常壓電漿氮化技術具有高效、低能耗和低成本等優勢。此外，該技術還具有處理效率高、處理後表面平整度好以及處理後材料變形小等優點。儘管常壓電漿加工具有快速加工的優勢，但後續加工結果的量測卻需要耗費大量時間成本。此外，對於特定加工結果所需的加工參數，僅能依靠過往經驗分析，不利於準確地符合所需加工品質的預期。
基於上述情況，本研究提出了一種基於人工智慧的加工品質預測方法，通過資料前處理增強資料以及待預測結果的相關性增強模型的預測性能表現，並藉由集成式學習結合各種增強模型預測性能的方法進而預測不同材料的氮化層厚度和加工後材料硬度。
兩種集成式學習之AISI D2氮化層厚度預測最大偏差分別為6.846%以及5.354%，SS304氮化層厚度預測最大偏差分別為12.376%以及11.95%，AISI D2加工後硬度預測最大偏差為5.882%以及7.259%，SS304加工後硬度預測最大偏差為8.329%以及8.62%。研究結果表明，該模型能夠準確預測材料的加工品質，並且能夠自動從資料集中提取對於預測模型訓練有利的特徵並僅須5秒便可取得實際加工品值結果，具有提高生產效率的特點。
常壓電漿氮化技術具有高效、低能耗和低成本等優勢。此外，該技術還具有處理效率高、處理後表面平整度好以及處理後材料變形小等優點。儘管常壓電漿加工具有快速加工的優勢，但後續加工結果的量測卻需要耗費大量時間成本。此外，對於特定加工結果所需的加工參數，僅能依靠過往經驗分析，不利於準確地符合所需加工品質的預期。
基於上述情況，本研究提出了一種基於人工智慧的加工品質預測方法，通過資料前處理增強資料以及待預測結果的相關性增強模型的預測性能表現，並藉由集成式學習結合各種增強模型預測性能的方法進而預測不同材料的氮化層厚度和加工後材料硬度。
兩種集成式學習之AISI D2氮化層厚度預測最大偏差分別為6.846%以及5.354%，SS304氮化層厚度預測最大偏差分別為12.376%以及11.95%，AISI D2加工後硬度預測最大偏差為5.882%以及7.259%，SS304加工後硬度預測最大偏差為8.329%以及8.62%。研究結果表明，該模型能夠準確預測材料的加工品質，並且能夠自動從資料集中提取對於預測模型訓練有利的特徵並僅須5秒便可取得實際加工品值結果，具有提高生產效率的特點。

The atmospheric pressure plasma nitriding technology possesses advantages such as high efficiency, low energy consumption, and low cost. Additionally, this technique offers benefits such as high processing efficiency, improved surface smoothness after treatment, and minimal material deformation. Despite the rapid processing advantages of atmospheric pressure plasma, measuring the results of post-processing requires a significant amount of time and cost. Moreover, determining the specific processing parameters for desired results relies solely on past experience and analysis, which hinders accurate attainment of the expected processing quality.
Given the aforementioned circumstances, this study proposed an artificial intelligence-based method for predicting processing quality. It enhances the predictive performance of the model by preprocessing data and strengthening the correlation with the predicted results. Furthermore, this study employed two ensemble learning approaches, combining various enhanced models' predictive performance to forecast the nitride layer thickness and material hardness after processing for different materials. The maximum deviations in the prediction of AISI D2 nitride layer thickness are 6.846% and 5.354%, while for SS304, they are 12.376% and 11.95%. As for AISI D2 hardness after processing, the maximum deviations are 5.882% and 7.259%, and for SS304, they are 8.329% and 8.62%. The research results demonstrate that this model accurately predicts material processing quality and can automatically extract advantageous features for training the predictive model from the dataset, obtaining actual processing values in just 5 seconds, thereby improving production efficiency.

