Multi-objective process parameters optimization of Ti-6Al-4V LPBF parts through a hybrid prediction–optimization approach

Abstract

In recent years, laser powder bed fusion (LPBF) has attracted increasing interest from industries and researchers due to its great potential in producing complex geometries. Many studies focus on the effect of process parameters on the final quality of the printed part. Nevertheless, due to the complexity of the LPBF process, it is hard to establish the mutual and intricate relationship between process parameters and part performance. For this reason, machine learning (ML) techniques have been adopted to obtain the best process condition regarding porosity or other optimization functions. Most of the time, it is necessary to satisfy more than one request to ensure the quality of the product. For this reason, a multi-objective (MO) optimization approach, which combines artificial neural networks (ANNs) and multi-objective metaheuristic algorithms, was developed. In this study, the MO optimization approach aimed to identify the optimal set of process parameters to assure the best compromise between strength and ductility for the LPBF of Ti-6Al-4V. Three multi-objective metaheuristics, namely non-dominated sorting genetic algorithms (NSGA-II), multi-objective grey wolf optimization (MOGWO), and multi-objective particle swarm optimization (MOPSO), were implemented and compared in terms of Pareto optimal solutions. Finally, the identified Pareto fronts were validated by three sets of additional experimental tests. The results show that a scan strategy of 0° must be avoided to obtain good results in terms of strength and ductility. Moreover, it is also clear that line energy density alone is not enough to predict the quality of the final part. The proposed model can help engineers in the design phase to meet multiple requests when considering the influence of several process parameters, as for the LPBF process.