Welding distortion prediction of ship plate frame structures based on a self-learning database

Welding distortion will increase the shipbuilding cost. Therefore, it is important to accurately predict the welding distortion of a structure during the design stage. In this paper, a novel welding distortion prediction method for ship plate frame structures based on a self-learning database was proposed. Based on a typical plate frame structure in a shipyard, the applicability of the method was verified by comparing the prediction with on-site measurements. To solve the problem of choosing the welding sequences of structures with too many welds, the Latin hypercube sampling method was selected to arrange the weld samples in sequence. The results showed that the distortion of the structure was mainly concentrated on the upper shell. The numerical simulation coincided with the measurement results, and the accuracy was within 20%. Moreover, a reasonable welding sequence can decrease the distortion of a structure by 11.2%.