SHIELD TUNNEL GASKET SECTION OPTIMIZATION BASED ON DEEP NEURAL NETWORK SURROGATE MODEL

The key to ensuring the waterproof design performance of a shield tunnel segment joint is a reasonable cross-sectional shape of an elastic rubber gasket. The frequent nonlinear calculations linked to large material deformation and contact analysis in the gasket section optimization design process limit the optimization efficiency. A structural optimization approach compiled with a deep neural network surrogate model is proposed, with the closure pressure and the ratio of effective contact pressure as the dual control objectives. The depth neural network surrogate model is utilized in the framework of the genetic algorithm to swiftly transform the section form into the contact stress field; Simultaneously, the use of transfer learning allows for the reuse of knowledge from surrogate models with variable sectional shapes, enabling for the creation of high-precision contact stress prediction models with few samples. As a result, the optimization efficiency of a sealing gasket structural section under its closure pressure constraint is significantly increased.