Multi objective optimization design of injection molding process parameters driven by multimodal data
摘要
注塑成型工艺参数优化中,多源数据利用率低、多目标冲突难以协调,影响制品质量稳定性.多模态数据存在量纲、采样率与维度的差异,且与翘曲变形、体积收缩率、残余应力之间呈强非线性、强耦合关系,难以直接融合并建立准确的工艺参数-质量映射模型,导致优化过程中翘曲变形增大、体积收缩分布不均、残余应力波动剧烈,直接影响产品合格率.为此,提出一种多模态数据驱动下注塑工艺参数多目标优化设计方法.该方法采集注塑机时序传感、网格化空间及静态质量三类数据,经Z-score 标准化与移动平均滤波去噪后,采用主成分分析依据累积贡献率筛选关键主成分,实现多模态数据融合降维.以降维后的特征为输入,构建 Kriging 代理模型,通过高斯相关函数与广义最小二乘法建立工艺参数到翘曲变形、体积收缩率、残余应力的非线性映射.以最小化翘曲变形、体积收缩率和残余应力为目标,采用改进非支配排序遗传算法进行多目标优化,结合快速非支配排序、拥挤距离计算、模拟二进制交叉与多项式变异算子,得到帕累托最优解集.通过综合满意度函数从帕累托最优解集中输出最优工艺参数决策.实验结果表明,所提方法通过多模态融合使翘曲、收缩方差、应力峰值分别降低8.6%、12.3%、9.7%,收敛快12.5%;最优参数(228℃/45 mm·s-1/85 MPa)翘曲和收缩预测误差小于 5%.所提方法有效提升了制品质量与效率.
In the optimization of injection molding process parameters,the low utilization of multi-source data and difficulty in coordinating multiple conflicting objectives affect the stability of product quality.Multimodal data present differences in scale,sampling rate,and dimension,and have strong nonlinear and strongly coupled relationships with warpage deformation,volumetric shrinkage,and residual stress.This makes it difficult to directly integrate the data and establish an accurate process parameter-quality mapping model,resulting in increased warpage deformation,uneven volumetric shrinkage distribution,and severe residual stress fluctuations during the optimization process,directly affecting the product pass rate.To address this,a multimodal data-driven multi-objective optimization design method for injection molding process parameters is proposed.This method collects three types of data:injection molding machine time-series sensor data,gridded spatial data,and static quality data.After Z-score normalization and moving average filtering for noise reduction,principal component analysis is used to select key principal components based on cumulative contribution rate to achieve multimodal data fusion and dimensionality reduction.The reduced-dimensional features are used as inputs to construct a Kriging surrogate model,establishing a nonlinear mapping from process parameters to warpage deformation,volumetric shrinkage,and residual stress using a Gaussian correlation function and generalized least squares method.Taking the minimization of warpage deformation,volumetric shrinkage,and residual stress as the objectives,the improved non-dominated sorting genetic algorithm is employed for multi-objective optimization,combining fast non-dominated sorting,crowding distance calculation,simulated binary crossover,and polynomial mutation operators to obtain a Pareto-optimal solution set.The optimal process parameter decision is output from the Pareto-optimal set using a comprehensive satisfaction function.Experimental results show that the proposed method reduces warpage,shrinkage variance,and stress peak by 8.6%,12.3%,and 9.7%,respectively,with a 12.5%faster convergence rate,through multimodal fusion.The optimal parameters(228℃/45 mm·s-1/85 MPa)achieve predicted warpage and shrinkage errors of less than 5%.The proposed method effectively improves product quality and efficiency.