Wire arc additive manufacturing (WAAM) introduces challenges in studying and predicting microstructures due to rapid solidification phenomena [1]. Accurate prediction of solidification microstructures at welding bead interfaces necessitates a comprehensive, full-field approach. To achieve this, precise knowledge of the transient temperature field during rapid solidification is crucial for solving a coupled thermal equation with a phase evolution equation [2],[3]. This study presents a novel method to expedite computation time by leveraging experimental data, extracted through image recognition, providing cost-effective insights into solidification microstructures [4].
To avoid solving a coupled thermal problem, we suggest identifying morphological texture zones from optical images of welding bead contours. The connectivity and continuity of these zones are studied using a convolutional neural network (CNN) method based on scanning electron microscope (SEM) images. Finally, modeling of the microstructure of a mm-sized weld bead is performed using Kobyashi-Warren-Carter phase field model [3], based on coupled solving of phase, temperature and crystallographic orientation variable. EBSD Images serve as input data to adjust and verify this latter method. Figure 1a illustrates the deposition strategy. Figures 1b and 1c show respectively a light optical observation and a simulation of crystal orientation in a typical bead. A mapping of several neighboring beads is also performed with information derived from the previous tools.
The combined use of minimal experimental input, image segmentation and artificial intelligence-based phase field modelling leads to the following outcomes:
●    the prediction of the crystal texture in the entire bead volume by a phase field approach,
●    the phase field simulation of the interfaces between neighboring beads considering the real process chronology,
●    a strongly reduced computation time.