基于PSO和Stacking集成学习的保险杠工艺优化

Optimization of Bumper Process Parameters Based on Integrated Learning of PSO and Stacking

结合保险杠在注塑成型过程中存在的体积收缩缺陷,采用Stacking集成学习方法建立了顶出时体积收缩率平均值的预测模型,并且对工艺参数进行优化。采用正交试验法设计试验方案,通过CAE软件分析并获得注塑成型过程温度、压力、时间等工艺参数与顶出时体积收缩率的平均值的样本数据。使用极限学习机结合岭回归、支持向量机回归、K近邻回归建立RSK-ELM集成模型,仿真实验表明,集成模型具有更高的预测精度。以降低顶出时体积收缩率的平均值为目标,基于建立的集成模型,运用粒子群算法对工艺参数优化问题进行求解,实验结果表明,使用优化的工艺参数,得到顶出时体积收缩率平均值为3.453%,与正交试验表中的下限相比,减少了3.94%,有效地降低了产品的收缩变形。因此,利用上述方法能提高产品的质量。

For the volumetric shrinkage defects of the bumper in the injection molding process,a prediction model of the average volumetric shrinkage during ejection was built using the Stacking integrated learning method.Then the optimal process parameters were obtained.The sample data of the average volumetric shrinkage during ejection and the injection molding process parameters were obtained by the Taguchi experiment design method and CAE software analysis.RSK-ELM integrated model,which combined ridge regression,support vector machine regression,k-nearest neighbor regression,and extreme learning machine,was established.The simulation results showed that the integrated model was of higher prediction accuracy.In order to reduce the average volumetric shrinkage during ejection,the particle swarm optimization algorithm was used to solve the process parameter optimization problem based on the established integrated model.The experiment results showed that the optimal average volumetric shrinkage during ejection was 3.453%,which was reduced by 3.94%compared with the lower limit of the Taguchi experiment table,and the shrinkage was effectively reduced.Therefore,the quality of the product could be improved by the use of the above methods.