Deep learning-enabled turbulence model optimization of solid motor

One of the primary methods for studying the ignition process of solid motors is numerical computation using turbulence models. However, the extensive use of empirical parameters in turbulence models often results in limitations that hinder the accurate description of the ignition process of solid motors using numerical methods. To address this issue, this study proposes a model optimization method based on deep learning for the ignition process of solid motors. This method optimizes two empirical parameters in the Spalart-Allmaras turbulence model. An improved ConvNeXt neural network was developed to identify the relationship between empirical parameters and ignition pressure data. This study creatively proposes a method of freezing the model through the frozen layer technique and updating the empirical parameters using an optimizer. Finally, the optimized empirical parameters are re-input into the Spalart-Allmaras turbulence model to analyze the optimization effect. The percentage error of the ignition pressure data output by the ConvNeXt neural network is less than 3%, demonstrating its ability to accurately capture the relationship between empirical parameters and ignition pressure data. The accuracy of the ignition pressure data calculated by the optimized turbulence model is improved by approximately 18% compared to the ignition pressure data calculated using the original empirical parameters. The method proposed in this study is effective in improving the accuracy of simulation calculations and has application value.