表面黏著技術製程是透過印刷鋼網，由印刷機將錫膏印刷於電路板焊盤上，再經由錫膏檢查機確認錫膏印刷狀態。之後再藉由實裝機吸取零件，經過機器判別其外型是否符合規格，符合規格的零件就會被精確地裝著於電路板印刷位置上，再經過迴焊爐，將零件熔接於電路板後，形成半成品。
近年來因人口老化、少子化及工廠人員流動率高等因素，造成勞動人力短缺及技術流失之困境。所以，如何將印刷機老師傅調整印刷參數的經驗及技術，透過機器學習及深度學習的方式，取得較佳錫膏厚度並建立預測錫膏厚度模組,透過此模組能精確地讓工程師預測錫膏厚度，這是一個非常重要的課題，也是本論文將研究及實現的主題。
本論文將取得實際SMT生產線上，由數十個機種經錫膏檢查機檢查輸出的錫膏厚度資料。從各種印刷參數組合所得到之錫膏厚度為大數據，透過深度學習來建立預測錫膏厚度的模型。讓SMT生產線工程師可於模型內輸入印刷參數，用以預測錫膏厚度，來達到未來開發模組的目標。



關鍵字：表面黏著，檢查機，錫膏，鋼網，實裝機

The surface mounting technology (SMT) process involves printing solder paste onto the circuit board pads through a stencil by a printing machine, and then confirming the printing status of the solder paste by a solder paste inspection machine (SPI). After that, the parts are picked up by the mounter machine, and the machine judges whether the shape of the parts meets the specifications or not. Parts that meet the specifications are precisely mounted on the printed circuit boards. Finally, the parts are fused to the circuit board through a soldering reflow to form a semi-finished product.
In recent years, due to the aging population, childlessness, and high turnover rate of factory workers, there is a shortage of labor manpower. Therefore, how to present the experience and technology of printing machine masters in a digital way and predict the thickness of solder paste accurately is an important topic. This is also the theme that this paper will study and realize.
In this paper, we obtain the solder thickness data from dozens of products in an actual SMT production line through solder paste inspection. The solder paste thickness obtained from various combinations of printing parameters is used as the big data to build a model for predicting the solder paste thickness through deep learning. This allows SMT line engineers to input printing parameters into the model to predict solder paste thickness to achieve the goal of developing modules in the future.


