進行金屬積層製造時，存在許多因素會導致加工品質不穩定，當金屬粉末熔融不穩定時，易導致工件粗糙度、密度或結構的變異。以工件內流道為例，當表面粗糙度過大時，將會影響其流道內部流速；而各積層的粗糙度過小時，亦將增加工件結構的孔隙率。因此，如何基於已知的加工參數，估測各積層的粗糙度，實為繼續加工的重要依據。
本研究基於同軸融池影像，開發可估測粗糙度之關鍵特徵搜尋與應用方法。在關鍵特徵尋找上，根據製程參數、同軸融池影像、與同軸溫度特徵，利用LASSO (Least absolute shrinkage and selection operator)算法可篩選出與粗糙度關聯度高之關鍵特徵。在特徵應用上，基於雷射功率、雷射速度與工件位置等關鍵特徵，利用類神經網路，可先行預測下一層之熔池長度，再進而下一層的粗糙度。在系統整合上，本研究建立一資料庫與網頁服務，可提供線上檢視與離線分析之用。
在研究成果上，以SUS420粉末為例，當雷射功率160瓦特至270瓦特，以及雷射速度600mm/s至850mm/s的製程參數範圍內。本方法可基於當層資訊，估測下一層之粗糙度，粗糙度MAPE 5%以內。因此，本研究所開發之關鍵特徵搜尋與應用方法可適用於金屬積層製造之表面粗糙度預測。

During manufacturing process of metal additive manufacturing (AM), there exist factors could lead to variations of workpiece roughness, density, and porosity, and reduce workpiece quality due to unstable melting of metal powder. For a workpiece with coolants example, the inner speed of coolant will be decreased with high inner roughness, while porosity may be increased with undersized layer roughness. Hence, how to estimate layer roughness based on having parameters is essential for achieving a stable AM process.
This study based on melt-pool image develops a key feature selection and application method for estimating layer roughness. In key feature selection, the features with high correlation of roughness are selected from process parameters, melt-pool images, and melt-pool temperatures by using LASSO (Least absolute shrinkage and selection operator) algorithm. In feature application, the mean melt-pool length of the next-layer can be estimated and applied to evaluate the next-layer roughness based on neural network served with the key features including laser power, laser speed, and workpiece locations of current layer. In system integration, the database and web services are provided for online usage and offline analysis.
In terms of research results, taking SUS420 powder as an example, the ranges of process parameters are laser power of 160 to 270 watts, and the laser speed of 600 to 850 mm/s. The proposed method presents the roughness of MAPE of the next layer to be less than 5% based on current layer information. Therefore, the developed key feature selection and application method can predict surface roughness of metal additive manufacturing.

摘要 i
Abstract ii
誌謝 iii
目錄 iv
表目錄 vi
圖目錄 vii
第1章 緒論 1
1.1 研究背景 1
1.2 研究目的 4
1.3 研究架構 6
第2章 理論方法 7
2.1 特徵計算方法 7
2.2 特徵挑選方法 10
2.3 模型建立方法 11
2.4 模型品質檢測 13
第3章 系統方法 14
3.1 系統架構 14
3.1.1 系統概念 14
3.1.2 系統流程 16
3.2 資料蒐集 18
3.3 特徵萃取 25
3.3.1 關鍵特徵尋找 25
3.4 模型建立 27
3.4.1 類神經模型之建立 27
3.5 Node-red介面設計建立 28
3.5.1 Node-red介面設計建立 28
第4章 案例研究與實驗分析 30
4.1 實驗資料擷取 31
4.1.1 實驗一資料擷取 31
4.1.2 資料分析與關鍵特徵擷取 34
4.2 建立粗糙度預測模型 36
4.2.1 建立模型(一) 36
4.2.2 建立模型(二) 40
4.2.3 建立模型(三) 45
4.3 粗糙度預測模型比較 48
4.3.1 粗糙度模型比較 48
4.3.2 粗糙度模型說明 50
4.4 溫度模型之實驗設計 51
4.4.1 溫度與粗糙度關聯性 51
4.4.2 實驗資料擷取 53
4.4.3 建立BPT_M模型(一) 54
4.4.4 建立BPT_M模型(二) 56
第5章 結論與未來展望 58
5.1 結論 58
5.2 未來展望 59
參考文獻 60

