Research On Image Segment In Defect Testing Of Turbine Blade

The NDT (Non Destructive Testing) of engine turbine blades plays an important role in aviation. The traditional methods are difficult for inspection of a hollow turbine blade due to the complexity of its inner structure. ICT (Industrial Computed Tomography) as a new technique is developing in 1980's. Especially the development of technology of 3D-CT accelerates the application of ICT which used in NDT. This paper focuses on the technique of hollow turbine blades testing in reverse engineering. Through analyzing the existing methods, we proposed a testing method based on the 3D-CT. The experiment showed good results in resolving of testing of turbine blade.First, the paper summarizes some testing methods, including algorithms based on threshold, edge detection, clustering, DSA and statistic and presents the applicable range of these methods.After investigating blade testing systems in factory, we study the defects of turbine blades thoroughly including void, crack and etc. In order to identify correctly, we summarize definition of typical defect, which provide evidence in detecting and classifying different type of defect.By generating ICT image sequence from turbine blade CAD model in which exist defects, we provide raw data for defects extraction.In order to solve the testing problem of turbine blade, this paper proposed an image segmentation method which use morphological parameters to segment and extract the defects from the image. Different from existing methods, we use some parameters to direct gray and morphology of defect. Depended on these parameters, we implement the sort of typical defects automatically.Image segmentation plays an important role in defect testing. In order to acquire good detecting results, this paper also put some work on image segmentation and chooses the method of otsu improving. Then we do some work on evaluation of segmentation result and validating result correctly. Finally, we test this detecting method through simulation image and get the correct result.