Key words: SMT, SPI, solder, mask, mounter

目錄
摘要 i
Abstract ii
致謝 iii
目錄 iv
圖目錄 vi
表目錄 viii
公式目錄 ix
第一章 緒論 1
1.1 研究背景與動機 1
1.1.1 SMT印刷製程及錫膏檢查 1
1.1.2 有／無錫膏檢查機SPI流程 1
1.2 研究目的與方法 3
1.3 研究架構與流程 6
第二章 文獻探討與回顧 8
2.1文獻LIST 8
2.1.1 人工智慧 10
2.1.2類神經網路(Neural Network) 10
2.1.3深度學習(Deep learning) 12
2.1.4機器學習(Machine learning) 13
2.1.5 深度學習與機器學習的應用 13
2.2 萃取參考文獻精髓 14
2.3 本論文探討內容與文獻間差異 17
第三章 研究方法 19
3.1錫膏厚度控制實證與研究條件及方法簡介 19
3.1.1 SMT印刷製程重要性說明 19
3.1.2 研究範圍與對象 23
3.1.3錫膏厚度控制實證與研究測試條件 28
3.1.4 研究及實證方法 32
3.2 錫膏厚度標準 34
第四章 錫膏厚度控制實證與研究 36
4.1 錫膏厚度控制實證與研究所需之原始資料 36
4.2 觀察及研究對象選定 37
4.3針對SPI輸出資料實施訓練及測試 39
4.4 SPI輸出資料經軟體訓練及測試後所呈現各參數相關圖 40
4.4.1 相關圖 40
4.4.2預測結果(By Prediction Chart) 42
4.4.3 依據錫膏檢查機輸出資料彙整後實際測試情況 43
4.4.4 xgb/lasso/Enet 預測模型說明 44
4.4.5 決定較佳預測模型 46
4.4.6 評估預測值精度的方法 48
第五章 結論與未來發展 52
5.1 結論 52
5.2 未來發展 52
參考文獻 54 
圖目錄
圖1-1 Printer and SPI process 1
圖2-1 類神經網路示意圖 12
圖2-2 RSM分析示意圖 17
圖3-1 印刷工程不良示意圖 19
圖3-2 零件實裝後不良示意圖 20
圖3-3 迴流焊後不良示意圖 20
圖3-4 錫膏印刷品質魚骨圖 22
圖3-5 印刷電路板Gerber data 27
圖3-6 印刷電路板錫膏印刷／零件實裝／Reflow後零件結合示意圖 28
圖3-7 錫膏印刷機外觀示意圖 29
圖3-8 錫膏檢查機(SPI)外觀示意圖 29
圖3-9 Chip尺寸／鋼網開口尺寸及錫膏形式對應表 30
圖3-10印刷鋼網(Mask)實際圖片 30
圖3-11為圖3-10紅框處之放大圖 30
圖3-12較佳錫膏厚度預測流程圖 33
圖4-1 測試PCB基板Gerber示意圖 38
圖4-2 測試PCB基板實際裝著示意圖 39
圖4-3 Machine learning總點數示意圖 40
圖4-4 Machine learning後各參數之間關係示意圖 41
圖4-5 Machine learning後預測錫膏厚度趨勢圖 42
圖4-6 Machine learning後全體錫膏厚度mean data示意圖 43
圖4-7 利用預測模型進行錫膏厚度預測 44
圖4-8 xgb預測流圖示意圖 45
圖4-9 利用三種預測模型判斷較佳模型 47
圖4-10 五種評估預測精度model Python code 50 
表目錄
表2-1 文獻Data 9
表2-2 鋼網厚度與開口面積比 15
表3-1 錫膏印刷參數對印刷的影響及效果說明 25
表3-2 印刷機印刷後脫模參數 26
表3-3 21種產品參數及PCB pad點數總表 31
表3-4 日本推薦不同厚度鋼網所對應之錫膏厚度及公差 34
表3-5 台灣SMT產業界不同厚度鋼網所對應之錫膏厚度及公差 35
表4-1 客戶生產線錫膏機印刷參數管制表 36
表4-2 錫膏厚度經SPI檢查後輸出之樣本資料 37
表4-3 五種評估預測精度模型評估預測精度總表 51
 
公式目錄
式2-1 文獻3 MAPE公式 16
式4-1 MAE公式 48
式4-2 MSE公式 48
式4-3 MASLE公式 49
式4-4 MedAE公式 49
式4-5 MAPE公式 50

Alelaumi, S., Wang, H., Lu, H., & Yoon, S. W. (2020). A predictive abnormality detection
model using ensemble learning in stencil printing process. IEEE Transactions on
Components, Packaging and Manufacturing Technology, 10(9), 1560-1568.
Al-Refaie, A. (2009). Optimizing SMT performance using comparisons of efficiency
between different systems technique in DEA. IEEE Transactions on Electronics
Packaging Manufacturing, 32(4), 256-264.
Alzubi, J., Nayyar, A., & Kumar, A. (2018, November). Machine learning from theory to
algorithms: an overview. In Journal of physics: conference series (Vol. 1142, p.
012012). IOP Publishing.
Bonaccorso, G. (2017). Machine learning algorithms. Packt Publishing Ltd.
Burrell, J. (2016). How the machine ‘thinks’: Understanding opacity in machine learning
algorithms. Big data & society, 3(1), 2053951715622512.
Huang, C. Y., Shen, L. C., Greene, C., & Yang, C. C. (2021). Parameter Optimization of Pre-tin Printing Process of Wireless Communication Module. IEEE Transactions on Components, Packaging and Manufacturing Technology, 11(7), 1137-1147.
Hu, X. (2018, September). Analysis of Materials and Processes Affecting SMT Printing
Quality. In 4th Workshop on Advanced Research and Technology in Industry
(WARTIA 2018) (pp. 39-45). Atlantis Press.
Khader, N., & Yoon, S. W. (2018). Stencil printing process optimization to control solder paste volume transfer efficiency. IEEE Transactions on Components, Packaging and Manufacturing Technology, 8(9), 1686-1694.
Khader, N., & Yoon, S. W. (2018). Online control of stencil printing parameters using reinforcement learning approach. Procedia Manufacturing, 17, 94-101.

