Data-mining based machine learning (ML) method is emerging as a strategy to predict aluminium (Al) alloy properties with the promise of less intensive experimental work. However, ML models for wrought Al alloys are limited due to the difficulty in feature digitalization of the variety of manufacturing processes. Hence, most previous studies were constrained to specific alloy designations, which impeded the applicability of those ML models to broader wrought Al alloys. In the present work, we propose a novel feature engineering, called procedure-oriented decomposition (POD), assisting prediction framework to address the complexity introduced by manufacturing processes for wrought Al alloys. In this model, both chemical compositions and manufacturing processes are integrated as features. Correlation mapping of these features to the wrought Al alloys mechanical properties is established using the support vector regressor (SVR) model. The prediction framework demonstrates a high prediction accuracy and potential to design new alloys.

基于数据挖掘的机器学习 (ML) 方法正在成为一种预测铝 (Al) 合金性能的策略，并有望减少密集的实验工作。然而，由于各种制造工艺的特征数字化困难，锻造铝合金的 ML 模型受到限制。因此，以前的大多数研究都局限于特定的合金名称，这阻碍了这些 ML 模型对更广泛的锻造铝合金的适用性。在目前的工作中，我们提出了一种新的特征工程，称为面向过程的分解（POD），辅助预测框架来解决锻造铝合金制造工艺引入的复杂性。在这个模型中，化学成分和制造过程都被整合为特征。使用支持向量回归 (SVR) 模型建立这些特征与锻造铝合金机械性能的相关映射。该预测框架展示了高预测精度和设计新合金的潜力。