An improved method for the quantification of microstructures via optical microscopy

Abstract

Quantitative optical microscopy is a powerful tool for microstructural characterization of relatively large areas of multiphase materials, but accurate quantification often requires skillful sample preparation, imaging, and analysis techniques. Measuring phase fractions in a microstructure is done by segmenting images between features of interest and everything else and can be accomplished by a variety of techniques. With existing methods, in many cases, optimal contrast between the features of interest and the surrounding microstructural constituents is difficult to obtain and inconsistent across both individual images and a sample surface due to normal variations in materials and equipment, which can introduce measurement errors. A new machine-learning tool, the Labkit ImageJ plugin, provides a user-friendly method for training models for quantitative metallography and can be generally applied to study a wide variety of multiphase materials. Sørensen–Dice Indices were calculated to provide accuracy metric and show the improvement in image segmentation using Labkit to other common contrast-based thresholding methods versus manually-segmented representative images throughout the temperature range of this study. An example using intercritically heat treated HY-80 steel containing a duplex microstructure is presented using three different optical microscopy quantitative methods to show the improvement in reliability from the Labkit plugin. These quantitative measurements are augmented by differential scanning calorimetry and compared to CALculation of PHAse Diagram (CALPHAD) models to highlight the importance of experimental data for the design of heat treatments. A method for calibrating CALPHAD models to better predict industrial heat treatments is presented and shown to improve the accuracy of phase-fraction predictions, which is crucial for execution of intercritical heat treatments of steels.