表面黏著技術為近年來電子工業能夠蓬勃發展之主要原因之一，透過回焊爐中不同溫區的溫度設定，使印刷電路板(PCB)經過回焊爐後形成特定溫度曲線，是製程中品質好壞重要的指標之一。
本研究分析過期生產PCB資料，對不同機種機台規格與回焊爐區位溫度設定，建立一雙向長短期記憶神經網路模型以預測溫度曲線。模型使用之序列對序列(Sequence-to-Sequence, Seq2Seq)架構因為個案公司資料型態，需使用短輸入序列預測長輸出序列。本研究放大輸入序列長度並加上位置編碼(Positional encoding)後，透過動態移動視窗注意力機制，對每個輸出序列的時間點(Time step)都自動決定其注意力機制的視窗中心點與視窗範圍，以態移動視窗注意力機制雙向長短期記憶神經網路模型預測不同機種規格與參數設定下，PCB在回焊爐內不同區位溫度設定對輸出溫度曲線的影響範圍。進而協助工程師進行生產流程作業設定，也能幫助釐清區位溫度設定對溫度曲線中各段之間的交互影響。
根據本研究的成果顯示，使用原始資料進行訓練時，機型A的均方根誤差為11.85393，其預測溫度曲線規範檢驗準確率達86.21%，機型B的均方根誤差為14.39509，其預測溫度曲線規範檢驗準確率達79.55%。使用擴增資料進行訓練時，機型A的均方根誤差為12.00008，其預測溫度曲線規範檢驗準確率達79.31%，機型B的均方根誤差為14.62653，其預測溫度曲線規範檢驗準確率達78.64%。因此在資料數量不足的情況下，透過適當的資料擴增，並且加入動態移動視窗注意力機制，在解決以短序列預測長序列的問題時可以得到較好的預測值。而其預測之溫度曲線在經過規範檢驗時，所檢測的檢驗準確率結果也較為精準。

Surface-mount technology (SMT) is one of the main reasons for the vigorous development of the electronic industry in recent years. Printed circuit board (PCB) can form a reflow profile through different setting sets of the reflow oven, which is an important criterion for the quality of the process.
This research analyzes expired PCB data, and establishes a Bidirectional Long Short-Term Memory (Bi-LSTM) neural network model to predict the temperature curve for different machine and temperature settings of the reflow oven. Due to the data type, we use Sequence-to-Sequence (Seq2Seq) as the model framework to predict long output sequences by using short input sequences. After enlarging the input sequence and adding positional encoding, we implement adaptive sliding window attention mechanism to predict the influence range of different reflow oven locations, automatically determine the center point and window length of the attention mechanism in each time step.
Our result shows that when we use original data to train our model, the root mean square error of model A is 11.85393, with an accuracy rate of 86.21%. The root mean square error of model B is 14.39509, with an accuracy rate of 79.55%. When we use augmented data to train our model, the root mean square error of model A is 12.00008, with an accuracy rate of 79.31%. The root mean square error of model B is 14.62653, with an accuracy rate of 78.64%. This experimental result demonstrates that adaptive sliding window attention mechanism Bi-LSTM model, comparing with previous model, performs better in both precision and accuracy.

[1] 陳俊凱，「SMT 自動化生產線製造彈性能力評估模式之發展」，樹德科技大學經營管理研究所，碩士論文，民國 99 年。
[2] Bhunia, S., & Tehranipoor, M. (2019). Chapter 4-Printed Circuit Board (PCB): Design and Test. Hardware Security, 81-105.
[3] Sarvar, F., & Conway, P. P. (1998). Effective modeling of the reflow soldering process: basis, construction, and operation of a process model. IEEE Transactions on Components, Packaging, and Manufacturing Technology: Part C, 21(2), 126-133.
[4] Amir, D. (1994). Expert system for SMT assembly. In Proceedings of the Surface Mount International Conference and Exposition-Technical Program (pp. 691-699). San Jose.
[5] Su, Y. Y., Srihari, K., & Emerson, C. R. (1997). A profile identification system for surface mount printed circuit board assembly. Computers & industrial engineering, 33(1-2), 377-380.
[6] Lau, J. H., & Pao, Y. H. (1997). Solder joint reliability of BGA, CSP, flip chip, and fine pitch SMT assemblies. McGraw-Hill Professional Publishing.
[7] Walker, A., & Baldwin, D. F. (1999). Initial investigations into low-cost ultra-fine pitch solder printing process based on innovative laser printing technology. IEEE Transactions on Electronics Packaging Manufacturing, 22(4), 303-307.
[8] Tsai, T. N. (2012). Thermal parameters optimization of a reflow soldering profile in printed circuit board assembly: A comparative study. Applied Soft Computing, 12(8), 2601-2613.