[1]D. P. Dowling, F. T. O’Neill, S. J. Langlais, and V. J. Law, “Influence of dc Pulsed Atmospheric Pressure Plasma Jet Processing Conditions on Polymer Activation,” Plasma Processes and Polymers, vol. 8, no. 8, pp. 718–727, 2011, doi: 10.1002/ppap.201000145.
[2]M. Donegan, V. Milosavljević, and D. P. Dowling, “Activation of PET Using an RF Atmospheric Plasma System,” Plasma Chem Plasma Process, vol. 33, no. 5, pp. 941–957, Oct. 2013, doi: 10.1007/s11090-013-9474-4.
[3]E. Moritzer, C. Budde, and C. Leister, “Effect of atmospheric pressure plasma pre-treatment and aging conditions on the surface of thermoplastics,” Weld World, vol. 59, no. 1, pp. 23–32, Jan. 2015, doi: 10.1007/s40194-014-0174-3.
[4]Y. Jin, C. Ren, L. Yang, J. Zhang, and D. Wang, “Atmospheric Pressure Plasma Jet in Ar and O2/Ar Mixtures: Properties and High Performance for Surface Cleaning,” Plasma Science and Technology, vol. 15, Nov. 2013, doi: 10.1088/1009-0630/15/12/08.
[5]K. Fricke, H. Steffen, T. von Woedtke, K. Schröder, and K.-D. Weltmann, “High Rate Etching of Polymers by Means of an Atmospheric Pressure Plasma Jet,” Plasma Processes and Polymers, vol. 8, no. 1, pp. 51–58, 2011, doi: 10.1002/ppap.201000093.
[6]M. Crnjac, I. Veža, and N. Banduka, “From Concept to the Introduction of Industry 4.0”.
[7]A. Rojko, “Industry 4.0 Concept: Background and Overview,” International Journal of Interactive Mobile Technologies, vol. 11, no. 5, pp. 77–90, Nov. 2017, doi: 10.3991/ijim.v11i5.7072.
[8]B. P. Santos, F. Charrua-Santos, and T. M. Lima, “Industry 4.0: An Overview,” 2018.
[9]P. Zheng et al., “Smart manufacturing systems for Industry 4.0: Conceptual framework, scenarios, and future perspectives,” Front. Mech. Eng., vol. 13, no. 2, pp. 137–150, Jun. 2018, doi: 10.1007/s11465-018-0499-5.
[10]P. O’Donovan, K. Leahy, K. Bruton, and D. T. J. O’Sullivan, “An industrial big data pipeline for data-driven analytics maintenance applications in large-scale smart manufacturing facilities,” Journal of Big Data, vol. 2, no. 1, p. 25, Nov. 2015, doi: 10.1186/s40537-015-0034-z.
[11]“Data-driven smart manufacturing | Elsevier Enhanced Reader.” https://reader.elsevier.com/reader/sd/pii/S0278612518300062?token=5C07BF5E763F039CD4CC3340066289E8D107D0F0E4D4759641CCAA11423C6646D17B6D424B7832483AB6D2D6DE0EE76B&originRegion=us-east-1&originCreation=20230415131453 (accessed Apr. 15, 2023).
[12]Y. Lecun, L. Bottou, Y. Bengio, and P. Haffner, “Gradient-based learning applied to document recognition,” Proceedings of the IEEE, vol. 86, no. 11, pp. 2278–2324, Jan. 1998, doi: 10.1109/5.726791.
[13]I. J. Goodfellow et al., “Generative Adversarial Networks.” arXiv, Jun. 10, 2014. doi: 10.48550/arXiv.1406.2661.
[14]A. L. Samuel, “Some Studies in Machine Learning Using the Game of Checkers,” IBM Journal of Research and Development, vol. 3, no. 3, pp. 210–229, Jul. 1959, doi: 10.1147/rd.33.0210.
[15]A. Shirzadi et al., “A comparative study between popular statistical and machine learning methods for simulating volume of landslides,” CATENA, vol. 157, pp. 213–226, Oct. 2017, doi: 10.1016/j.catena.2017.05.016.
