基于多种代理模型的光纤激光切割质量预测

Quality Prediction of Fiber Laser Cutting Using Multiple Surrogate Models

为了提高光纤激光切割质量的预测精度,分别使用了多项式响应面法、克里金法和支持向量回归法三种代理模型方法,建立了激光功率、切割速率和辅助气体压力等加工参数与挂渣量之间的预测模型。使用拉丁超立方抽样法产生30组训练数据,利用正交试验法产生25组验证数据用以验证预测模型的精度。结果表明,利用克里金法和支持向量回归法建立的模型均成功地预测了挂渣质量,其中克里金法建立的模型预测精度最高,并利用该模型分析了激光切割加工参数对挂渣量的影响规律。

In order to improve the prediction accuracy of fiber laser cutting quality, three surrogate models such as polynomial response surface method, Kriging method and support vector regression method are used to establish the prediction model which includes three input variables of laser power, cutting rate and auxiliary gas pressure and one output of dross weight. Latin Hypercube Sampling is used to generate 30 sets of training data, and Taguchi method is used to generate 25 sets of testing data to validate the accuracy of the prediction model. The results show that the Kriging based model achieves the highest accuracy among three models, as a result, it is adopted to analyze the influence of cutting parameters on dross weight.