In-situ observation of sintering process of tape-cast ceramics utilizing deep learning and its application to the prediction for multilayer deformation

Abstract

The co-sintering of multilayer ceramics, produced by tape casting, is a cost-effective process for manufacturing electrodes in solid oxide fuel cells (SOFCs). Hence, understanding the deformation kinetics of multilayer ceramics during co-sintering, originating from the different shrinkage rates of individual layers, has become critical. In-situ monitoring of the sintering process using an optical device is a promising non-contact method that enables the deformation tracking of both monolayer and multilayer ceramics. However, the material emits thermal radiation at various levels during sintering, which leads to a time-consuming image processing procedure, because the intensity of thermal radiation changes significantly when shifting from low to high temperatures. This study presents deep learning techniques that employ convolutional neural networks to segment and detect objects in the image processing associated with in-situ monitoring. We verify that the DeepLab V3+ network accurately segments the sintered body, even when images are severely affected by mixed brightness or noise. Furthermore, the presented network can also quantify shrinkage profiles over a wide temperature range within a short time. In addition, we demonstrate that the YOLO V4 network can monitor the warpage behavior of co-sintered laminates by evaluating the curvature profile over a broad range of temperatures without requiring image binarization, despite the limitations of curvature detection. Finally, we demonstrate how the developed networks can potentially be utilized to predict the time evolution of warpage deformation in SOFC anode/electrolyte bilayer systems.

Keywords

• solid oxide fuel cells
• sintering
• co-sintering
• deep learning
• segmentation
• object detection