預測保養是基於人工智慧演算法去試圖預測設備可能發生故障的時間以便提前進行更換或維修。在國巨的積層陶瓷電容製程中會經過一道燒結程序，該程序所使用之推板式窯爐必須控制在恆溫使得陶瓷順利結晶，而當其中的二矽化鉬加熱元件無預警斷裂便會造成溫度驟降而影響成品率。本研究期望建立機器學習模型來預測加熱元件的剩餘使用壽命，找到最佳的修復時間。但是，預測保養應用於實際工廠中最大的挑戰就是資料收集不易，難以取得足夠的設備損壞資料供機器學習模型訓練，因此我們另外訓練生成對抗網路來學習生成大量類似於實際數據的合成數據，並以合成數據訓練長短期記憶模型及支持向量迴歸模型，兩模型的平均預測能力都很強，MAE皆趨近於0，但由RMSE及R2的結果顯示SVR的擬合效果更出色，R2可達0.634。根據此預測結果便能在加熱元件斷裂約20天前就進行維護工作，除了能避免無預警的斷裂，也免去定期更換而產生不必要的浪費。

Predictive Maintenance (PdM) based on machine learning algorithm detects possible machine failure in advance to avoid sudden breakdown. This research is a case study for real-life industrial heating process. Molybdenum disilicide (MoSi2) is a common heating material used in a pusher kiln due to its high thermal resistance. Our purpose is to construct a machine learning model that can predict the remaining useful life of MoSi2 so that sudden breakdown of heating elements can be avoided. However, insufficient failure experience is a common challenge when dealing with predictive maintenance in real-life scenario. Therefore, Generative Adversarial Network (GAN) is applied to generate data which is similar to the real one but with enough failure experiences so that it is feasible to be used as training data. As for regression model, Long Short-term Memory (LSTM) and Support Vector Regression (SVR) are selected and compared the performance. Determine the best model for predicting remaining useful life of MoSi2.

[1] C.Groba, S.Cech, F.Rosenthal, A.Gössling, Architecture of a predictive maintenance framework, Proc. - 6th Int. Conf. Comput. Inf. Syst. Ind. Manag. Appl. CISIM 2007. (2007) 59–64. https://doi.org/10.1109/CISIM.2007.14.
[2] A.K.S.Jardine, D.Lin, D.Banjevic, A review on machinery diagnostics and prognostics implementing condition-based maintenance, Mech. Syst. Signal Process. 20 (2006) 1483–1510. https://doi.org/10.1016/j.ymssp.2005.09.012.
[3] T.Zonta, C.A.daCosta, R.daRosa Righi, M.J.deLima, E.S.daTrindade, G.P.Li, Predictive maintenance in the Industry 4.0: A systematic literature review, Comput. Ind. Eng. 150 (2020) 106889. https://doi.org/10.1016/j.cie.2020.106889.
[4] O.Surucu, S.A.Gadsden, J.Yawney, Condition Monitoring using Machine Learning: A Review of Theory, Applications, and Recent Advances, Expert Syst. Appl. 221 (2023) 119738. https://doi.org/10.1016/j.eswa.2023.119738.
[5] A.Baru, Three Ways to Estimate Remaining Useful Life for Predictive Maintenance, MathWorks. (2018).
[6] I.Goodfellow, J.Pouget-Abadie, M.Mirza, B.Xu, D.Warde-Farley, S.Ozair, A.Courville, Y.Bengio, Generative adversarial networks, Commun. ACM. 63 (2020) 139–144. https://doi.org/10.1145/3422622.
[7] H.Lu, V.Barzegar, V.P.Nemani, C.Hu, S.Laflamme, A.T.Zimmerman, GAN-LSTM Predictor for Failure Prognostics of Rolling Element Bearings, in: 2021 IEEE Int. Conf. Progn. Heal. Manag. ICPHM 2021, Institute of Electrical and Electronics Engineers Inc., 2021. https://doi.org/10.1109/ICPHM51084.2021.9486650.
[8] Wattco, Electric heating elements, (2023). https://www.wattco.com/electric-heating-elements/.
[9] C.Zhe Ming, Using Activation Functions in Neural Networks, (2022).
[10] D.P.Kingma, J.L.Ba, Adam: A method for stochastic optimization, 3rd Int. Conf. Learn. Represent. ICLR 2015 - Conf. Track Proc. (2015) 1–15.
[11] ＡＩ專欄 - 最新消息 - 國立政治大學人工智慧跨域研究中心, (n.d.).
[12] Understanding LSTM Networks -- colah’s blog, (2016) 1–8. http://colah.github.io/posts/2015-08-Understanding-LSTMs/.
[13] 支持向量回归（SVR）的详细介绍以及推导算法_与秃顶斗争！的博客-CSDN博客, (n.d.).
[14] A.Khosravi, S.Nahavandi, D.Creighton, A.F.Atiya, Comprehensive review of neural network-based prediction intervals and new advances, IEEE Trans. Neural Networks. 22 (2011) 1341–1356. https://doi.org/10.1109/TNN.2011.2162110.
[15] Confidence intervals for LOWESS models in python - James Brennan, (n.d.).
[16] N.Li, Y.Lei, J.Lin, S.X.Ding, An Improved Exponential Model for Predicting Remaining Useful Life of Rolling Element Bearings, IEEE Trans. Ind. Electron. 62 (2015) 7762–7773. https://doi.org/10.1109/TIE.2015.2455055.
[17] D.Hazra, Y.C.Byun, Synsiggan: Generative adversarial networks for synthetic biomedical signal generation, Biology (Basel). 9 (2020) 1–20. https://doi.org/10.3390/biology9120441.
[18] E.Brophy, Z.Wang, Q.She, T.Ward, Generative Adversarial Networks in Time Series: A Systematic Literature Review, ACM Comput. Surv. 55 (2023). https://doi.org/10.1145/3559540.