Multi-fidelity Gaussian process modeling of a thin-walled structure for laser powder bed fusion (LPBF) process window

Abstract

The laser powder bed fusion (LPBF) method has recently been adopted by industry due to its advantageous representation of complex shapes such as tire sipes. In this study, thin-walled specimens inspired by tire sipes are investigated. During the LPBF manufacturing process, many process variables affect the resulting product properties. There is no method to explicitly define the relationship between input parameters and output properties. Material density is the most fundamental property to examine in additive manufacturing. Numerous studies have been conducted on predicting porosity in metal-based additive manufacturing, primarily focusing on density data obtained exclusively through the Archimedes method. With the Archimedes method, data can deviate due to certain measurement conditions, such as fluid temperature and sample size. Due to the uncertainty of the Archimedes method, the proposed study additionally considers a high-fidelity micrographic observation approach. However, high-fidelity data are not as easy to obtain as low-fidelity data. This research proposes a multi-fidelity Gaussian process model to contribute to the construction of a robust data-driven model. The proposed multi-fidelity Gaussian process modeling framework is performed successfully to construct a process window cost-efficiently.