Automatic Nondestructive Detection of Damages in Thermal Barrier Coatings Using Image Processing and Machine Learning

July 22, 2022

Andrew Sprague (1), Pouya Tavousi (2), Sina Shahbazmohamadi (2), Zahra Shahbazi (1)
Microscopy and Microanalysis. Volume 28 (22 July 2022). DOI:


Thermal barrier coatings (TBCs) are multilayer coatings meant to insulate gas turbine engine metal component and allow it to operate at elevated temperatures. Typically, a TBC is made from four layers: a ceramic topcoat, a thermally grown oxide (TGO), an aluminum-containing bond coat, and a superalloy substrate. Figure 1 shows a representative image of TBC layers. After certain hours of service time the ceramic topcoat eventually spalls off which can result in the exposure of substrate to melting temperatures. The delamination of the topcoat is attributed to several reasons: i) The growth and linkage of cracks within topcoat ii) the growth in undulations within bond coat and TGO exceeding strains of thermal expansion coefficient mismatch and iii) growth of the TGO thickness. [1] In order to fully understand the failure mechanism of TBC systems and predict their life, one needs to study the evolution of cracks and the TGO interfacial surface geometry as a function of hour of operation. 3D Xray Microscopy (XRM) allows us to obtain such information non-destructively at various intervals of heat treatment corresponding to engines’ operation. However, lack of quantitative information does not allow us to develop or confirm constitutive relationships or failure mechanisms. Therefore, it is critical to assign materials to voxels in XRM images. This process is known as segmentation. Segmentation of TBCs, however, is not trivial. The top coat Zirconia-based composition significantly attenuates the Xray photons. Higher energy X-rays are usually used along with aggressive filtering to avoid beam hardening effects. However, even at high energies, the contrast between the top coat material and cracks and voids changes from slice to slice. Also, detection of rough aluminum-based interfacial layer, TGO, proves to be difficult and discerning that from cracks and voids close to the interface is very difficult. Previous efforts of segmenting TBCs systems have all been manual which is both time consuming and labor intensive. [1] This work aims to automate the detection of cracks in the topcoat and the TGO interfacial geometry of a heat-treated TBC sample using image processing and machine learning. To achieve this, TBC image data was first solicited from [1], where a cyclically heat-treated APS 7 wt.% Yttria-stabilized-zirconia TBC was imaged using 3D XCT [1]. In total, 1,007 2D X-ray micrographs were taken, serving as this work's TBC image dataset. In an attempt to identify cracks from this data, the 3D visualization software, Dragonfly was used [4].

How Our Software Was Used


Author Affiliation

(1) Department of Mechanical Engineering, Manhattan College, Bronx, NY, United States
(2) Department of Biomedical Engineering, University of Connecticut, Storrs, CT, United States