Session: 04-22-01: Machine Learning-Based Modeling, Prediction, and Optimization of Advanced Manufacturing and Materials System with Multiphysical Phenomena

Paper Number: 166488

Surrogate Multi-Objective Optimization of LPBF Additive Manufacturing for Improved Part Quality and Fatigue Life 

In Additive Manufacturing (AM), process parameters play a crucial role in determining the quality and properties of the final part, as they are inherently related to the microstructure of the printed component. Microstructure is directly related to mechanical properties such as fatigue, tensile string, and hardness. Therefore, optimizing process parameters in AM is essential for ensuring part quality and mechanical strength. Poor choice of process parameters can result in defects that weaken fusion, leading to porosity and geometric inaccuracy, which adversely affect the fatigue life and structural integrity of the part. In this study, we present a framework for optimization of AM process parameters to improve part quality and mechanical strength. Our goal is to minimize discrepancies in part quality between as-designed and as-built by following a two-step approach. The proposed approach integrates the design of experiments such as Latin hypercube sampling, AM process simulation, cyclic load analysis, surrogate modeling, and multi-objective optimization. In the first step, we simulate an AM process, such as a Laser Powder Bed Fusion (LPBF) process, to capture the effects of process parameters such as laser power, scan speed, hatch spacing, and layer thickness influence on the fabricated part. In the second step, the as-built geometry is subjected to mechanical analysis to determine structural integrity and performance. This methodology creates a direct relationship between process parameters, structure, and mechanical properties. To address the computational cost of this high-fidelity technique, we constructed Gaussian process (GP) surrogate models for efficient multi-objective optimization. The optimization model considers key mechanical properties like fatigue life and geometric accuracy subject to porosity constraints. We used the Genetic algorithm (GA) implanted in MATLAB to optimize and systematically explore the design space, identify the best values for the parameters, and yield a Pareto solution that satisfies the competing objectives. The results indicate that the optimal process parameters obtained through this approach result in superior part quality compared to a part produced with unoptimized process parameters. Also, sensitivity analysis is performed on the surrogate model to understand the effect of each process parameter on the part qualities. By analyzing and optimizing the effects of process parameters, our framework provides a cost-effective solution for evaluating and optimizing AM part qualifications, reducing the need for extensive experimental trials. This study contributes to the existing literature by improving part quality, mechanical strength, and overall reliability, all of which are important for a variety of engineering applications and certification processes.

 

Presenting Author: Betelhiem Mengesha University of Maryland

Presenting Author Biography: Betelhiem N. Mengesha is a mechanical engineering doctoral student at the University of Maryland. She was awarded the National Science Foundation (NSF) Graduate Research Fellowship in 2024. Her research focuses on leveraging additive manufacturing processes to optimize heat exchanger designs while ensuring quality control measures. She aims to create a holistic digital twin that integrates design optimization with the manufacturing process and qualification.

Authors:

Betelhiem Mengesha University of Maryland
Vikrant C. Aute University of Maryland
Shapour Azarm University of Maryland