工作機台在運作時其工作效率會隨時間而慢慢衰退，所以針對機台工作效率的診斷方法非常重要，現今的診斷方法大多只能區分出正常與損壞的運作狀態。本研究運用智慧化之非監督式深度學習方法進行機台工作狀態診斷，主要目的有二：發現滾珠螺桿隨著時間變化之性能衰退，在滾珠螺桿出現性能衰退至門檻值之前給予及時補油修正與補償。由於機器性能衰退難以用人工方式清楚判別，本研究使用了長短期記憶網路編碼器(LSTM-ED)，將滾珠螺桿的振動訊號代入非監督式深度學習網路中，藉由神經網路的資料辨別能力，得到滾珠螺桿在四次長時間重負載運轉下健康指數(Health index)隨運行距離的變化。
本研究分析滾珠螺桿的訊號，是加速規量測上螺帽部位的振動訊號。潤滑條件為試驗開始前注脂12c.c.，且連續運行到150公里前都沒有再注脂。為了要得到滾珠螺桿的健康指數，我們依下列步驟處理：首先是振動訊號前處理(Preprocess)，包含等速段訊號擷取與正規化(Normalization)、零平均處理(Zero-mean process)等。接下來代入深度學習模型做健康診斷，藉由模型的重構誤差得到滾珠螺桿性能衰退曲線，也就是健康指數的變化。除了討論健康指數的下降趨勢外，還有三個方法進行結果分析，找出主導健康指數變化的因素：首先，在試
驗前後都有量測滾珠螺桿的預壓扭矩，利用預壓扭矩與健康指數變化作圖，訂定健康指數起始與結束的標準。第二，利用快速傅立葉轉換，觀察球通頻率及各元件的缺陷頻率判定哪些接觸配對為影響整體系統健康指數的主導因子。最後利用經驗模態分解法將原始訊號由高頻拆分至低頻，將主導磨耗及振動行為之高頻訊號單獨代入模型運算，分析高頻訊號隨運行距離之健康指數變化，用來對照整體健康指數的變化，瞭解高頻振動訊號對整體健康指數的影響。
健康指數能顯示滾珠螺桿在尚未發生機件損壞前的性能衰退現象，尤其能對滾珠螺桿的磨潤狀況有不錯的分析評估能力。而高頻磨耗對整體健康指數的影響也會隨運行距離有不同的變化。本研究也將健康指數量化分析滾珠螺桿的磨潤狀態，發現隨著運行距離的增加，滾珠螺桿的效能衰退速度有變快的趨勢。由於使用的是非監督式的深度學習模型，在處理複雜的振動訊號時，模型能自行學習找出判斷性能的標準，因此能應用於各種振動訊號的分析。使用本研究成果，在機台性能(健康指數)下降至門檻值之前便給予及時的補償與補充潤油，達到智慧化診斷及省時省力的目的。

The efficiency of the machine will degrade with the increase in working time. It is important to diagnose the working state of the machine. This study focuses on intellectual diagnosis for the performance of ball screw system by using an unsupervised deep learning method. There are two purposes of this study：first, figure out the performance change of ball screw with working time, then giving instant lubrication. Because the degradation of ball screw system is difficult to figure out by manual methods. This study can show the degradation by using a unsupervised deep learning method named：Long Short-Term Memory Encoder Decoder (LSTM-ED). With the data analysis ability of the neural network, we can obtain Health Index (HI) of ball screw which shown the degradation while ball screw was working.
This study quantified the health index with an exponential function. We found that the performance declined faster as the accumulated running distance increase. Also, we decomposed raw data into several Intrinsic Mode Functions (IMFs) by using Empirical Mode Decomposition (EMD) and then trained model with each IMF. Finally, we can obtain HI of each frequency region. The health index has a great ability to display the lubrication status of ball screw. We can combine preload torque with health index to explain the overall situation. With the result of this study, we can compensate for machine lubrication immediately before the damage happened.

[1] Igarashi, T., Tokunaga, Y., & Ookuma, K.“Studies on the sound and vibration of a ball screw. Sound characteristics of a ball screw”. JSME International Journal. Ser. 3, Vibration, Control Engineering, Engineering for Industry, 31(4), 732–738, 1988.
