臺灣博碩士論文加值系統

現今工業發達，於生產中，機器持續損耗老舊，當機器內部零件出現問題時，往往會影響產品品質，造成成本上的大量損耗，為了降低生產成本，如何知道機器損耗程度，並且提前更換老舊設備或零件，成為越來越重要的課題。隨著工業物聯網逐漸崛起，在現今機器設備中，已有加裝眾多感測器，我們可以透過感測器對工業設備進行監測並透過得到的資料進行預防性維護,以此降低生產時的成本,在現今論文中,已有提出一些方法,在探討這個問題,但多數論文的資料並非真實資料,且資料完整度都較高,所以我們想利用真實資料,來驗證這些方法在真實世界的可行度,並將這些結果整理下來並進行比較。

本研究將方法整理下來並利用真實資料實驗出結果，首先資料進行切割並分別取特徵值，透過如相關係數分析、主成分分析、等特殊方法，選擇出重要參數，降低噪音特徵值對後續處理的影響，接著將最後特徵值，透過機器學型模型或是比較的方式，得出機器的剩餘使用生命(RUL)，以此獲得最終答案。

Nowadays, in many industries, with the problems of higher costs in production and the effect on product quality, the topic of when to replace worn parts before operation process stops is becoming more important. As IOT developing, by collecting data from machines, real-time information helps prognostics of machines. In papers, there are some solutions about this topic, but most of the data in them are from simulation or experimental environment, in order to verify the feasibility of the framework from these papers, the data in real world were used to get the final results and show the answer we want to know.

The remaining useful life information was showed by a series of processes. After the implementation of feature extraction on different domain, new features were denoised by Principle Component Analysis, correlation coefficient or other solutions, and final features were got as the input of final step. With some Machine Learning Algorithms, or comparing processes, the RUL prediction shows the health condition of machine, and increases efficiency of maintenance on machines.

摘要 i
ABSTRACT ii
致謝 iii
Ⅰ. 介紹 1
Ⅱ. 相關研究 2
Ⅲ. 架構 3
1. 訊號切割 3
2. 特徵值擷取 4
3. 選取特徵值 5
4. 最終結果取得 6
a. 機器學習模型 6
b. 數值比較 7
Ⅳ. 資料介紹 9
Ⅴ. 實驗結果 10
1. 機器學習模型 10
a. 資料集1 10
b. 資料集2 14
2. 數值比較 16
a. 資料集1 16
b. 資料集2 20
Ⅵ. 結論 25
Ⅶ. 參考文獻 26

[1] Z.Bouzidi, L.Terrissa, A.Lahmadi, N.Zerhouni, S.Ayad, "Health Index-Based Prognostics for Remaining Useful Life Predictions in Electrical Machines", 2017 4th International Conference on Control, Decision and Information Technologies (CoDIT) on, 5-7 April 2017.
[2] F.Yang, M. Habibullah, T.Zhang, Z.Xu, P.Lim, S.Nadarajan,"Health Index-Based Prognostics for Remaining Useful Life Predictions in Electrical Machines", IEEE Trans. Ind. Electron., vol. 63, no. 4, pp. 2633 – 2644, April 2016.
[3] C.Zhang, G.S.Hong, H.Xu, K.C.Tan, J.H.Zhou, H.L.Chan, H.Li, "A data-driven prognostics framework for tool remaining useful life estimation in tool condition monitoring", Emerging Technologies and Factory Automation (ETFA), 2017 22nd IEEE International Conference on, 12-15 Sept. 2017.
[4] S.Laddada, T.Benkedjouh, M.O.Si-Chaib, R.Drai, "A data-driven prognostic approach based on wavelet transform and extreme learning machine", IEEE Electrical Engineering Boumerdes, 2017 5th International Conference on, 29-31 Oct .2017.
[5] 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", 2nd ACM SIGKDD Workshop on ML for PHM, arXiv preprint arXiv:1709.01073 (2017).
[6] J. Deutsch, D. He, "Using Deep Learning-Based Approach to Predict Remaining Useful Life of Rotating Components", IEEE Transactions on Systems, Man, and Cybernetics: Systems, Vol. 48, no. 1, pp. 11 – 20, Jan. 2018.