Abstract

Machine learning has revolutionized the way complex problems are solved in engineering. In the current work, machine learning methodology has been applied for predictive modeling of surface roughness generation during machining of Al-Mg based metal matrix composites (MMCs) reinforced with micro boron carbide and multiwalled carbon nanotube particles. Machine learning was used for parameter estimation of modeling structures such as auto regressive with exogenous variables (ARX), auto regressive moving average with exogenous variables (ARMAX), Box Jenkins (BJ) and Output Error (OE). The identified models were validated on the basis of FIT, final prediction error (FPE) and mean squared error (MSE). The PID, fractional order PID (FOPID), complex order PID (COPID) and model predictive controllers (MPC) were employed to effectively control machined surface roughness based on the best performing predictive models. Primary results indicate that: (1) CNT MMCs generate surface roughness comparable to that due to the micro MMCs with tenfold higher reinforcement fractions (2) ARX441 and ARMAX3331 are the best performing predictive models for the nano and micro MMCs respectively (3) PID and MPC are the best controllers for micro and nano MMC systems respectively considering the peak overshoots as the foremost performance metric (safety), followed by settling time (productivity).

摘要

机器学习彻底改变了工程中解决复杂问题的方式。在目前的工作中，机器学习方法已被应用于对用微碳化硼和多壁碳纳米管颗粒增强的 Al-Mg 基金属基复合材料 (MMC) 的加工过程中的表面粗糙度生成进行预测建模。机器学习用于建模结构的参数估计，例如带有外生变量的自回归 (ARX)、带有外生变量的自回归移动平均 (ARMAX)、Box Jenkins (BJ) 和输出误差 (OE)。确定的模型在 FIT、最终预测误差 (FPE) 和均方误差 (MSE) 的基础上进行了验证。PID，分数阶 PID (FOPID)，复杂阶 PID (COPID) 和模型预测控制器 (MPC) 用于根据性能最佳的预测模型有效控制加工表面粗糙度。初步结果表明：(1) CNT MMC 产生的表面粗糙度可与具有 10 倍增强率的微型 MMC 相媲美 (2) ARX441 和 ARMAX3331 分别是纳米和微型 MMC 的最佳预测模型 (3) PID 和MPC 是微型和纳米 MMC 系统的最佳控制器，分别将峰值过冲视为最重要的性能指标（安全性），其次是稳定时间（生产率）。