The Digital Weldability Test: Predicting Laser Weld Soundness from Input Parameters Using Machine Learning

This thesis investigates the weldability and mechanical performance of nickel-based superalloys, with a primary focus on Inconel 718, using laser beam welding (LBW) techniques and artificial neural network (ANN) modeling. Nickel-based superalloys are widely utilized in high-temperature and corrosive environments such as aerospace, power generation, and chemical industries due to their exceptional mechanical strength, thermal stability, and resistance to oxidation and corrosion. However, their welding presents significant challenges, including crack formation, microstructural instability, and the formation of undesirable intermetallic phases such as Laves phases. The study begins with a comprehensive review of nickel-based alloys, their compositions, and conventional and advanced welding techniques. Particular emphasis is placed on laser beam welding, which offers advantages such as reduced heat input, narrow heat-affected zones, improved weld precision, and minimized distortion compared to traditional methods. The research compiles an extensive database from existing literature, including welding parameters such as laser power, welding speed, alloy composition, and material thickness, alongside corresponding weld quality outcomes. A detailed analysis is conducted to understand the relationship between process parameters and weld quality, including the influence of aluminum and titanium content, laser power, and welding speed on mechanical properties such as tensile strength and hardness. Experimental findings from literature indicate that laser welding can produce high-quality joints with reduced defects, although challenges such as reduced ductility and phase formation persist under certain conditions. To address the complexity and nonlinearity of welding parameter optimization, an artificial neural network model is developed using MATLAB. The ANN is trained on the compiled dataset to predict weld quality based on input parameters. The model demonstrates strong learning capability, effective convergence behavior, and satisfactory generalization in predicting weld outcomes. Performance evaluation using training results, confusion matrices, and classification metrics confirms the reliability and efficiency of the proposed model. Overall, this research highlights the potential of combining advanced laser welding techniques with data-driven approaches such as ANN to optimize welding parameters and improve the performance of nickel-based superalloy joints. The findings contribute to enhanced process understanding and provide a foundation for intelligent welding system development in advanced manufacturing applications.