[16]“Comparing conventional statistical models and machine learning in a small cohort of South African cardiac patients | Elsevier Enhanced Reader.” https://reader.elsevier.com/reader/sd/pii/S2352914822002404?token=45D95960AB8EFBA3E93776F7726A7D4EB5798B04BA72DD98D97347AC354B535FC737174E5CF7AF8F5AB323FEBA2BCBB5&originRegion=us-east-1&originCreation=20230423133732 (accessed Apr. 23, 2023).
[17]M. Nalbant, H. Gokkaya, and İ. Toktaş, “Comparison of Regression and Artificial Neural Network Models for Surface Roughness Prediction with the Cutting Parameters in CNC Turning,” Modelling and Simulation in Engineering, vol. 2007, p. e92717, Jun. 2007, doi: 10.1155/2007/92717.
[18]C. Cortes and V. Vapnik, “Support-vector networks,” Mach Learn, vol. 20, no. 3, pp. 273–297, Sep. 1995, doi: 10.1007/BF00994018.
[19]W. S. McCulloch and W. Pitts, “A logical calculus of the ideas immanent in nervous activity,” Bulletin of Mathematical Biophysics, vol. 5, no. 4, pp. 115–133, Dec. 1943, doi: 10.1007/BF02478259.
[20]D. P. Kingma and J. Ba, “Adam: A Method for Stochastic Optimization.” arXiv, Jan. 29, 2017. Accessed: Jul. 28, 2023. [Online]. Available: http://arxiv.org/abs/1412.6980
[21]M. Kopec et al., “Effect of plasma nitriding process on the fatigue and high temperature corrosion resistance of Inconel 740H nickel alloy,” Archiv.Civ.Mech.Eng, vol. 22, no. 1, p. 57, Jan. 2022, doi: 10.1007/s43452-022-00381-0.
[22]J. C. Díaz-Guillén et al., “Improved Mechanical Properties, Wear and Corrosion Resistance of 316L Steel by Homogeneous Chromium Nitride Layer Synthesis Using Plasma Nitriding,” J. of Materi Eng and Perform, vol. 29, no. 2, pp. 877–889, Feb. 2020, doi: 10.1007/s11665-020-04653-9.
[23]J. C. Díaz-Guillén et al., “Duplex plasma treatment of AISI D2 tool steel by combining plasma nitriding (with and without white layer) and post-oxidation,” Surface and Coatings Technology, vol. 385, p. 125420, Mar. 2020, doi: 10.1016/j.surfcoat.2020.125420.
[24]M. Godec, B. Podgornik, A. Kocijan, Č. Donik, and D. A. S. Balantič, “Use of plasma nitriding to improve the wear and corrosion resistance of 18Ni-300 maraging steel manufactured by selective laser melting,” Sci Rep, vol. 11, no. 1, p. 3277, Feb. 2021, doi: 10.1038/s41598-021-82572-y.
[25]Q. Xie, M. Suvarna, J. Li, X. Zhu, J. Cai, and X. Wang, “Online prediction of mechanical properties of hot rolled steel plate using machine learning,” Materials & Design, vol. 197, p. 109201, Jan. 2021, doi: 10.1016/j.matdes.2020.109201.
[26]B. N. Galimzyanov, M. A. Doronina, and A. V. Mokshin, “Machine learning-based prediction of elastic properties of amorphous metal alloys,” Physica A: Statistical Mechanics and its Applications, vol. 617, p. 128678, May 2023, doi: 10.1016/j.physa.2023.128678.
[27]M. S. Ozerdem and S. Kolukisa, “Artificial neural network approach to predict the mechanical properties of Cu–Sn–Pb–Zn–Ni cast alloys,” Materials & Design, vol. 30, no. 3, pp. 764–769, Mar. 2009, doi: 10.1016/j.matdes.2008.05.019.
[28]G. Khalaj, “Artificial neural network to predict the effects of coating parameters on layer thickness of chromium carbonitride coating on pre-nitrided steels,” Neural Comput & Applic, vol. 23, no. 3, pp. 779–786, Sep. 2013, doi: 10.1007/s00521-012-0994-2.
