose a data science framework embedded with data mining techniques. The purpose is to find out the key factors of the vibration signal data affecting the remaining useful life (RUL) under a certain experiment condition. We build up prediction models to predict the RUL, then make some policies of predictive maintenance according to the results.
Due to the large-scale of the raw dataset, it is difficult for the hardware and methods to analyze. Therefore, we apply an algorithm called sliding window (SW) to reduce the amount of raw data but keep the main information. In our proposed framework, we take the transformed dataset into two parts: (i) Predict the failure time interval by observing the unstable signal revealed in the early stage. (ii) Predict the RUL by taking the random time interval sampling. In part (i), we emphasize the changes in the early stage when motors begin operating. By using ordinary least squares (OLS) and piecewise linear segmentation, we can extract statistical and physical features. Stepwise regression is used to select the important features afterward. Further, we build up two prediction models which are OLS and partial least squares regression (PLS) to predict the failure time interval. Besides, a stochastic process simulation called geometric Brownian motion (GBM) is also added to the comparison of prediction models. Finally, we can compare the three models with different conditions. In part (ii), we consider the situation in reality that people cannot make sure that the data they collect is from the early stage. Thus, we provide a random interval sampling method to maintain the uncertainty in reality. After that, feature extraction and selection processes may be the same as part (i). We use OLS, PLS and support vector regression (SVR) to construct the prediction models. However, heteroscedasticity reveals in the features, we use the weighted least squares regression (WLS) or feasible generalized least squares regression (FGLS) to fix it. Finally, we discuss and compare the results from the two parts, and give an advice for predictive maintenance (PdM) to avoid any loss caused by the failure of the motors as possible as we can.
The empirical study goes with Department of Mechanical Engineering, National Taiwan University of Science and Technology. With their experiment data and our proposed research framework, the results show that the large-scale dataset can be simplified, and we can predict the RUL in the two different conditions with the corresponding techniques. Decision maker can decide the time to maintain/repair or change the components with the results.

Abonyi, J., Feil, B., Nemeth, S., Arva, P. (2005). Modified Gath–Geva clustering for fuzzing segmentation of multivariate time-series, Fuzzy Sets and Systems, Data Mining Special Issue, 149, 39–56.
Asr, M.Y., Ettefagh, M.M., Hassannejad, R., Razavi, S.N. (2017). Diagnosis of combined faults in Rotary Machinery by Non-Naive Bayesian approach. Mechanical Systems and Signal Processing, 85, 56-70.
Bastos, P., Lopes, I., Pires, L. (2014). Application of data mining in a maintenance system for failure prediction. Safety, Reliability and Risk Analysis: Beyond the Horizon – Steenbergen et al. (Eds), Taylor & Francis Group, London, ISBN 978-1-138-00123-7.
Byington, C.S., Watson, M.J., Sheldon, J.S., Swerdon, G.M. (2009). Shaft coupling model-based prognostics enhanced by vibration diagnostics. Insight:Non-Destructive Testing and Condition Monitoring 51, 420–425.
Cui, L.R., Loh, H.T., Xie, M. (2004). Sequential inspection strategy for multiple systems under availability requirement. European Journal of Operational Research 155, 170–177.
Fisher,C. and Baines, N.C. (1988). Multi-sensor condition monitoring systems for gas turbines”, Journal of Condition Monitoring, 1, 57–68.
Fu, T.C. (2011). A review on time series data mining. Engineering Applications of Artificial Intelligence, 24, 164-181.
Hansen, R.J., Hall, D.L., Kurtz, S.K. (1995). A New approach to the challenge of machinery prognostics. Journal of Engineering for Gas Turbines and Power 117, 320–325.
Heng, A., Zhang, S., Tan, C.C., Mathew, J. (2009). Rotating machinery prognostics: State of the art, challenges and opportunities. Mechanical Systems and Signal Processing 23, 724–739.
Hess, A. (2002). Prognostics, from the need to reality-from the fleet users and PHM system designer/developers perspectives. Proceedings of the IEEE Aerospace Conference, Big Sky, MT, USA, vol.6 pp.2791–2797.
Hocking, R. R. (1976). The analysis and selection of variables in linear regression. Biometrics, vol. 32, pp. 1-49.
Jardine, A. K.S., Lin, D., Banjevic, D. (2006). A review on machinery diagnostics and prognostics implementing condition-based maintenance. Mechanical Systems and Signal Processing, 20, 1483-1510.
Kemerait, R.C. (1987). New cepstral approach for prognostic maintenance of cycle machinery. Proceedings of the IEEE Southeast Conference, pp. 256–262.
Keogh, E. (1997a.) A fast and robust method for pattern matching in time series databases. In: Proceedings of the Ninth IEEE International Conference on Tools with Artificial Intelligence, pp.578–584.
