Hybrid prediction-optimization approaches for maximizing parts density in SLM of Ti6Al4V titanium alloy

• Published in: Journal of intelligent manufacturing Vol. 33; no. 7; pp. 1967 - 1989
• Main Authors: Costa, A., Buffa, G., Palmeri, D., Pollara, G., Fratini, L.
• Format: Journal Article
• Language: English
• Published: New York Springer US 01.10.2022; Springer Nature B.V
• Subjects: Advanced manufacturing technologies; Algorithms; Artificial neural networks; Business and Management; Control; Design of experiments; Heuristic methods; Laser beam melting; Machines; Manufacturing; Mathematical models; Maximization; Mechatronics; Optimization; Physical properties; Porosity; Prediction models; Process parameters; Processes; Production; Response surface methodology; Robotics; Sensitivity analysis; Specific gravity; Titanium alloys; Titanium base alloys; Variance analysis; Artificial neural network; Response surface methodology; Optimization; Predictive model; Metaheuristic algorithms
• ISSN: 0956-5515; 1572-8145; 1572-8145
• DOI: 10.1007/s10845-022-01938-9

It is well known that the processing parameters of selective laser melting (SLM) highly influence mechanical and physical properties of the manufactured parts. Also, the energy density is insufficient to detect the process window for producing full dense components. In fact, parts produced with the same energy density but different combinations of parameters may present different properties even under the microstructural viewpoint. In this context, the need to assess the influence of the process parameters and to select the best parameters set able to optimize the final properties of SLM parts has been capturing the attention of both academics and practitioners. In this paper different hybrid prediction-optimization approaches for maximizing the relative density of Ti6Al4V SLM manufactured parts are proposed. An extended design of experiments involving six process parameters has been configured for constructing two surrogate models based on response surface methodology (RSM) and artificial neural network (ANN), respectively. The optimization phase has been performed by means of evolutionary computations. To this end, three nature-inspired metaheuristic algorithms have been integrated with the prediction modelling structures. A series of experimental tests has been carried out to validate the results from the proposed hybrid optimization procedures. Also, a sensitivity analysis based on the results from the analysis of variance was executed to evaluate the influence of the processing parameter and their reciprocal interactions on the part porosity.