[29]A. F. Yetim, M. Y. Codur, and M. Yazici, “Using of artificial neural network for the prediction of tribological properties of plasma nitrided 316L stainless steel,” Materials Letters, vol. 158, pp. 170–173, Nov. 2015, doi: 10.1016/j.matlet.2015.06.015.
[30]J. Pribbenow, M. Mejauschek, P. Landgraf, T. Grund, G. Bräuer, and T. Lampke, “Neural network for prediction of hardness profiles for steel alloys after plasma nitriding,” IOP Conf. Ser.: Mater. Sci. Eng., vol. 480, no. 1, p. 012019, Feb. 2019, doi: 10.1088/1757-899X/480/1/012019.
[31]A. Zhecheva, S. Malinov, and W. Sha, “Simulation of microhardness profiles of titanium alloys after surface nitriding using artificial neural network,” Surface and Coatings Technology, vol. 200, no. 7, pp. 2332–2342, Dec. 2005, doi: 10.1016/j.surfcoat.2004.10.018.
[32]S. Dean, T. Filetin, I. Zmak, and D. Novak, “Determining Nitriding Parameters with Neural Networks,” Journal of Astm International, vol. 2, May 2005, doi: 10.1520/JAI12213.
[33]D.-C. Feng, W.-J. Wang, S. Mangalathu, G. Hu, and T. Wu, “Implementing ensemble learning methods to predict the shear strength of RC deep beams with/without web reinforcements,” Engineering Structures, vol. 235, p. 111979, May 2021, doi: 10.1016/j.engstruct.2021.111979.
[34]Q.-F. Li and Z.-M. Song, “High-performance concrete strength prediction based on ensemble learning,” Construction and Building Materials, vol. 324, p. 126694, Mar. 2022, doi: 10.1016/j.conbuildmat.2022.126694.
[35] Z. Zhang et al., “An improved stacking ensemble learning model for predicting the effect of lattice structure defects on yield stress,” Computers in Industry, vol. 151, p. 103986, Oct. 2023, doi: 10.1016/j.compind.2023.103986.
[36]https://www.facebook.com/sag.micro, “Scanning Electron Microscope (SEM)- Definition, Principle, Parts, Images - Microbe Notes,” Mar. 11, 2022. https://microbenotes.com/scanning-electron-microscope-sem/ (accessed Jun. 30, 2023).
[37]“Vickers-Hardness-Testing.png (457×261).” https://www.hardnessgauge.com/wp-content/uploads/2020/11/Vickers-Hardness-Testing.png (accessed Apr. 29, 2023).
[38]郭兆渝, “常壓電漿噴射束於回火麻田散鐵鋼表面滲氮處理之製程溫度、時間與硬質層生成相互關係之研究,” 國立臺灣科技大學, 台北市, 2022. [Online]. Available: https://hdl.handle.net/11296/vpm84g
[39]張祐錡, “常壓電漿氮化處理對AISI 304不鏽鋼抗衝擊磨損和耐蝕性能之研究,” 國立臺灣科技大學, 台北市, 2021. [Online]. Available: https://hdl.handle.net/11296/g9v5wm
[40]J. Bergstra and Y. Bengio, “Random Search for Hyper-Parameter Optimization”.
[41]J. Snoek, H. Larochelle, and R. P. Adams, “Practical Bayesian Optimization of Machine Learning Algorithms.” arXiv, Aug. 29, 2012. doi: 10.48550/arXiv.1206.2944.
[42]S. Falkner, A. Klein, and F. Hutter, “Practical Hyperparameter Optimization for Deep Learning,” May 2023, Accessed: Jun. 15, 2023. [Online]. Available: https://openreview.net/forum?id=HJMudFkDf
[43]X. He, K. Zhao, and X. Chu, “AutoML: A survey of the state-of-the-art,” Knowledge-Based Systems, vol. 212, p. 106622, Jan. 2021, doi: 10.1016/j.knosys.2020.106622.