Pressure Distribution Prediction for Submarine Using Unet-Based AI Model

In this study, a preprocessing method for submarine hull form and hydrodynamic pressure distribution were developed to enable an AI model to predict pressure distribution for the corresponding hull form. An Unet-based symmetric model including convolution layers was trained. The training hull form data set was generated by modifying DARPA Suboff 5475 model and the corresponding pressure distribution data set was calculated by solving RANS equation. The AI model was designed to extract the feature maps of body, sail, and stern parallelly to learn their hydrodynamic interaction. As a result, the trained model's predicted pressure distribution closely matched the CFD results, showing not only similar contours but also a total resistance coefficient within 2% error margin. However, the model shows limitations in predicting highly non-linear behavior of pressure distribution for the combination of two deformation on stern: afterbody and stern appendages.