This study presents the development of deep learning (DL) models to predict two critical welding Key Performance Characteristics (KPCs) - welding depth and average pore volume - in hairpin welding. The models utilize a wide range of welding Key Input Characteristics (KICs) as inputs, including welding beam geometries, feed rates, path repetitions, and bright light weld ratios.
The dataset was obtained from a series of hairpin welding experiments conducted on 12 unique welding geometry configurations, each with 3-4 welding paths. The welding depth and average pore volume were measured using computed tomography (CT) scans.
Exploratory data analysis revealed strong correlations between certain KICs, such as the bright light weld ratio, and the welding depth. The pore volume was found to be strongly correlated with the number of path repetitions.
Two DL models were developed - one to predict welding depth and another to predict average pore volume. The welding depth prediction model achieved a Mean Absolute Error (MAE) of 0.1079 on the validation dataset, while the pore volume prediction model achieved an MAE of 0.0641. The results demonstrate the capability of DL in capturing the complex nonlinear relationships between the welding input parameters and the critical weld characteristics.
This research represents a significant step towards implementing data-driven quality assurance in laser welding processes, moving beyond the current trend of relying solely on defect classification. The developed models can enable better control and optimization of welding outcomes, leading to improved productivity and quality in industrial applications.
To Another Language
from source content
arxiv.org
Deeper Inquiries