Width Prediction of Intersecting Line Welds in Pipeline Pressure Vessels Based on VAE-PSO-DNN

Precise control of the weld width at intersecting lines between pipes and pressure vessels is crucial for ensuring the sealing performance, structural integrity, and service life of the equipment. However, under complex welding conditions, the coupling effects of multiphysics fields render the weld formation mechanism highly intricate. Traditional inspection methods suffer from limited precision and poor timeliness, often failing to meet practical production requirements. To address this challenge, this paper proposes a VAE-PSO-DNN-based method for predicting the weld width at pipe-pressure vessel intersecting lines, constructing an integrated predictive model. This model leverages the strengths of data augmentation, the Particle Swarm Optimization (PSO) algorithm, and Deep Neural Network (DNN) for ensemble prediction. During the model training phase, the RMSprop adaptive learning rate algorithm was employed to ensure efficient and stable convergence. The results demonstrate that the proposed model exhibits outstanding performance in the weld width prediction task. Its Root Mean Square Error (RMSE), Mean Absolute Error (MAE), Mean Bias Error (MBE), and Coefficient of Determination (R2) reached 0.0349, 0.0282, −0.0029, and 0.9980, respectively. Compared to non-optimized DNN models, Artificial Neural Networks (ANN), and traditional machine learning algorithms, these metrics show significant improvement.