Predictions of compression capacity of randomly corroded spherical shells based on artificial neural network

Spherical pressure hull is a key buoyancy component of deep-sea submersibles. Corrosion on the surface of spherical shell can significantly reduce its compression capacity. Pitting corrosion is a typical form of corrosion in steel structures. Artificial neural network (ANN) is utilized to predict the compression capacity of spherical shells with random corrosion and determine influencing factors of corroded spherical shells. Corroded diameter D-C/D, corroded thickness T-C/T, and mass loss ratio chi are utilized to characterize corrosion severity. The influence of the corroded diameter D-C/D and the corroded thickness T-C/T on the prediction accuracy is investigated in this study. The applicability of trained ANN for spherical shells with different geometric sizes is validated. Results indicated that ANNs can be utilized for high-precision compressive capacity prediction and the corroded diameter D-C/D, corroded thickness T-C/T, mass loss ratio chi, and the corroded area can be used as input variables.