Adaptive segmentation method in radiographic testing for turbine blades based on spatial entropy

Abstract

During the radiographic testing with multiple voltage exposures, the problem of image redundancy has serious influences on the speed of analysis. In this paper, by the calculation of spatial entropy and the extraction of microtopography features, a new segmentation method for an unpredictable free-form surface in turbine blades is presented to reduce the testing redundancy. First, the entropy calculation is applied to a set of radiography images of the target object. The optimized image for surface segmentation is determined according to global entropy values and exposure parameters. Then, the obtained image is set as the reference image for the segmentation process. Through the spatial entropy calculation, the reference radiographic image is divided into several regions by geometric features of the tested object. The different thickness range of free-form surfaces is indicated by those regions through the analysis of spatial entropy distribution in the reference image. By the statistical advantage of entropy calculation, the selected region is self-adaptive to the unpredictable free-form surface in the blade. A nickel-based alloy turbine blade is used to validate the segmentation method in the radiographic testing. The processed image quality is assessed by using the American Society for Testing and Materials image quality indicator to address its capability for the detection of defects, where the resolution of the image is not affected by the down-sampled effect of entropy calculation. The experimental results show that the image redundancy in the multiple exposure testing is reduced to less than 30% during the inline testing, while the dynamic range in each extracted region with an optimal image is significantly improved.