基于机器学习的铝合金电弧增材薄壁构件成形质量预测及多目标优化

Machine Learning-Based Forming Quality Prediction and Multi-objective Optimization of Aluminum Alloy Thin-Walled Components in Wire Arc Additive Manufacturing

摘要:

电弧增材制造（WAAM）在航空航天领域具有重要应用价值，但其热输入不稳定性导致铝合金薄壁构件几何符合度差与内部缺陷多的问题突出。针对传统方法在多物理场耦合优化中的局限性，本研究提出数据驱动解决方案：通过构建工艺参数（电流、扫描速率、送丝速率）与成形质量（路径/层间壁厚一致性、孔隙率）的数据集，建立反向传播（BP）神经网络模型，并融合遗传算法（GA）优化原始模型，结合第二代非支配排序遗传算法（NSGA-II）进行成形质量多目标寻优。结果表明：优化后的GA-BP模型显著提升了沿路径壁厚一致性和孔隙率的预测精度，但层间壁厚一致性预测优化效果有限。通过NSGA-II获得的50组Pareto解集提出4类优化策略，验证试验结果表明模型预测误差为8.89%，准确地实现了成形质量指标的协同优化。

Abstract:

Wire arc additive manufacturing (WAAM) holds significant application value in the aerospace field, but the instability of heat input leads to prominent issues such as poor geometric conformity and numerous internal defects in aluminum alloy thin-walled components. To address the restrictions of traditional methods in multi-physics coupling optimization, this study proposed a data-driven solution by constructing a dataset of process parameters (current, scanning speed and wire feed rate) and forming quality (path/interlayer wall thickness consistency and porosity). A back propagation (BP) neural network model was established and optimized using the genetic algorithm (GA), combined with the non-dominated sorting genetic algorithm II (NSGA-II) for multi-objective optimization. The results show that the optimized GA-BP model significantly improves the prediction accuracy of path wall thickness consistency and porosity, but its optimization effect on interlayer wall thickness consistency prediction is restricted. Four types of optimization strategies are proposed based on the 50 Pareto solution sets obtained through NSGA-II, and validation tests indicate the model prediction error of 8.89%, accurately achieving the collaborative optimization of forming quality indicators.