Prediction and optimization method for welding quality of components in ship construction

Area of Science:

• Industrial Engineering
• Materials Science
• Manufacturing Technology

Background:

• Welding is critical in shipbuilding, representing 70% of workload and 40% of costs.
• Current welding quality prediction methods lack accuracy and timeliness for intelligent manufacturing.
• Ship assembly-welding processes require improved quality control for efficiency and predictability.

Purpose of the Study:

• To develop a data and model-driven method for predicting welding quality in ship construction.
• To address limitations in existing methods, focusing on efficiency, timeliness, and predictability.
• To enhance intelligent welding control through accurate quality prediction.

Main Methods:

• Analysis of welding quality influence factors and process parameter correlations.
• Establishment of a data collection architecture and feature dimensionality reduction for monitoring.
• Development of a fused prediction model using adaptive simulated annealing, particle swarm optimization, and back propagation neural networks.

Main Results:

• Identified key influence factors and established correlations between process parameters and welding quality.
• Implemented a robust data collection and monitoring system.
• Achieved over 90% prediction accuracy in 74 plate welding experiments.

Conclusions:

• The proposed data-driven method significantly improves welding quality prediction accuracy and timeliness.
• The fused optimization and neural network model effectively enhances intelligent welding control in shipbuilding.
• This approach offers a reliable solution for dynamic quality control in complex industrial welding processes.