Prediction and Optimization of Load-Bearing Capacity in Resistance Spot Welded Titanium Joints Using Neural Networks and Genetic Algorithms

Abstract

Abstract

This study investigates the mechanical performance of resistance spot-welded titanium lap joints made of Grade 1 and Grade 5 alloys. Experimental tests were combined with artificial neural network modeling to predict joint load-bearing capacity based on welding current and welding time. Three models were developed for Grade 1/Grade 1, Grade 1/Grade 5, and Grade 5/Grade 5 joints. The mixed Grade 1/Grade 5 joint achieved the highest predictive accuracy, with an R2 value of 0.9289. Statistical evaluation confirmed high model reliability, with mean relative errors between four and six percent. The most accurate model was optimized using a genetic algorithm. The algorithm identified an optimal parameter set consisting of a welding current of 2.89 kA and a welding time of five pulses. This configuration produced a predicted load-bearing capacity of 3.2 kN, which meets the required threshold of three kilonewtons. Contour maps showed that the optimal point lies near the boundary of the high-strength region and corresponds to the lowest welding current and shortest welding time that still ensure sufficient joint quality. The results demonstrate that combining neural network modeling with evolutionary optimization is an effective approach for designing efficient welding processes for dissimilar titanium joints.