Multi-objective accelerated process optimization of mechanical properties in laser-based additive manufacturing: Case study on Selective Laser Melting (SLM) Ti-6Al-4V

Abstract

Process optimization of Laser-Based Additive Manufacturing (LBAM) systems is often complicated by the tradeoff between different mechanical properties as well as the relative density window. For instance, parts with similar relative densities can have noticeably different tensile mechanical properties (e.g., elongation-to-failure, yield strength, ultimate tensile strength, Young’s modulus). This phenomenon can be attributed to the variation of size and distribution of fabrication-induced voids within the final parts. To overcome the aforementioned challenge, we applied an efficient sequential multi-objective process optimization framework to optimize the quality of LBAM-fabricated parts with respect to multiple non-correlated mechanical properties within the optimal relative density regime. The applied Multi-objective Accelerated Process Optimization (m-APO) method indirectly accounts for the effect of size and distribution of voids on the final parts’ mechanical properties. The m-APO decomposes the master multi-objective optimization problem into a sequence of single-objective sub-problems constructed from mathematically convex combination of individual unknown objective functions. At each step, the m-APO smartly maps the experimental data from previous sub-problems to the remaining sub-problems. Therefore, the information captured from previous sub-problems is leveraged to accelerate the master multi-objective process optimization problem. The m-APO exhibited capability to achieve a set of process parameter setups, resulting in the best trade-off between conflicting mechanical properties in the optimal window. The m-APO methodology is employed to maximize relative density and elongation-to-failure of Ti-6Al-4 V parts fabricated by Selective Laser Melting (SLM) system. The results show that the m-APO achieved the optimal process parameter setups while reducing the time-and cost-intensive experiments by 51.8%, compared with an extended full factorial design of experiments plan.