Research On Defects Detection And Classification Based On X-ray Image For Aviation Titanium Alloy Castings

Defects identification of aviation titanium alloy castings using X-ray inspection imaging is a common method to evaluate its internal quality.To ensure the safety and reliability and use performance of aviation products,it is usually necessary to quantitatively output information such as defect location,quantity,type and size.However,the existing manual recognition methods have problems such as unstable recognition results,low quality inspection efficiency,and low utilization of X-ray system.For these reasons,this dissertation studies the computer-aided identification method of detecting and classifying defects for aviation titanium alloy castings based on X-ray images,develops an intelligent recognition system for X-ray images of casting defects,and applies it to aviation titanium alloy precision casting enterprises.Firstly,a defect target detection method based on selective search and defect image feature analysis is proposed.According to the characteristics of X-ray imaging,quality enhancement preprocessing is performed on the image from four aspects: noise,brightness,contrast,and sharpness.Based on the selective search algorithm,the suspicious regions of the image are recommended,and the maximum size and edge curvature characteristics of the real defects are statistically analyzed.Then establish a standard for removing false defects,and achieve a 99.7% recall rate of true defects and a 91.76% precision of detection.Secondly,a convolutional neural network model is constructed to classify the target regions of defects.The model data set is created on the acquired raw X-ray image data,including 5 categories of low-density holes,shrinkage,linear defects,high-density inclusions,and casting body structure.A 7-layer convolutional neural network architecture is built,the relevant parameters of the model are set up and trained,and we get preliminary classification prediction results.Furthermore,the feature learning is optimized from three aspects of data expansion,optimization of the learning rate and improvement of model overfitting.By comparing with several typical convolutional neural network models including network before optimization,Alex Net,VGG16,and VGG19,the classification availability of the current model is verified.Based on the 50 original-size images after the targets location,the accuracy of the overall target classification accuracy is 87.65%.Finally,an intelligent recognition system for X-ray images of casting defects is designed and developed,which is applied to an aviation titanium alloy precision casting enterprise.The system functions include image reading and display,defect detection and classification,result output and storage.After a comparative analysis with the actual recognition process of the enterprise,the example results show that compared with the manual identification method,the intelligent recognition system for X-ray images of casting defects can obtain higher stability results,shorten the identification time from about 15 minutes to about 1 minute,and improve the utilization rate of the X-ray system.