Prediction of Bead Geometry with Changing Welding Speed Using Artificial Neural Network

Area of Science:

• Materials Science
• Manufacturing Engineering
• Robotics

Background:

• Weld bead geometry is critical for industrial design and quality control.
• Predicting bead geometry in dynamic welding processes is challenging due to complex parameter interactions.
• Existing models struggle to account for real-time variations in welding parameters.

Purpose of the Study:

• To develop an artificial neural network (ANN) model for predicting weld bead geometry.
• To investigate the impact of spatially varying welding speed on bead geometry.
• To enhance quality detection in gas metal arc welding (GMAW) through accurate geometry prediction.

Main Methods:

• Utilized a welding robot for gas metal arc welding (GMAW) with stochastically varied welding speeds.
• Employed transient response tests to determine the spatial influence of welding speed on bead geometry.
• Developed an ANN model with spatial welding speed sequences as input and bead width/reinforcement as outputs.
• Measured bead geometry using a structured laser light sensor and polynomial fitting.

Main Results:

• Demonstrated that changing welding speed has a significant spatial influence on bead geometry (10 mm backward to 22 mm forward).
• The developed ANN model (33-6-2 structure) achieved high prediction accuracy.
• Achieved 99% accuracy on the training dataset and 96% accuracy on the test dataset.

Conclusions:

• The ANN model effectively predicts weld bead geometry under dynamic welding conditions.
• Accurate prediction of bead width and reinforcement is feasible despite changing welding speeds.
• This approach offers a robust solution for real-time quality assessment in automated welding.