A new approach for the automatic evaluation of the solidification structure in steel using orientational entropy filtering



    Corresponding author
    1. Faculty of Engineering and Information Sciences, University of Wollongong, Wollongong, Australia
    • Correspondence to: Stefan Griesser, Faculty of Engineering and Information Sciences, University of Wollongong, Northfields Avenue, Wollongong, NSW 2522, Australia. Tel: +61-2-4221-5718; fax: +61-2-4221-3662; e-mail: sg045@uowmail.edu.au

    Search for more papers by this author
  • P. O'LEARY

    1. Institute for Automation, University of Leoben, Leoben, Austria
    Search for more papers by this author


We present a new method for the automatic evaluation of the dendritic solidification structure in metallurgical micrographs of solidified steel. The evaluation of the microstructural parameters such as the primary dendrite arm spacing and the primary grain size are of high importance due to their direct relationship with the internal quality and mechanical properties of the cast product. Given the repeated geometric features in the micrographs and the regular pattern in colour intensity, we applied a filter mask to determine the local entropies within the masks in order to detect the centre coordinates of each individual dendrite. The orientation of the dendrites was determined by rotating the filter mask over each pixel to find the orientation which corresponds to the lowest entropy value. The segmentation of the microstructure was then performed via Delaunay tessellation and subsequent transformation of the triangular mesh into a rectangular grid, enabling the determination of the desired microstructural parameters.

Lay Description

The application of image processing software to automatically evaluate microstructural features in metallurgical polished micrographs has become a key element within the research and development of new materials as well as quality management of existing processes. Because of the relationship between the material properties and their internal microstructure, the knowledge of the microstructural parameters is often used to predict the materials properties or to draw conclusions on the conditions during the materials production process.

During the solidification of alloys, tree-like crystals (called dendrites) form in the liquid melt and grow until the cast product has completely solidified. The structure that is formed by these dendrites is commonly referred to as the primary or the solidification microstructure. In many alloys, further cooling results in additional phase transformations that lead to the formation of different microstructural features, which is referred to as secondary or tertiary microstructure.

Existing machine-vision based systems have concentrated on the evaluation of secondary and tertiary microstructures, such as grain boundary detection and grain size measurement. In contrast to primary grain boundaries, these secondary grain boundaries are directly visible in the micrographs and can be easily detected via contrast enhancements. Primary grain boundaries however are defined as regions where groups of dendrites with different orientations collide, so that they can only be detected indirectly by finding local disorientations of neighbouring dendrites.

We propose a novel approach for the determination of the position and orientation of the individual dendrites, which is then used to identify the individual primary grains (i.e. groups of dendrites with the same orientation) and the resulting primary grain boundaries. A beneficial side effect of using this technique is that the primary dendrite arm spacing, which is another important microstructural parameter, can also be automatically determined.

The evaluation of the primary solidification structure is a novel field of research in materials engineering due to the experimental complexity involved, which contributes to the lack of existing data in the relevant literature. The application of purposely developed image processing software such as the one presented in this work, will help to quickly generate data-sets of primary microstructural parameters, which can then be used for the study of microstructure evolution or to minimize quality defects in the materials production process.