 

[1]S. Kleszczynski, A. Ladewig, K. Friedberger, J. zur Jacobsmühlen, D. Merhof , and G. Witt, “Position Dependency of Surface Roughness in Parts from Laser Beam Melting Systems,” Proceedings of the 26th Internation Solid Free Form Fabrication (SFF) Symposium, 2015.
[2]L. Hitzler, J. Hirsch, M. Merkel, W. Hall, and A. Öchsner, “Position dependent surface quality in Selective Laser Melting,” Materialwissenschaft und Werkstofftechnik, vol. 48, pp. 327-334, 2017.
[3]Giovanni Strano, Liang Hao, Richard M. Everson, and Kenneth E. Evans, “Surface roughness analysis, modelling and prediction in selective laser melting,”Journal of Materials Processing Technology ,vol.213, Issue 4, pp.589-597, 2013.
[4]A. Safdar, W. He, L.-Y. Wei, A. Snis, and L. Chávez de Paz, “Effect of process parameters settings and thickness on surface roughness of EBM produced Ti-6Al-4V,” Rapid Prototyping Journal, vol. 18, pp. 401-408, 2012.
[5]J. K. Algardh et al., “Thickness dependency of mechanical properties for thin-walled titanium parts manufactured by Electron Beam Melting (EBM)®,” Additive Manufacturing, vol. 12, pp. 45-50, 2016.
[6]Mohammad A, Mohammed MK, Alahmari AM, “Effect of laser ablation parameters on surface improvement of electron beam melted parts,” Int. J. Adv. Manuf. Tech., pp.1-12, 2016.
[7]F. Calignano, D. Manfredi, E.P. Ambrosio, L. Iuliano, and P. Fino, “Influence of process parameters on surface roughness of aluminum parts produced by DMLS,” Int. J. Adv. Manuf. Tech., vol. 67, pp.2743-2751, 2013.
[8]D. Dai et al., “Influence of scan strategy and molten pool configuration on microstructures and tensile properties of selective laser melting additive manufactured aluminum based parts,” Optics & Laser Technology, vol. 99, pp. 91-100, 2018.
[9]P. J. DePond et al., “In situ measurements of layer roughness during laser powder bed fusion additive manufacturing using low coherence scanning interferometry,” Materials & Design, vol. 154, pp. 347-359, 2018.
[10]A. Spierings, N. Herres, and G. Levy, “Influence of the particle size distribution on surface quality and mechanical properties in AM steel parts,” Rapid Prototyping Journal - RAPID PROTOTYPING J, vol. 17, pp. 195-202, 2011.
[11]X. Han, H. H. Zhu, X. J. Nie, G. Q. Wang, and X. Y. Zeng, “Investigation on Selective Laser Melting AlSi10Mg Cellular Lattice Strut: Molten Pool Morphology, Surface Roughness and Dimensional Accuracy,” Materials, vol. 11, pp. 392 ,2018.
[12]I. Yadroitsev, P. Krakhmalev and I. Yadroitsava, “Selective laser melting of Ti6Al4V alloy for biomedical applications: temperature monitoring and microstructural evolution,” J. Alloys Comp., vol. 583, pp. 404-409, 2011.
[13]A. Hussein, L. Hao, C. Yan, and R. Everson, “Finite element simulation of the temperature and stress fields in single layers built without-support in selective laser melting,” M Mater. Des., vol.52, pp. 638-647, 2013.
[14]Y.L. Li, D.D. Gu, “Parametric analysis of thermal behavior during selective laser melting additive manufacturing of aluminum alloy powder,” Mater. Des., vol. 63, pp.856-867, 2014.
[15]黃熾宏，應用AVM於智慧積層製造量測，國立成功大學製造資訊與系統研究所碩士論文，2020。
[16]R. Tibshirani, “Regression shrinkage and selection via the lasso,” Journal of the Royal Statistical Society: Series B (Methodological), vol.58, pp.267-288, 1996