Tire wear performance is a critical factor for vehicle safety, economic efficiency, and environmental impact. While physics-based simulations using the Finite Element Method (FEM) can predict tire wear, they suffer from high computational costs, especially in the early design stages where numerous tread patterns must be evaluated. This high cost is primarily due to the iterative nature of wear simulations, which require repeated dynamic analyses for various driving modes. To address this challenge, we propose a deep learning method for rapid prediction of tire wear energy distribution, a key indicator of wear. Our model utilizes a U-Net-based architecture that integrates multiple data types: 2D spatial data, such as contact pressure and tread depth distributions, and 1D structural data, including rubber thickness and complex modulus, etc. The model was trained on a dataset of wear energy distributions generated by FEM simulations. Evaluation on a test dataset demonstrated the model's high accuracy, achieving a Mean Absolute Error (MAE) of 1.95 J/m2 and a Structural Similarity Index (SSIM) of 0.927. These results indicate that our proposed method can capture the key features of wear energy distribution and could serve as an alternative to simulations.
Furuhashi et al. (Wed,) studied this question.