Predicting and Managing weld induced Distortion of thin-walled steel structures

The drive towards lighter ships and thinner plate is restricted by the significant increase in distortion as the plate thickness decreases. Although welding has been the preferred process for metal joining for the last fifty years, distortion of the welded structures remains a major problem - typically, for a 6 mm thick plate distortions can be on the order of 60 mm. A recent study by the US Naval Sea Systems Command has estimated that the cost of distortion can up be to $3.4 million (approx. 2 million) per ship. While it is not expected that distortion can be eliminated completely, a reduction in the magnitude of the distortion will reduce significantly the costs to UK industry.In this work, the neural network approach, in conjunction with experimental measurements and the finite element method will be used to study the relationship between distortion of welded ferritic steel plates and the design parameters. The two key aspects of the problem, which will be investigated, are the interaction of process and production parameters in causing distortion and the influence of pre-existing (residual) stresses in the plate. By modelling the distortion process using material, design and welding parameters, the parameters can be optimised to minimise the resulting distortion. An existing artificial neural network (ANN) will be extended to allow examination of the distortion of the welded plate. The ANN will be trained, using results from measured plate and those obtained from a finite element code, and validated by the experimental work undertaken in the research program, to enable it to estimate plate distortion under a wide range of conditions. The combined effort will thus identify the parameters which cause distortion, assess the significance of each parameter and propose techniques to reduce distortion in welded plate.