[9] Hochreiter, S., & Schmidhuber, J. (1997). Long short-term memory. Neural computation, 9(8), 1735-1780.
[10] Wang, Y., Huang, M., Zhu, X., & Zhao, L. (2016, November). Attention-based LSTM for aspect-level sentiment classification. In Proceedings of the 2016 conference on empirical methods in natural language processing (pp. 606-615).
[11] Mikolov, T., Karafiát, M., & Burget, L. (2010, September). J. ˇCernocky, and S. In Khudanpur,“Recurrent neural network based language model,” in Eleventh annual conference of the international speech communication association (Vol. 110).
[12] Ma, Y., Peng, H., & Cambria, E. (2018, April). Targeted aspect-based sentiment analysis via embedding commonsense knowledge into an attentive LSTM. In Thirty-second AAAI conference on artificial intelligence.
[13] Byeon, W., Breuel, T. M., Raue, F., & Liwicki, M. (2015). Scene labeling with lstm recurrent neural networks. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (pp. 3547-3555).
[14] Theis, L., & Bethge, M. (2015). Generative image modeling using spatial lstms. Advances in Neural Information Processing Systems, 28, 1927-1935.
[15] Yuan, X., Li, L., Shardt, Y. A., Wang, Y., & Yang, C. (2020). Deep learning with spatiotemporal attention-based LSTM for industrial soft sensor model development. IEEE Transactions on Industrial Electronics, 68(5), 4404-4414.
[16] Schuster, M., & Paliwal, K. K. (1997). Bidirectional recurrent neural networks. IEEE transactions on Signal Processing, 45(11), 2673-2681.
[17] Baldi, P., Brunak, S., Frasconi, P., Soda, G., & Pollastri, G. (1999). Exploiting the past and the future in protein secondary structure prediction. Bioinformatics, 15(11), 937-946.
[18] Siami-Namini, S., Tavakoli, N., & Namin, A. S. (2019, December). The performance of LSTM and Bi-LSTM in forecasting time series. In 2019 IEEE International Conference on Big Data (Big Data) (pp. 3285-3292). IEEE.
[19] Liu, J., Shahroudy, A., Xu, D., & Wang, G. (2016, October). Spatio-temporal lstm with trust gates for 3d human action recognition. In European conference on computer vision (pp. 816-833). Springer, Cham.
[20] Cho, K., Van Merriënboer, B., Bahdanau, D., & Bengio, Y. (2014). On the properties of neural machine translation: Encoder-decoder approaches. arXiv preprint arXiv:1409.1259.
[21] Cho, K., Van Merriënboer, B., Gulcehre, C., Bahdanau, D., Bougares, F., Schwenk, H., & Bengio, Y. (2014). Learning phrase representations using RNN encoder-decoder for statistical machine translation. arXiv preprint arXiv:1406.1078.
[22] Malhotra, P., Ramakrishnan, A., Anand, G., Vig, L., Agarwal, P., & Shroff, G. (2016). LSTM-based encoder-decoder for multi-sensor anomaly detection. arXiv preprint arXiv:1607.00148.
[23] Sutskever, I., Vinyals, O., & Le, Q. V. (2014). Sequence to sequence learning with neural networks. In Advances in neural information processing systems (pp. 3104-3112).
[24] Bahdanau, D., Cho, K., & Bengio, Y. (2014). Neural machine translation by jointly learning to align and translate. arXiv preprint arXiv:1409.0473.
[25] Luong, M. T., Pham, H., & Manning, C. D. (2015). Effective approaches to attention-based neural machine translation. arXiv preprint arXiv:1508.04025.
[26] Mirsamadi, S., Barsoum, E., & Zhang, C. (2017, March). Automatic speech emotion recognition using recurrent neural networks with local attention. In 2017 IEEE International conference on acoustics, speech and signal processing (ICASSP) (pp. 2227-2231). IEEE.
[27] Guo, Y., Ji, J., Lu, X., Huo, H., Fang, T., & Li, D. (2019). Global-local attention network for aerial scene classification. IEEE Access, 7, 67200-67212
[28] Song, S., Lan, C., Xing, J., Zeng, W., & Liu, J. (2017, February). An end-to-end spatio-temporal attention model for human action recognition from skeleton data. In Proceedings of the AAAI conference on artificial intelligence (Vol. 31, No. 1).
[29] Vaswani, A., Shazeer, N., Parmar, N., Uszkoreit, J., Jones, L., Gomez, A. N., ... & Polosukhin, I. (2017). Attention is all you need. In Advances in neural information processing systems (pp. 5998-6008).