When cone penetration testing is used to evaluate the undrained strength of cohesive soils, the conventional cone factor shows considerable dispersion and is not suitable for partially drained penetration. To address this problem, an inversion method for undrained strength based on physics-informed neural network was proposed. Spherical cavity expansion theory provides the physical constraint. Separate neural networks were constructed for the elastic and plastic zones. We incorporated the cavity wall limit pressure into the loss function as observational data, enabling direct inversion of undrained strength. The method was then applied to silty clay and clayey silt using the relationship between cone tip resistance and normalized penetration rate. Comparisons with analytical solutions from spherical cavity expansion theory confirmed the numerical reliability of the proposed method. Centrifuge and field tests were conducted on four cohesive soils. The inverted undrained strength showed errors generally within 20%. These results validate the method's applicability across different soil types and drainage conditions.

Most offshore wind turbines at home and abroad use large diameter monopile foundation, and the control condition of large diameter monopile design is usually the response under horizontal load. The traditional API method is mainly used for slender piles, and the PISA method developed in recent years uses four kinds of springs to represent the reaction force of soil to piles, which is more suitable for monopiles with low aspect ratio in principle, but there is great controversy about the calculation of ultimate horizontal resistance in PISA method. The typical sand parameters in the South China Sea are selected for finite element analysis, and the key parameters such as soil relative density, pile diameter and aspect ratio are changed, and the response curves of pile foundation are compared. It is found that the three widely used expressions of ultimate horizontal resistance have obvious limitations. Based on the finite element analysis of a large number of variable parameters, it is suggested to express the change of ultimate horizontal resistance with depth in two stages.