Khan, A., Baharudin, B., Lee, L. H., & Khan, K. (2010). A review of machine learning
algorithms for text-documents classification. Journal of advances in information
technology, 1(1), 4-20.
Li, M. H. C., Al-Refaie, A., & Yang, C. Y. (2008). DMAIC approach to improve the capability of SMT solder printing process. IEEE Transactions on Electronics Packaging Manufacturing, 31(2), 126-133.
Mannan, S. H., Ekere, N. N., Ismail, I., & Currie, M. A. (1995). Flow processes in solder
paste during stencil printing for SMT assembly. Journal of Materials Science:
Materials in Electronics, 6, 34-42.
Mannan, S. H., Ekere, N. N., Ismail, I., & Lo, E. K. (1994). Squeegee deformation study in the stencil printing of solder pastes. IEEE Transactions on Components, Packaging, and Manufacturing Technology: Part A, 17(3), 470-476.
Pan, J., Tonkay, G. L., Storer, R. H., Sallade, R. M., & Leandri, D. J. (2004). Critical variables of solder paste stencil printing for micro-BGA and fine-pitch QFP. IEEE Transactions on Electronics Packaging Manufacturing, 27(2), 125-132.
Ramli, M. I. I., Mohd Salleh, M. A. A., Mohd Sobri, F. A., Narayanan, P., Sweatman, K.,
& Nogita, K. (2019). Relationship between free solder thicknesses to the solder
ability of Sn–0.7 Cu–0.05 Ni solder coating during soldering. Journal of Materials
Science: Materials in Electronics, 30, 3669-3677.
Ray, S. (2019, February). A quick review of machine learning algorithms. In 2019
International conference on machine learning, big data, cloud and parallel
computing (COMITCon) (pp. 35-39). IEEE.
Rusdi, M. S., Abdullah, M. Z., Chellvarajoo, S., Abdul Aziz, M. S., Abdullah, M. K., Rethinasamy, P., ... & Santhanasamy, D. G. (2019). Stencil printing process performance on various aperture size and optimization for lead-free solder paste. The International Journal of Advanced Manufacturing Technology, 102, 3369-3379.
Sarker, I. H. (2021). Machine learning: Algorithms, real-world applications and research
directions. SN computer science, 2(3), 160.
Singh, A., Thakur, N., & Sharma, A. (2016, March). A review of supervised machine
learning algorithms. In 2016 3rd international conference on computing for
sustainable global development (INDIACom) (pp. 1310-1315). Ieee.
Yang, T., Tsai, T. N., & Yeh, J. (2005). A neural network-based prediction model for fine pitch stencil-printing quality in surface mount assembly. Engineering Applications of Artificial Intelligence, 18(3), 335-341.
劉鋒,胡天英,陳俊霖, & 但晨. (2021). Independent Variable Selection of High-Dimensional Data in Cox Regression Model. Statistics and Application, 10(2), 183-192.
NVIDIA,(2023/7).甚麼是XGBoost ?數據科學(NVIDIA術語表)
https://www.nvidia.cn/glossary/data-science/xgboost/