Abstract: Aluminum wheels are subject to stringent lightweight requirements and feature complex geometries, which significantly increases manufacturing difficulty. Low pressure die casting (LPDC), as the most widely used manufacturing technology for aluminum wheels, has long faced industry challenges in accurate mold temperature prediction and casting quality control. Based on big data from aluminum wheel low-pressure die casting production, this study establishes a machine learning model combining convolutional neural network and long short-term memory network (CNN-LSTM), revealing the intrinsic relationship among casting process, mold temperature and casting quality.The results show that the CNN-LSTM model can accurately predict the temperature at the wheel spoke position of the mold under different die casting processes, with a root mean square error of 2.82 ℃ between measured and predicted mold temperatures. Taking the mold temperature curve of qualified aluminum alloy wheels as the control target, timely adjustment of process parameters during casting can effectively avoid the formation of casting defects. By comparing the casting quality of aluminum wheels produced under the traditional casting process and the novel machine learning model-driven casting process, the mechanical properties of wheel spokes are significantly improved by the novel process. The tensile strength and elongation at fracture are increased by 19.5 MPa and 3.8 percentage points compared with the traditional process.