Quality assurance by automatic processing of metallographic images

The periodical inspection of micrographs is an essential part of the quality monitoring process in the production of engine parts. Several new methods for the automatic extraction of quality relevant features are described in this paper to allow image processing based evaluations for materials where no automatic inspection was available up to now. Besides already existing solutions for the treatment of inclusions, a new approach for the detection of parallel lines (twins) will be presented. A special type of a line Hough transformation with the distance between parallel lines as the third dimension in the accumulator space increases both, the robustness and the accuracy compared to other algorithms and can reduce the computation time for images with a high number of parallel line segments. The computation of the distribution of  phase particles in images with weak grain boundaries which are often interrupted by gaps and recognizable only as accumulations of  phase precipitates in a typical shape, requires sophisticated methods. An adapted version of a grayscale image reconstruction can improve the contrast of boundaries while reducing non boundary structures inside of grains at the same time. Repetitive application of the grayscale reconstruction in combination with morphological operations and intermediate classification of contours produces an improved boundary image. Twins and scratches will be erased from the boundary image while protecting boundary - twin intersections and closing boundary gaps. This image is used for the computation of the phase percentage, the distribution of the distance of  phase particles to the next grain boundary, the evaluation of grain size and the characterization of anisotropic grain shape.