Swarm Intelligence-based Modeling and Multi-objective Optimization of Welding Defect in Electron Beam Welding

The process parameters involved in electron beam welding have major influences on spiking and penetration efficiency of the joint. In addition to this, the input-output relationships of electron beam welding are nonlinear and complex in nature. Therefore, adaptive neuro-fuzzy inference system (ANFIS)-based input-output modeling had been attempted to predict the spiking severity in electron beam welded joints. Swarm-based optimization algorithms, like grey wolf optimizer, particle swarm optimization, and bonobo optimizer (BO), were used for optimizing the ANFIS architecture and predicting the response precisely. Multi-objective bonobo optimization (MOBO), Multi-objective grey wolf optimization, and Multi-objective particle swarm optimization algorithms had been used for solving the conflicting multi-objective criteria problems associated with the study. The input process parameters were accelerating voltage, beam current, scan speed, focusing distance, and beam oscillation parameters, whereas mean weld-bead penetration and its standard deviation were considered as the responses of the system. The irregular penetration, that is, spiking was expressed in terms of the standard deviation of weld penetration. The accuracy level of optimization and modeling had been tested with some test cases obtained through the real experiments. MOBO had shown the better accuracy in predicting the optimized set of input parameters, which could satisfy both the spiking and penetration criteria, while BO-ANFIS had shown the superior efficiency in predicting the response with minimum error.