Interpretable prediction of ultimate strength of composite pressure hulls using data-driven symbolic regression

Pressure hulls of submersibles are typically manufactured from multilayered composite cylindrical shells to withstand deep-sea hydrostatic pressure. Accurate prediction of their ultimate strength (UTS) remains challenging due to complex post-buckling behavior and strong coupling among multiple design parameters. This work presents an interpretable data-driven approach that establishes universal expressions for predicting the UTS of composite pressure hulls. Experimentally calibrated finite element analysis (FEA) is first performed to build a comprehensive failure database covering various design parameters. Then, feature engineering and symbolic regression are employed to extract critical descriptors and derive explicit UTS formulas. The formulas are validated against published data, with errors below 13%. Furthermore, comparisons between the derived and existing empirical formulas, such as ASME 2007 and NASA SP-8700, based on our FEA database demonstrate that the proposed formulas deliver markedly superior performance. Unlike the black box nature of conventional AI-based models, the derived formulas possess clear physical interpretability, directly revealing the relationships between design variables and the corresponding UTS. This enables direct design of lightweight or high-performance composite pressure hulls through rational parameter selection. As a result, this work provides a practical strategy for developing data-enhanced predictive models to support the design and failure evaluation of composite hulls.