The welding station of T-beam in ship section has been operated by robot, but the actual working time differs greatly from the theoretical time predicted by off-line programming software, so it is difficult to reflect the benefits of quantitative dispatching in ship shop. Based on this, this paper proposes a multi-type weld linear model, based on machine learning, training the weights of various types of weld, get the model function, forecast vessel segment T-beam welding station operation time, and can be extended to sub assembly, unit assembly, painting and working time is linear or nearly linear distribution in the workshop.