Intelligent manufacturing system and hybrid generative-attention framework for quality prediction of high-performance aluminum alloys

High-performance aluminum alloys are extensively used in aerospace, rail transit, and new-energy vehicles owing to their high specific strength, corrosion resistance, and formability. Meeting the stringent reliability standards in these sectors relies on the melting–casting process; however, characterized by complex multi-process coupling and small, imbalanced datasets, this process poses significant challenges for conventional modeling approaches in achieving accurate quality classification. To address these challenges, we develop a SCADA–MES integrated intelligent manufacturing data management system tailored to the full aluminum-alloy melting and casting route, providing dynamic, high-dimensional, and cross-process data support for predictive modeling. Building on this end-to-end infrastructure, we propose a deep-learning-based quality classification method that integrates Conditional Tabular Generative Adversarial Network (CTGAN), Self-Attention and Inter-sample Transformer (SAINT), and Convolutional Neural Network (CNN). CTGAN performs data augmentation by generating synthetic samples consistent with the empirical distribution under categorical and process-condition constraints, thereby alleviating the issues of data scarcity and imbalance. SAINT employs a dual-channel attention mechanism to capture nonlinear dependencies across both features and samples, while CNN extracts stable local patterns and attenuates noise. Evaluations of the algorithm’s performance in comparison with the baseline approach indicate the proposed method achieves approximately 5% improvement in all key metrics such as accuracy, macro F1 and recall, which demonstrates outstanding accuracy and stability in multi-stage quality prediction across the entire melting-casting process and thus can provide a practical pathway for data-driven quality control of high-performance aluminum alloy.