Distortion due to heat input from welding is well known, but difficult to predict. It causes rework, adds cost and may affect strength. The paper addresses the complexity of the problem using a neural network model as a predictor. This gave good agreement with experimental distortion data. The sensitivity of results to different variables, including chemical composition, is reported.

Ship plate distortion results in a decrease in strength and aesthetics, has a detrimental effect on the fabrication process, as well as an increase in production cost due to the need for correction. Welding heat has been regarded as a major contributor to distortion. Assessing the distortion problem, without making assumptions, is extremely difficult due to the complexity of the variables involved, e.g., welding procedures, materials, design, and geometry. This paper develops a neural network model to predict the topology of ship plate distortion. The developed model presents good agreement with actual distortion data that have been obtained. In terms of variable sensitivity, chemical composition should be taken into account.