[2]Tokunaga, Y., Igarashi, T., & Suglura, T.“Studies on the sound and vibration of a ball screw. 2nd report, Characteristics and generating mechanism of pulse sound”. Transactions of the Japan Society of Mechanical Engineers Series C, 55(520), 2945–2950, 1989.
[3]王建文，“滾珠導螺桿低頻噪音源之鑑別與改善研究”，國立清華大學動力機械工程學系碩士論文，2006。.
[4]洪梓豪，“頻譜分析與碎形理論運用於滾珠螺桿傳動系統之訊號分析及運轉狀況鑑定”，國立成功大學機械工程學系碩士論文，2014。
[5]巫承駿，“運用訊號分析技術探討牙型係數與軌道表面粗糙度對於環狀式類滾珠軸承傳動磨潤行為之研究”，國立成功大學機械工程學系碩士論文，2018。
[6]Pineda, Fernando J. “Generalization of Back-Propagation to Recurrent Neural Networks.” Physical Review Letters, vol. 59, no. 19, Sept. 1987, pp. 2229–2232
[7]G. E. Hinton and R. S. Zemel, “Autoencoders, minimum description length, and Helmholtz free energy, ” in Advances in Neural Information Processing Systems 6 (NIPS'93), (D. Cowan, G. Tesauro, and J. Alspector, eds.), pp. 3-10, Morgan Kaufmann Publishers, Inc., 1994.
[8]S. Hochreiter and J. Schmidhuber, “Long short-term memory. ” , Neural computation, 9(8):1735–1780, 1997.
[9]G. E. Hinton and R. R. Salakhutdinov, “Reducing the Dimensionality of Data with Neural Networks.” Science, vol. 313, no. 5786, pp. 504–507, 2006.
[10]A. Krizhevsky, I. Sutskever, and G. E. Hinton. “Imagenet classification with deep convolutional neural networks. ”, In NIPS, 2012.
[11]N. Srivastava, E. Mansimov, and R. Salakhutdinov. Unsupervised learning of video representations using LSTMs. In ICML, 2015.
[12]A. G. Howard, M. Zhu, B. Chen, D. Kalenichenko, W. Wang, T. Weyand, M. Andreetto, and H. Adam. Mobilenets: Efficient convolutional neural networks for mobile vision applications. arXiv:1704.04861, 2017.
[13]Rashid, M., Grosvenor, R. “Fault Diagnosis of Ball Screw Systems Through Neural Networks”. Comadem 97: 10th International congress and exhibition on condition monitoring and diagnostic engineering management: Espoo,142–151, 1997.
[14]Y. Yang, D. Yu, J. Cheng. “A Fault Diagnosis Approach for Roller Bearing Based on IMF Envelope Spectrum and SVM.” Measurement, vol. 40, no. 9-10, pp. 943–950, 2007.
[15]R. Zhao , R. Yan, Z. Chen, K. Mao, P. Wang, and R. X. Gao, “ Deep learning and its applications to machine health monitoring: A survey,” Submitted to IEEE Trans. Neural Netw. Learn. Syst., 2016. [Online]. Available: http://arxiv.org/abs/1612.07640.
[16]N. Gugulothu, V. TV, P. Malhotra, L. Vig, P. Agarwal, G. Shroff, “ Predicting remaining useful life using time series embeddings based on recurrent neural networks. ”, arXiv:1709.01073, 2017.
[17]P. Malliavin and G. Letac, “傅立葉分析及頻譜分析”。
[18]Cooley, James W., and John W. Tukey. “An Algorithm for the Machine Calculation of Complex Fourier Series.” Mathematics of Computation, vol. 19, no. 90, pp. 297–301 , Jan. 1965.
[19] 李天龍，“以FFT為架構建立之諧波參數分析”，國立中山大學碩士論文，1999。
[20]L. Y. Lu, “Fast Intrinsic Mode Decomposition of Time Series Data with Sawtooth Transform”, ORACLE Technical Report, 2007.
