In float glass production, thermomechanical stresses that develop during the cooling process critically affect product quality. Deviations from the optimal parameters often lead to significant production losses. Currently, this process is manually controlled, which is error-prone due to the complexity and interconnectivity of the parameters involved. In this work, we explore the use of process data to predict thermomechanical stresses in the cooling area. Our approach began with an analysis of this data to identify key features, followed by data cleaning to prepare for AI applications. We then trained various AI models, adjusting their hyperparameters to optimize performance. These models were benchmarked to select the most effective one, iteratively focusing on models that delivered superior benchmark results until a final model was established. Our research demonstrates promising results, with the potential for initial prototype testing. Implementing this model could streamline quality control, enhance output optimization, and, if extended to automatically control the cooling process, could substantially reduce production losses. Such advancements could significantly impact global glass production.