基于机器学习耦合模型预测FDM零件的表面粗糙度  

Predicting Surface Roughness of Parts Manufactured by the Fused Deposition Modeling Based on Coupled Machine Learning models

作　　者：赵陶钰[1] 邵鹏华 ZHAO Taoyu;SHAO Penghua(Department of Computer and Information Engineering,Shanxi Youth Vocation College,Taiyuan 030020,China;Coal Industry Taiyuan Design and Research Institute Group Co.,Ltd.,Taiyuan 030024,China)

机构地区：[1]山西青年职业学院计算机与信息工程系,山西太原030020 [2]煤炭工业太原设计研究院集团有限公司,山西太原030024

出　　处：《塑料工业》2024年第5期116-123,共8页China Plastics Industry

基　　金：山西省自然科学基金青年项目(202203041345225)。

摘　　要：熔融沉积工艺(FDM)制造的零件表面粗糙度高,不仅影响了零件外观,还降低了性能。采用响应面实验设计,研究了层高(A)、填充密度(B)、喷嘴温度(C)、床层温度(D)和打印速度(E)对聚乳酸(PLA)零件表面粗糙度的影响。同时,将遗传算法(GA)与决策树(DT)、人工神经元网络(ANN)两种机器学习模型相结合,预测了零件的表面粗糙度。结果表明,A、B、C和E是显著影响零件表面粗糙度的主效应,A×B、A×C、A×E、B×C、B×E、C×E是影响显著的交互效应。GA+DT耦合模型预测PLA零件表面粗糙度的准确性更高,预测值与实验值的相关系数(R2)、均方误差(MSE)和平均绝对误差(MAE)分别为0.952、0.132和0.234,优于GA+ANN的0.823、1.561和1.759。GA+DT模型的预测值与实验值的Pearson相关系数为0.984,而GA+ANN模型仅为0.903,这表明GA+DT模型在预测PLA零件表面粗糙度时准确度更高。The surface roughness of parts manufactured by the fused deposition modeling(FDM)is high,which affects the appearance of the parts and decreases the performances.The response surface design was used to investigate the effects of layer height(A),filling density(B),nozzle temperature(C),bed temperature(D),and printing speed(E)on the surface roughness of polylactic acid(PLA)parts.At the same time,combining genetic algorithm(GA)with decision tree(DT)and artificial neural network(ANN),the surface roughness of the parts was predicted.The results show that A,B,C,and E have significant impacts on the surface roughness of parts,A×B,A×C,A×E,B×C,B×E,C×E are significant interaction effects.The GA+DT coupled model has higher accuracy in predicting the surface roughness of PLA parts,and the correlation coefficient(R 2),mean square error(MSE),and mean absolute error(MAE)values between predicted and experimental values are respectively 0.952,0.132,and 0.234,which are better than these of GA+ANN coupled model(0.823,1.561,and 1.759).The Pearson correlation coefficient between the predicted values by the GA+DT coupled model and the experimental results is 0.984,while that between the predicted values by the GA+ANN coupled model and the experimental results is 0.903,indicating that the GA+DT coupled model has higher accuracy in predicting the surface roughness of PLA parts.

关 键 词：决策树 人工神经元网络 遗传算法 熔融沉积 表面粗糙度 聚乳酸 

分 类 号：TQ320[化学工程—合成树脂塑料工业]