Process parameters design of squeeze casting through SMR ensemble model and ACO

Process parameters are key to the production and cast quality of squeeze casting (SC). The exiting methods to obtain the process parameters are selected directly based on experimental results or based on the optimization through single models, which require more experiments, are expensive, time-consuming, and less adaptable. In this study, a process-parameter design method of SC based on ensemble learning is proposed, and a design and optimization framework for SC process parameters based on multi-model ensemble is established. Based on this framework, using the support vector machine (SVM), multivariate linear regression (MLR), and random forest (RF) model as the unit models, and an improved model assembling strategy ( R 2 weight assignment), the ensemble model (SMR) for optimizing the SC process parameters is established. Then, to obtain the optimal SC process parameters, the ant-colony-optimization (ACO) algorithm is adopted to solve the SMR model. The two application cases show that the proposed ensemble strategy is reasonable and effective; the ensemble model had higher prediction accuracy and stronger generalization ability, even in small data sample situations. Furthermore, it effectively improved the quality of cast by its designed process parameters, and the shrinkage porosity of Case 1 cast, using the designed process parameters, was reduced to 1.316%. Compared with the conventional methods for designing SC process parameters, the method based on the ensemble model is more efficient and accurate, and reduces the cost.