在機台加工中，需要正確的預測刀具磨耗時間點以防止加工工件損壞。而其中，預測刀具磨耗主要有使用圖像檢測跟訊號檢測兩種。圖像檢測需要使用攝影機，且需要清晰、無髒汙的圖像才能預測。然而，由於加工環境因素，攝影機架設及取像皆極為困難。因此，本文基於長短期記憶網路(Long Short-Tern Memory Network)，預測刀具訊號之上下界。其架構包含訊號處理，長短期記憶網路訓練，計算刀具訊號上下界三個部分。若訊號超過我們預測之上下界，代表應注意刀具狀態，或需立即更換刀具。其中，長短期記憶網路訓練時間約3小時。訓練完成的網路模型所繪出之上下界可隨著訊號位置及變化量做調整，由此也可證實模型預測出之上下界對於訊號磨耗預測有一定的效用。另外，在訊號資料處理中，本文提出了分割(segment)訊號，並算出每段分割後之訊號內的變化量，以作為預測訊號上下界範圍之標準，並且能加速模型訓練以達到業界上對於運算速度之需求。

In machine processing we need to predict anomaly time precisely to prevent work piece broken. Among them, there are two main methods for tool anomaly predication. One is image detection, other is signal detection. Image prediction need to use camera, and the picture is required to clear and no noise. However, due to the working environment, it is very hard to setup camera and get the clear picture. Therefore, this paper will base on long short-tern memory network to predict the upper and lower boundary. The structure contains three parts, signal preprocessing, training of long short-tern memory network and calculate upper and lower boundary. If the signal exceeds to our boundary, it means the tool situation should be warned, or even change the tool instantly. Besides, the training time for long short-tern memory network is about 3 hours. The boundary that drawn from trained network model will adjust by the signal location and variety, prove that the method is functional. For signal preprocessing, the paper uses segmentation to calculate the amount of change for each segment to give the reference for signal upper and lower boundary. In addition, segmentation can also speed up our training period to meet the industrial require.

[1] Chandola, Varun, Arindam Banerjee, and Vipin Kumar. "Anomaly detection: A survey.", ACM computing surveys (CSUR) 41.3, pp. 15, 2009
[2] Babu, Giduthuri Sateesh, Peilin Zhao, and Xiao-Li Li. "Deep convolutional neural network based regression approach for estimation of remaining useful life.", International conference on database systems for advanced applications. Springer, Cham, 2016.
[3] Shao, Haidong, et al. "A novel deep autoencoder feature learning method for rotating machinery fault diagnosis.", Mechanical Systems and Signal Processing 95, pp. 187-204, 2017.
[4] Zhao, Rui, et al. "Deep learning and its applications to machine health monitoring." Mechanical Systems and Signal Processing 115, pp. 213-237, 2019.
[5] Guo, Liang, et al. "Multifeatures fusion and nonlinear dimension reduction for intelligent bearing condition monitoring.", Shock and Vibration, 2016.
[6] Junbo, Tan, et al. "Fault diagnosis method study in roller bearing based on wavelet transform and stacked auto-encoder.", The 27th Chinese Control and Decision Conference (2015 CCDC). IEEE, 2015.
[7] Wang, Peng, Ruqiang Yan, and Robert X. Gao. "Virtualization and deep recognition for system fault classification.", Journal of Manufacturing Systems 44, pp. 310-316, 2017.
[8] Babu, Giduthuri Sateesh, Peilin Zhao, and Xiao-Li Li. "Deep convolutional neural network based regression approach for estimation of remaining useful life.", International conference on database systems for advanced applications. Springer, Cham, 2016.
[9] D. Verstraete, A. Ferrada, E.L. Droguett, V. Meruane, M. Modarres, Deep learning enabled fault diagnosis using time-frequency image analysis of rolling element bearings, Shock and Vibration, AtricleID 506765, 2017.
[10] Verstraete, David, et al. "Deep learning enabled fault diagnosis using time-frequency image analysis of rolling element bearings.", Shock and Vibration, 2017
[11] Malhotra, Pankaj, et al. "Multi-sensor prognostics using an unsupervised health index based on lstm encoder-decoder.", arXiv preprint arXiv:1608.06154, 2016