Related Experiment Video
Updated: Jun 6, 2026
09:17
Surrogate Model Development for Digital Experiments in Welding
Published on: March 28, 2025
Optimization of BP neural network based on Harris Hawk algorithm to predict resistance spot welding quality
Shuwan Cui1,2, Xuan Zhou1, Haijun Xu3
1School of Mechanical and Automotive Engineering, Guangxi University of Science and Technology, Liuzhou, 545006, China.
Scientific Reports
|June 4, 2026
Summary
A new Harris Hawk Herd Algorithm Optimized BP Neural Network (HHOBP) model accurately predicts resistance spot welding (RSW) quality. This advanced AI approach enhances weld monitoring in industries like automotive and aerospace.
Area of Science:
- Materials Science and Engineering
- Artificial Intelligence in Manufacturing
- Quality Control and Assurance
Background:
- Resistance spot welding (RSW) is crucial for automotive and aerospace manufacturing, but weld quality is paramount.
- Current methods for monitoring RSW quality can be insufficient for ensuring high-integrity joints.
- Predictive modeling offers a path to enhance online monitoring and quality control in RSW processes.
Purpose of the Study:
- To develop an advanced neural network-based online monitoring model for resistance spot welding (RSW) quality.
- To utilize the Harris Hawk Herd Algorithm for optimizing a Backpropagation (BP) neural network for enhanced prediction accuracy.
- To evaluate the performance of the proposed HHOBP model against other established machine learning techniques.
Main Methods:
- A neural network model was developed using welding current, voltage, air pressure, and time as input parameters.
- The Harris Hawk Herd Algorithm was employed to optimize the BP neural network (HHOBP) for predicting weld nucleus diameter.
- The HHOBP model's predictive accuracy was benchmarked against PSO-BP, GWO-BP, CS-BP, GA-BP, traditional BPNN, and support vector machine models.
Main Results:
- The HHOBP model achieved a high R-squared coefficient of 0.9981, demonstrating excellent weld quality prediction capabilities.
- HHOBP outperformed other models across various error metrics, including MAE (93.06%–96.15%), MSE (99.43%–99.81%), RMSE (92.18%–95.46%), and MAPE (87.54%–96.83%).
- The study highlights the superior modeling and prediction accuracy of the HHOBP approach for online RSW quality monitoring.
Conclusions:
- The Harris Hawk Herd Algorithm Optimized BP Neural Network (HHOBP) provides a highly accurate and reliable method for online resistance spot welding quality monitoring.
- This AI-driven approach can significantly improve weld integrity and process control in critical manufacturing industries.
- The HHOBP model represents a substantial advancement in predictive quality assessment for resistance spot welding applications.