Description

Context and objective:

Wire Arc Additive Manufacturing (WAAM) is a metal additive manufacturing process that enables the production of large-scale structures with complex geometries. Mastery of this process helps address key challenges related to industrial sovereignty and innovation (topology optimization, functional integration, etc.). However, the resulting materials typically exhibit highly heterogeneous microstructures, residual stresses, occasional volumetric defects, and significant surface roughness. Previous studies [1], [2] have shown that this roughness is particularly critical for high-cycle fatigue performance.

Systematic machining of surfaces, however, severely limits the design freedom enabled by WAAM. As a result, surface finishing treatments (such as water jetting or hammer peening) are often preferred to partially smooth surface irregularities without completely removing them. The goal is therefore to develop a method to predict the effect of such surfaces on fatigue performance.

This is the main objective of the NEREIDS project (Numerical Evaluation of Roughness Effect on In-service Durability of Structures): to establish a fast method for predicting the fatigue performance of a WAAM component while accounting for its surface topography. This study is part of a project call from the Centre Interdisciplinaire Mers et Océan (CIMO) of the Institut Polytechnique de Paris. CIMO promotes, among other things, research on materials, their assemblies and hydrodynamic structures, as well as design, architecture, and propulsion systems.

Content of the study:

To develop a robust predictive tool, reliable experimental data are required. Within the NEREIDS project, the approach is based on data obtained from a thermometric technique: thermoelastic stress analysis (TSA). This technique involves monitoring the evolution of the surface temperature field of a specimen subjected to cyclic loading using an infrared camera. Depending on the test configuration, and based on the local heat equation, the measured temperature signal can be related to surface stress ; this is known as thermoelastic coupling. Consequently, any evolution of the mechanical field at the surface can be monitored using infrared imaging. This is particularly relevant for tracking the initiation and propagation of sruface fatigue cracks subjected to Mode I loading.

Applying this technique to WAAM surfaces is especially valuable, as it allows the estimation, over large areas, of the initiation times of multiple cracks and the monitoring of their propagation. This is useful, for example, for identifying models based on linear elastic fracture mechanics. However, conventional TSA practice only exploits a small portion of the information contained in infrared recordings. Moreover, traditional infrared image processing methods are highly sensitive to measurement noise, heat conduction effects (blurring), crack interactions, crack closure, ... These approaches also rely heavily on threshold definitions, which are strongly operator-dependent parameters.

In this context, the project proposes the development of an innovative post-processing method based on neural networks. This method aims to fully exploit the richness of infrared data and thereby provide more robust crack monitoring, as well as enable access to additional indicators, such as the presence of process-induced residual stresses near cracks. Furthermore, such a post-processing protocol would not, a priori, be limited to WAAM surfaces.

The results obtained from this method will then be used to develop a predictive approach for the fatigue performance of as-built WAAM structures. This will involve correlating experimental results obtained through the new post-processing method (crack initiation sites, propagation kinetics, preferred paths, etc.) with surface mechanical fields derived from simulations. The objective is to identify meaningful correlations between these datasets. By combining this data comparison with the selection of an appropriate fatigue criterion, the ultimate goal is to generalize the approach to other types of surfaces.

Skills required

The ideal candidate has a background in materials mechanics and a strong interest in numerical methods. Experience with deep learning techniques is not required but would be highly appreciated. The candidate should demonstrate autonomy, curiosity, and a proactive mindset.

Bibliography

[1] M. Renault, L. Bercelli, C. Doudard, B. Levieil, J. Beaudet, et S. Calloch, « Infrared imaging surface roughness criticality assessment of Wire Arc Additive Manufactured specimens », Procedia Structural Integrity, vol. 57, p. 22‑31, 2024, doi: 10.1016/j.prostr.2024.03.004. [2] L. Bercelli, C. Doudard, S. Calloch, V. Le Saux, et J. Beaudet, « Thermometric investigations for the characterization of fatigue crack initiation and propagation in Wire and Arc Additively Manufactured parts with as‐built surfaces », Fatigue Fract Eng Mat Struct, vol. 46, no 1, p. 153‑170, janv. 2023, doi: 10.1111/ffe.13854.

Keywords

Material durability; Thermometry; Additive manufacturing; Surface roughness; Machine learning