Research On Ultrasonic Field Visualization And Classification Of CFRP Defect Types

The carbon fiber reinforced polymer(CFRP)has been widely used in the aerospace field.During its manufacturing and in-service process,it is susceptible to flaws due to factors such as process manufacturing and fatigue impact,thereby reducing the safety performance of CFRP structural parts.Ultrasonic non-destructive testing(NDT)is one of the common methods in the testing of CFRP materials.However,in the actual detection process,due to different materials,shapes,etc.,the corresponding detection schemes are different.The ultrasound field simulation can simulate the flaw detection schemes and guide the detection work.After ultrasonic testing,the interpretation of flaw detection results depends on professional knowledge and experience,and automatic classification of defects can assist the interpretation of results.The basic visualization of ultrasonic field and automatic classification methods of CFRP defect types were studied.Firstly,the relationship between the reflection and transmission coefficients and the incident angle was analyzed.The radiated and focused fields of the transducer were simulated respectively based on the multi-Gaussian beam(MGB)model.And the effects of probe frequency,size and interface shape on the radiated field were analyzed.Secondly,CFRP was detected by ultrasonic instrument,and the convolutional neural networks(CNN)model was constructed to automatically identify the defect types of CFRP materials.In the one-dimensional CNN model,multi-convolutional blocks were used to simultaneously extract as well as enhance the diversity of extracted features.Subsequently,one-dimensional residual blocks were stacked to simplify the training of networks.Bayesian optimization algorithm was used to optimize the network hyper-parameters(learning rate and momentum parameter of stochastic gradient descent).Considering the interference of noise to the ultrasonic signals during actual detection,the one-dimensional CNN model was improved.The idea of dilated convolution was introduced in the network,and the multi-scale network structure was constructed.Finally,based on the proposed defect classification model,the user interface of the defect classification model was designed,which can facilitate the updating of data set and training and testing of defect classification models.By simulating the ultrasonic field,the effects of different transducer parameters and interface shape on the ultrasonic field during NDT are analyzed.This shows that the visualization of the ultrasonic field can guide the design of the detection schemes.In terms of CFRP defect classification,the data set of ultrasonic A-Scan signals is constructed.And three flaw types of delamination,void and defect-free are automatically distinguished by the proposed CNN model.According to experiment results,the evaluation indicators of the proposed method are better than the comparative methods.The improved CNN model can also classify the ultrasonic A-scan signals with noise.Moreover,under different noise intensities,the classification accuracy of the improved CNN is higher than the network before the improvement.