Abstract: The microstructure and mechanical properties of low-pressure cast aluminum alloy wheels directly affect their service life, but subsequent quality inspection is usually required to obtain their microstructure and property data, making online prediction difficult. This paper addresses this challenge by proposing a data-driven machine learning method to construct a prediction path from mold temperature to microstructure and then to mechanical properties. By sampling data on the mold temperature-microstructure characteristics-mechanical properties of the target wheel, a multimodal dataset including mold temperature curves, secondary dendrite arm spacing (SDAS), yield tensile strength, and elongation at break is constructed. A comparative study of microstructure and property prediction models is then conducted. The results show that in the SDAS prediction task, the Transformer-BiGRU nested model performs best, with an R2 of 0.932 on the test set. In the mechanical property prediction task, the random forest model provides the best prediction effect for yield strength, while the support vector regression model provides the best prediction effect for tensile strength and elongation. The constructed predictive framework can rapidly predict microstructure and properties based solely on online mold temperature. It can accurately characterize the correlation between mold temperature, solidification structure, and mechanical properties during low-pressure casting of aluminum alloys, with the prediction error of each indicator consistently controlled within 10%. This data-driven machine learning prediction method provides an effective approach for online quality control and process optimization of low-pressure cast aluminum alloy wheels, and is of great significance for subsequent tuning of aluminum wheel process parameters and improvement of production efficiency.