Improved CNN for Predicting Line Heating Forming Deformation of Complex Hull Curved Plate

Shipbuilding requires numerous complex curved plates, with manual line heating forming as the primary processing method for the outer plates of the hull. To solve the problem of automation in curved plate forming, this research investigates intelligent deformation prediction for hull curved plate based on improved convolutional neural network (CNN). A numerical calculation model based on the strain direct boundary was established. The feasibility of the model was verified by comparing finite element numerical calculation results with experimental data. To achieve fast prediction of curved plate forming, a deformation prediction model for curved plate forming based on CNN was established. To further enhance the prediction accuracy of the CNN model, particle swarm optimization (PSO) and whale migration algorithm (WMA) were introduced to improve the weights of CNN, PSO-CNN model, and WMA-CNN model were established. Compared with the CNN model, the MAPE of the PSO-CNN model and the WMA-CNN model decreased by 0.31 and 1.36, respectively, for predicting the shrinkage. The MAPE decreased by 1.31 and 4.05, respectively, for predicting the deflection. This results in the WMA-CNN model effectively improving the prediction accuracy compared with the CNN model and PSO-CNN model. The WMA-CNN model is applied for the validation of the case. The relative error of the cases is all within 5%, indicating that the WMA-CNN model is applicable to the intelligent prediction of curved plate forming. This research provides reference for automation forming of hull curved plate.