應用機器學習於鈦與鎂異種金屬銲接參數最佳化

由於輕量化、高強度鎂合金與輕量化、高強度、耐腐蝕、無磁性鈦合金有優越的特性，隨著工業的快速發展以及消費需求的改變導致現今應用非常普及，兩者的接合便非常重要，鎢極惰性氣體（tungsten inert gas, TIG）銲接是非常方便普及的接合技術，鈦與鎂合金材料銲接雖然有許多優異機械性質，但是鎂合金的熔點低（大約為649℃），鈦的熔點為1660℃，兩者熔點差異甚大，且在異種金屬TIG銲接上其可銲性之條件範圍狹窄，會有接合介面和融熔銲道形成硬而脆的介金屬化合物的困難點。一般對於銲接參數設定並沒有公式可循，完全憑藉專家過去的知識和經驗來設定，一旦超出專家經驗範圍，便無法有效設定最佳參數；因此，本研究提出一套機器學習分析流程，解決鈦與鎂合金異種金屬銲接多重品質特性的問題，提高整體銲接強度。利用理想解類似度順序偏好法（technique for order preference by similarity to ideal solution）與機器學習的演算法（包含隨機森林[random forest]、支援向量迴歸[support vector regression]、線性迴歸[linear regression]、類神經網路[artificial neural network, ANN]），並將以均方根誤差（root mean square error）指標評估各種模型之績效，訓練最佳化參數設計函數架構，結合完全排列組合法，找出鈦與鎂合金異種金屬銲接最佳化參數水準組合。研究結果顯示最佳模型為ANN，鈦與鎂異種金屬銲接最佳參數水準組合為銲接移行速度14.3 cm/min（280 rpm）、銲接電流170 A、氬氣流量11 L/min、鎢棒工作間隙2 mm、鎢棒凸出量2 mm。銲接移行速度是最重要的關鍵因子，且拉伸強度可以達到299 Mpa，本模式改善異種金屬銲接脆化性質與國內外相關多重品質特性實驗設計法，可學習異種金屬鈑材銲接參數與品質特性之關係，以利智慧製造設備參數設定決策輔助之應用，提升銲接相關產業產品品質和銲接效率。

關鍵字

鈦合金；鎂合金；隨機森林；支援向量迴歸；類神經網路

並列摘要

Due to the excellent characteristics of lightweight, high-strength magnesium alloy, and light weight, high-strength, corrosion-resistant, non-magnetic titanium alloy, the joining of titanium and magnesium alloy is gradually important, with the rapid development of industry, as well as changes in consumer demand. Tungsten inert gas (TIG) welding is convenient and general joining technology. The welding of different metal materials such as titanium and magnesium alloys have superior mechanical characteristics, but the melting point of titanium and magnesium alloys are 1660℃ and 649℃ respectively, the difference is very large, also the feasible set for the heterogeneous metal materials welding parameters of the TIG welding has many difficulties due to some hard and crisp inter-metallic compounds created within the weld line. Normally, the setting for welding parameters does not have the formula to follow; it usually depends on experts' past knowledge and experiences. Once exceeding the rule of thumb, it becomes impossible to feasibly set up the optimal parameters. So, this research proposed a machine learning analysis process to deal with the multiple quality characteristics problem of improving welding strength for titanium and magnesium alloy heterogeneous metal materials. It used technique for order preference by similarity to ideal solution and machine learning algorithms (containing random forest, support vector regression, linear regression, and artificial neural network [ANN]) algorithms to evaluate the performance of these models by the indices of root mean square error, this research method employed all permutation combinations to search the optimal parameters combination of titanium and magnesium alloy heterogeneous metal materials welding. The research results showed that the best model is ANN, and the best combination of parameters for titanium and magnesium heterogeneous metal welding were welding travel speed 14.3 cm/min (280 rpm), welding current 170 A, argon flow 11 L/min, tungsten rod working gap 2 mm, tungsten rod protrusion 2 mm. Welding travel speed was important factor, and the tensile strength could reach 299 MPa. This model was to improve the welding crisp of heterogeneous metal and previous experimental methods for multiple characteristics. Additionally, the model could learn the relationship between the welding parameters and the quality responses of different heterogeneous metal materials to facilitate future applications in the decision-making of parameter settings for automatic welding equipment of intelligent manufacturing. The research results could improve the product quality and welding efficiency of relevant welding industries.

並列關鍵字

titanium alloy ； magnesium alloy ； random forest ； support vector regression ； artificial neural network

參考文獻

張志平、劉孝先，2014，應用柔性演算法於航太鋁合金銲接參數最佳化之研究，品質學報，21(3)，205-216。doi:10.6220/joq.2014.21(3).05

Jhang, J.-P., 2016, The optimal parameters design of multiple quality characteristics for the welding of aluminum magnesium alloy, Journal of Quality, 23(3), 201-211. doi:10.6220/joq.2016.23(3).04

Tong, L.-I. and Su, C.-T., 1997, Optimizing multi-response problems in the Taguchi method by fuzzy multiple attribute decision making, Quality and Reliability Engineering International, 13(1), 25-34. doi:10.1002/(SICI)1099-1638(199701)13:1<25::AID-QRE59>3.0.CO;2-B

Zang, C., Liu, J., Tan, C., Zhang, K., Song, X., Chen, B., et al., 2018, Laser conduction welding characteristics of dissimilar metals Mg/Ti with Al interlayer, Journal of Manufacturing Processes, 32, 595-605. doi:10.1016/j.jmapro.2018.03.019

Hwang, C.-L. and Yoon, K., 1981, Multiple Attribute Decision Making—Methods and Applications: A State-of-the-Art-Survey, Springer-Verlag, Berlin, Germany. doi:10.1007/978-3-642-48318-9

主題瀏覽