Keogh, E., Chakrabarti, K., Pazzani, M., Mehrotra, S. (2001). Dimensionality Reduction for Fast Similarity Search in Large Time Series Databases. Knowledge and Information Systems, 3: 263-286.
Keogh, E., Chu, S., Hart, D., Pazzani, M. (1993). Segmenting Time Series – A Survey and Novel Approach. Published by World Scientific.
Kurfess, Li, Y., Liang, T.R., S.Y. (2000). Stochastic prognostics for rolling element bearings. Mechanical Systems and Signal Processing 14, 747–762.
Lee, C.Y., Chen B.S. (2017). Mutually-exclusive-and-coolectively-exhaustivefeature selection scheme. Applied Soft Computing.
Lee, J., Ni, J., Djurdjanovic, D., Qiu, H., Liao, H. (2006). Intelligent prognostics and emaintenance. Computers in Industry 57, 476–489.
Lee, J., Wu, F., Zhao, W., Ghaffari, M., Liao, L., Siegel, D. (2014). Prognostics and Health Management Design for Rotary Machinery Systems – Reviews, methodology and applications. Mechanical Systems and Signal Processing, 42, 314-334.
Li, Y.G., Nilkitsaranant, P., (2009). Gas turbine performance prognostic for condition based maintenance. Applied Energy 86, 2152–2161.
Lin, J., Keogh, E., Lonardi, S., Chiu, B. (2003). A Symbolic Representation of Time Series, with Implications for Streaming Algorithms. 8th ACM SIGMOD Workshop on Research Issues in Data Mining and Knowledge Discovery, San Diego, CA, USA, June 13.
Meeker, W.Q., Escobar, L.A. (1998). Statistical Methods for Reliability Data. John Wiley & Sons.
Muir, D., Taylor, B. (1997). Oil debris monitoring for aeroderivative gas turbine. ASME Power Division (Publication) PWR32547–553.
O’brien, R.M. (2007). A Caution Regarding Rules of Thumb for Variance Inflation Factors. Quality & Quantity. 41: 673-690.
Orsagh, R., Roemer, M., Sheldon, J. (2004). A comprehensive prognostics approach for predicting gas turbine engine bearing life. Proceedings of ASME Turbo Expo, Vienna, Austria, vol.2, pp.777–785.
Qian, Y., Yan, R., Gao, R. X. (2017). A multi-time scale approach to remaining useful life prediction in rolling bearing. Mechanical Systems and Signal Processing, 83, 549-567.
Rashid, M. M., Gondal, I., Kamruzzaman, J. (2015). Condition Monitoring through Mining Fault Frequency from Machine Vibration Data. IEEE, Neural Networks (IJCNN), 2015 International Joint Conference, Killarney, 12-17 July 2015.
Si, X.S., Wang, W., Hu, C.H., Zhou, D.H. (2011). Remaining useful life estimation – A review on the statistical data driven approaches. European Journal of Operational Research, 213, 1-14.
Smola, A. and Vapnik, V. (1997). Support vector regression machines. Advances in neural information processing systems, vol. 9, pp. 155-161.
Teka, K. H. (2013). PARAMETER ESTIMATION OF THE BLACK-SCHOLES-MERTON MODEL”, submitted in partial fulfillment of the requirements for the degree, KANSAS STATE UNIVERSITY, Manhattan, Kansas.
Tenenhaus, M., Vinzi, V.E., Chatelin,Y.-M., Lauro, C. (2005). PLS path modeling. Computational statistics & data analysis, vol. 48, pp. 159-202.
Tilo Strutz. (2010). Data Fitting and Uncertainty: A Practical Introduction to Weighted Least Squares and Beyond, Vieweg and Teubner , Germany, ISBN:3834810223 9783834810229.
Wang, Z.Q., Hu, C.H., Si, X.S., Zhang, J.X., Wang, H.Y. (2013). A New Remaining Useful Life Prediction Approach for Independent Component Based on the Wiener Process and Bayesian Estimating Paradigm. 25th Chinese Control and Decision Conference (CCDC).
Yang, H., Mathew, J., Ma, L. (2003). Vibration Feature Extraction Techniques for Fault Diagnosis of Rotating Machinery – A Literature Survey. Asia Pacific Vibration Conference, Gold Coast, Australia, 12-14 November.
Zeng, J.-H., Lee, C.-Y., Lee, S.-Y., Kuo, P.-H., Lu R.-B. (2015). “Data Mining for Genome-Wide Association Studies: An Empirical Study of Han-Chinese Patients with Bipolar Disorder”. 2015 Chinese Institute of Industrial Engineers (CIIE) Conference & Annual Meeting, Taichung, Taiwan. (in Chinese).