PSI - Issue 38

Alexander Raßloff et al. / Procedia Structural Integrity 38 (2022) 4–11

5

2

A. Raßlo ff et al. / Structural Integrity Procedia 00 (2021) 000–000

Process characterisation

Microstructure characterisation

Material model

Process

Demonstrator

Mechanical properties

Process model

Defect analysis (CT)

Data analysis

Knowledge-based Engineering

Fig. 1. Schematic overview over processes within AMTwin for the data-driven prediction of PSP relationships.

1. Introduction

Additive manufacturing (AM) paves the way to customised, lightweight and innovative materials. The process of qualifying new materials with desired properties and their reliable utilisation requires more than ten years according to the National Science and Technology Council (2011). Understanding the relationships between process, structure and properties is essential for accelerating materials development. Deriving a large database of associated process structure and structure-property pairs based on experiments and augmented by numerical simulations allows for an inverse materials design. However, a major challenge in transforming AM into the e ffi cient and reliable commercial application can be seen in the lack of su ffi cient and systematic knowledge about process-structure-property (PSP) relationships. Hence, AM components often contain imperfections and inhomogeneities that cause premature failure especially under cyclic loading. Therefore, further development of methods for material qualification, structure and process simulation as well as component construction and quality control is necessary and addressed within the project AMTwin . The goal of AMTwin is the data-driven prediction of PSP relationships for the widely used metal alloy Ti-6Al 4V manufactured by laser powder bed fusion (LPBF). A multitude of steps as illustrated in Figure 1 is necessary to achieve this goal by accumulating enough data through data fusion, i.e. the augmentation of few experimental data by numerical simulations. A digital process chain and workflows are established to create a digital twin of the AM process and material. Starting with a detailed analysis of the Ti-6Al-4V powder, all data and metadata is coherently stored. Many specimens are built by design of experiment through variation of the LPBF process parameters to cap ture a significant range of microstructures and associated properties, allowing for both establishing process-structure relationships and the derivation of a process model for data augmentation. The microstructure is experimentally char acterised by light microscopy (LM) and electron back-scatter di ff raction (EBSD). The resulting images are to be statistically analysed and serve as basis for the in silico reconstruction of grain microstructures. The material’s me chanical properties are determined by static and cyclic tests. Capturing the pores by x-ray computed tomography (CT) is essential for a comprehensive microstructure analysis. Based on these experimental data, a process model is de rived that simulates the LBPF process. Additionally, methods are developed to reconstruct microstructures based on statistical and translation-invariant descriptors, see Seibert et al. (2021). A low-dimensional description in the form of few meaningful scalars that capture the crucial microstructural features are key for establishing the PSP relationships. The reconstructed SVEs – synthetic structures that are as a single one not representative for all possible instances of a microstructure associated with that statistic, but as a set – serve as simulation domains for the numerical crystal plas ticity (CP) simulations. Through modelling the material behaviour they yield mechanical properties such as fatigue indicator parameters (FIPs) or yield stress. Processing them yields the properties which are incorporated in the data base. Finally, the latter one can be used to establish the desired PSP relationships. As a show case of this approach, the current article aims at presenting an experimental-numerical approach and an exemplary demonstration for the investigation of the influence of pores on the fatigue properties by numerical simulation as illustrated in Figure 2. To that end, specimens of Ti-6Al-4V are manufactured and their microstructure statistically characterised on the basis of LM images and CT scans. SVEs are reconstructed and CP simulations are