Development of a mathematical model for coating thickness in continuos hot-dip galvanizing li̇ne

Galvanized steel is widely utilized in automotive manufacturing, construction, and household appliances due to its high corrosion resistance, mechanical strength, and long service life. The globally high per-capita consumption of galvanized steel highlights the industrial and economic significance of accurately controlling coating thickness. Achieving the desired coating mass is crucial not only for ensuring product durability under varying environmental conditions but also for minimizing zinc usage and improving production efficiency. These requirements make predictive modeling of the air-jet-assisted wiping process an essential component of modern galvanizing operations. In this thesis, a fully dimensionless and nozzle-geometry-independent mathematical model is developed for predicting coating thickness in continuous hot-dip galvanizing lines employing air-jet wiping. Conventional models relying on dimensional variables such as inlet pressure or the z/d ratio show limited generalizability because different nozzle geometries produce different exit jet velocities under identical operating conditions. To overcome this limitation, the fundamental dimensionless parameters governing the process are identified using the Buckingham Pi Theorem, enabling a physically consistent and geometry-independent modeling framework. To construct the physical model, the gas–jet interaction was first analytically examined, and an analytical model was developed to estimate the coating thickness based on the shear stress and pressure gradient acting on the strip. To compute these quantities, a two-dimensional, single-phase, RANS-based CFD model—considering only the air phase—was developed, and a mesh independence study was conducted to verify the numerical accuracy. In line with the objective of achieving a nozzle-geometry-independent formulation, the nozzle inlet pressure (𝑃 𝑗𝑒𝑡 ) was defined as a boundary condition and linked to the jet exit velocity (𝑉 𝑗𝑒𝑡 ) within the CFD simulations. Considering industrial operating conditions, CFD analyses were carried out for 990 distinct cases, and the results were nondimensionalized through a systematic dimensional analysis. Experimental measurements collected from Borçelik's CGL-3 line are statistically verified for reliability, enabling comparison between physical predictions and real plant observations. Since industrial data are naturally clustered within a narrow domain, CFD results covering a wider parameter space are incorporated to enhance the model's generalization capability. Residuals from the physics-based preliminary model are corrected through supervised learning, and five algorithms are evaluated. XGBoost provides the best predictive performance, yielding R² = 0,99 and a mean relative error of 2,35% during training, while previously unseen validation conditions result in a mean error of 4,91%. In addition, the uncertainty analysis revealed that the overall uncertainty is predominantly governed by the experimental parameters, while the contribution of the numerical component remains minimal due to the mesh-independent nature of the simulations. The final xix hybrid and dimensionless model offers a reliable, scalable, and nozzle-geometry-independent tool for coating thickness prediction in continuous galvanizing processes. Its physically grounded formulation and strong predictive accuracy make it highly suitable for integration into industrial process control strategies.