Research Of Detection Extraction Of Vane Defects Based On ICT Images

Hollow turbine blade is the key parts of aviation engine. It is very important to the security and reliability of flight that strictly quality inspection is made for it. Turbine blades quality inspection technology based on Cone Beam Volume CT has become hot point of current NDT research field. Image segmentation is either the key technology of defect recognition or the foundation of auto-identication. This paper does the research around this technique. The main research contents are as follows:1. Research of image segmentation based on Markov random field model: A cluster method based on statistical mode is introduced in this paper. This algorithm first uses an outlier rejection method to overcome the bad influence of noise and artifacts on follow disposal. Then it applied Markov Random field to clustering analysis method based on Gauss function. At last, using ICM (Iterative Conditional Mode) algorithm, the segmentation result is obtained by the MAP (maximum a posteriori) criterion. In order to improve the speed, a decisive algorithm based on game theory is introduced. The experimental results show this method can get higher precision than traditional image segmentations based on Markov random field model.2. Research of image segmentation based on fuzzy clustering: Fuzzy clustering is the best method to solve uncertain decision of the training samples. Aiming at the problem that image histogram contains much flat areas, this paper applies fuzzy theory to clustering method based on objective function. It proposed image segmentation algorithms based on Distance function and Gaussian function. Gaussian mixture model algorithm uses three-state Gaussian mixture model to match the distribution of pixels. Firstly this method is to find the parameters which are exactly match the corresponding distributing of each kind. Then segmentation is judged by mixture objective function. The experimental results show this method can obtain higher accuracy and faster implements. It can strictly locate defect areas.3. Research of image segmentation based on fuzzy ART algorithms: For the number of slice sequence image's data is quite large, the neural network method which fit to parallel compute is adopted. It combines fuzzy technology with ART neural network to achieve the segmentation of CT image. Based on detailed analyze of FART's net model, algorithm and parameter characters, a new method is proposed. It uses fuzzy character which composed of gray statistical information as the import of fuzzy ART neural network. Image segmentation is realized by classify grays. This method can automatically determine the number of classes and effectively suppress the noise in the given image. The experimental results demonstrate velocity and effectiveness of this method.