激光熔覆层形貌预测对比分析

Comparative analysis on predictions of the geometric form of laser clading

关键词:  激光熔覆  多元线性回归  遗传神经网络  形貌预测  对比分析

Key words:  laser cladding  multiple linear regression  genetic neural network  prediction of the form  comparative analysis

摘要:

将多元线性回归分析和遗传神经网络对比应用于激光熔覆层形貌的预测,确定了主要工艺参数(激光功率、扫描速率、送粉速率)和激光熔覆层形貌(熔覆层宽、高、基体熔深)之间的对应关系.结果表明,多元线性回归分析应用于激光熔覆层的形貌预测是可行的,五组检验数据的平均相对误差为6.05%;基于遗传算法优化的神经网络预测熔覆层形貌是可靠的,五组检验数据的平均相对误差为2.49%.二者相比较,前者应用较方便,能直观的获得熔覆层宽、高、熔深等参数与熔覆层形貌参数之间的函数关系;后者精度相对较高,但运算过程相对复杂,函数关系模糊.一般情况下推荐采用多元线性回归分析。

Abstract:

A comparison between the analytical methods of multiple linear regression analysis (MLRA) and genetic algorithm optimizing neural networks was made for predicting the geometric form of laser cladding. The corresponding relationship between main processing parameters (laser power, scanning volocity and powder mass flow rate) and the geometrric form of clad(ding width, height and depth of the penetration into she substrate) was affirmed.The result proved the feasiblity of using MLRA to predict the geometric form of laser ding and the averrage relative error of five test values was 2.49%. In comparison,the former is convenient in applic ation by which functional relationship between parameters such as width, height and depth of the penetration and so on. While the later produces a better precision and a invisible function relationship with a more complex operation process. Often the MLRA method is usuallyre commended.