[21]L. Y. Lu, “Fast Intrinsic Mode Decomposition of Time Series Data and filtering of time series dat”, ORACLE Technical Report, 2008.
[22]Z. K. Penga and W. T. Peter and F. L. Chu, “An improved Hilbert–Huang transform and its application in vibration signal analysis”, Journal of Sound and Vibration, vol. 286, pp.187–205.
[23]H. Norden and C. Zheng and L. Steven and M. C. Wu and H. H. Shih and Q. Zheng and N. C. Yen and C. C. Tung and H. H. Liu, “The Empirical Mode Decomposition and the Hilbert Spectrum for Nonlinear and Nonstationary Time Series Analysis”, Proc. R. Soc. Lond. A, vol. 454, pp.903-995, 1998.
[24]G. Gai, “The processing of rotor startup signals based on empirical mode decomposition”, pp.222-235, 2006.
[25]Lee, W. G., Lee, J. W., Hong, M. S., Nam, S.-H., Jeon, Y., & Lee, M. G. “Failure Diagnosis System for a Ball-Screw by Using Vibration Signals”. Shock and Vibration, pp.1-9, 2015.
[26]魏進忠，“單螺帽雙圈滾珠螺桿在預負荷及潤滑作用條件下運動機制與機械性能的理論分析及實驗印證”，國立成功大學機械工程學系博士論文，2003。
[27]黃淳紹，“運用頻譜分析與多重碎形理論於長時間滾珠螺桿系統運轉時訊號分析”，國立成功大學機械工程學系碩士論文，2013。
[28]Agatonovic-Kustrin, S, and R Beresford, “Basic Concepts of Artificial Neural Network (ANN) Modeling and Its Application in Pharmaceutical Research.” Journal of Pharmaceutical and Biomedical Analysis, vol. 22, no. 5, pp.717–727, 2000.
[29]S. Sharma, S. Sharma, A. Athaiya, “Activation Functions in Neural Networks.” International Journal of Engineering Applied Sciences and Technology, vol. 4, no. 2455-2143, pp. 310-316, 2017.
[30]S. Ruder,. An overview of gradient descent optimization algorithms. arXiv preprint arXiv:1609.04747,2016.
[31] V. K. Ojha, A. Abraham, V. Sn´aˇsel, “Metaheuristic Design of Feedforward Neural Networks: A Review of Two Decades of Research.” Engineering Applications of Artificial Intelligence, vol. 60, pp. 97–116, 2017.
[32]D. E. Rumelhart, G. E. Hintont, R. J. Williams, “ Learning representations by back-propagating errors.” Nature, vol .323, pp.533-536, 1986.
[33]F. Seide, G. Li, D. Yu, “Feature Engineering in Context-Dependent Deep Neural Networks for Conversational Speech Transcription.” 2011 IEEE Workshop on Automatic Speech Recognition & Understanding, 2011.
[34]H. W. Lin, M. Tegmark, D. Rolnick, “Why Does Deep and Cheap Learning Work So Well?” Journal of Statistical Physics, vol. 168, no. 6, pp. 1223–1247, 2017.
[35]D. Dong, X. Y. Li, F. Q. Sun, “Life Prediction of Jet Engines Based on LSTM-Recurrent Neural Networks.” 2017 Prognostics and System Health Management Conference (PHM-Harbin), 2017.
[36]W. Juan, H. Gao, “Degradation Assessment for the Ball Screw with Variational Autoencoder and Kernel Density Estimation.” Advances in Mechanical Engineering, vol. 10, no. 9, pp.1-12, 2018.
[37]P. Malhotra, A. Ramakrishnan, G. Anand, L. Vig, P. Agarwal, G. Shroff, “Multi-sensor prognostics using an unsupervised health index based on LSTM Encoder-Decoder.”, arXiv preprint arXiv:1608.06154, 2016.
[38]蔡博盛，“垂直傳動滾珠螺桿傳動性能與注脂條件影響研究”，國立虎尾科技大學機械與機電工程研究所碩士論文，2019。
[39]何慶賀，“運用不同特徵萃取方法結合支撐向量機於滾珠螺桿傳動系統之故障診斷”，國立成功大學機械工程學系碩士論文，2015。