Automated phase segmentation with quantifiable sensitivities of three-phase microstructures of solid oxide cell electrodes

Microstructures control the properties of materials and their performance in applications, including solid oxide cell (SOC) electrodes. Direct calculation of microstructural parameters from 2D or 3D images requires phase segmentation of experimental greyscale data. Subtle differences in segmentation outcomes can lead to significant differences in calculated microstructure parameters, especially for higher-order ones such as triple phase boundary (TPB) density. Unfortunately, it is nearly impossible to reproduce published segmentation results and their uncertainties are difficult to assess. We present a modification to typical watershed segmentation that addresses these issues. Markers are first placed based on the gradient distribution, using a single threshold parameter, and then grown by watershed to generate a marker segmented image. Phase-labeling is readily carried out from a post-watershed marker averaged greyscale distribution, using two threshold parameters. Sequentially performing these steps and simplifying the methods to determine the optimal values of the three threshold parameters enables the process to be automated and readily reproduced. With this new method, common segmentation errors can be avoided and sensitivity of computed microstructural properties around the optimal threshold values can be assessed. The method is applied to several large-scale 3D SOC electrode datasets, and the outputs are compared to published values. • Segmentation of a three-phase microstructure from 3D greyscale data was automated. • Phase and region identification are separated and use different data and methods. • Three threshold parameters are identified and optimized to remove user involvement. • Improper phase-IDs and resultant erroneous triple phase boundaries are reduced. • Quantitative sensitivity analysis of the output to the threshold parameters is done.