Computer Forecasting of Butt-welding Quality of Reinforcing Profiles Using Machine Learning Methods

The study investigates mathematical and computer-based modeling of butt welding of galvanized steel strips employed in the fabrication of reinforcing profiles for window frame systems. The motivation of the research lies in the necessity to improve weld quality and stabilize production processes in industrial window manufacturing. The primary aim is to establish predictive  models capable of accurately estimating the structural strength of welded profiles from critical welding parameters. Experimental datasets were processed and analyzed through state-of-the-art Data Science techniques, including regression analysis, logistic modeling, and machine learning. Python-based libraries (Pandas, Scikit-learn, Seaborn) were utilized for data preprocessing, cleaning, and visualization. Several regression models were proposed, with polynomial models of the second and fourth degree yielding the most adequate representation of the process. In particular, the fourth-degree model demonstrated superior predictive capability, confirming a complex nonlinear dependency between welding current, heating time, strip thickness, and weld strength coefficient. The results highlight the effectiveness of integrating regression-based methods and machine learning into digital manufacturing frameworks, enabling optimization of butt-welding regimes and providing valuable tools for industrial quality control and automated production systems.