【摘要】 注射压缩成型工艺(ICM)凭借低注射压力与均匀模具型腔压缩力已成为一种理想的聚合物透镜成型技术。然而，ICM工艺参数间存在复杂非线性交互关系，使得控制成型透镜质量变得十分困难。针对某款平凸透镜的注射压缩成型过程，以透镜成像相移和相移分布均匀度为质量设计目标，选取模具温度、熔体温度、注射时间、保压时间、保压压力、压缩距离和压缩速度等工艺参数为设计变量，进行Taguchi实验设计与Moldflow模拟分析。采用信噪比望小特性模型对实验模拟结果进行分析，结果表明，影响相移目标的重要工艺参数依次为保压时间、保压压力和注射时间，而影响均匀度目标的最重要工艺参数依次为保压压力、注射时间和熔体温度，两成像质量目标具有竞争关系，无法同时达到最优值。据此，采用融合空间信息注意力机制的卷积神经网络建立了成像质量目标可靠预测模型，运用快速精英非支配排序遗传算法并结合线性加权法获得了最佳工艺参数组。相较于初始成型条件，优化后的成像质量目标相移降低了64.82%，相移分布均匀度提高了5.76%，有效地提升了透镜的质量。更多还原

【Abstract】 Injection-compression molding(ICM) has gained prominence as an advanced polymer lens manufacturing technique due to its advantages of low injection pressure and uniform mold cavity compression. However,the complex and nonlinear interactions among ICM process parameters significantly hinder precise control over lens quality. ICM process for a plano-convex lens,with the goal of optimizing optical imaging performance in terms of phase shift and phase shift uniformity was investigated. Seven key process parameters,including mould temperature,melt temperature,injection time,packing time,packing pressure,compression distance,and compression speed were selected as design variables. A Taguchi design of experiments coupled with Moldflow simulations was employed to explore the process. The simulation results were evaluated using a smaller-the-better signalto-noise ratio model. The experimental simulation results were analyzed using a signal-to-noise ratio model with a small-scale characteristic. The results indicate that the important process parameters affecting the phase shift target are holding time,holding pressure,and injection time in order,while the most important process parameters affecting the uniformity target are holding pressure,injection time,and melt temperature in order. These two imaging quality objectives are competitive and cannot be simultaneously optimized. Accordingly,a reliable predictive model was constructed based on a convolutional neural network enhanced with spatial attention mechanism,and the optimal process parameters were obtained by using the fast elitist non-dominated sorting genetic algorithm,in conjunction with a linear weighted aggregation method. Compared with the initial processing conditions,the optimized settings lead to a 64.82% reduction in phase shift and a 5.76% improvement in uniformity,thereby significantly enhancing lens quality.更多还原