Application Of Neural Network Based On Sequence Signals In Nondestructive Testing Of Composite Materials

Composite materials have been widely used in various fields due to their excellent properties and ultra-high-cost performance.At present,the requirements for nondestructive testing efficiency and reliability of composite materials are becoming higher and higher.However,traditional feature extraction and data analysis methods have great limitations and cannot meet the needs of modern nondestructive testing.The main purpose of this paper is to realize the automatic defect detection of composite materials by using the neural network methods based on sequence signals,and to realize high efficiency and reliable automatic defect analysis,and further promote the development of artificial intelligence technology in the field of automated non-destructive testing.In this paper,the serial signals collected by air-coupled ultrasonic technique and infrared thermography technique are taken as research objects.The neural network algorithms based on sequence signal are used to realize the automatic feature extraction of sequence data,so as to realize the task of automatic defect recognition.The main work of this paper is as follows:Firstly,the basic principles of air-coupled ultrasonic technique and infrared thermography are introduced,and the theoretical basis of defect detection based on the serial data collected by these two techniques is analyzed.Then,three kinds of neural network algorithms which can be used to process sequence data are introduced to provide theoretical support for the following research.Subsequently,long short-term memory recurrent neural network model and one-dimensional convolutional neural network model based on air-coupled were designed,which can automatically detect the impact damage of fiber composites.Four kinds of fiber composites with different structures with impact damage were used to train and test the model.The results show that the trained LSTM-RNN model has strong generalization ability and high precision and recall rate on all three test specimens.In the study of pulsed infrared thermography for the automatic detection of non-planar carbon-fiber-reinforced polymer samples,the long short-term memory recurrent neural network model and artificial feedforward neural network are designed respectively.The original temperature-time series,the coefficients after thermographic signal reconstruction,and the sequenced signals processed by the thermographic signal reconstruction and first derivative are used to train and test the model.The quantitative analysis of the test results show that thermographic signal reconstruction and first derivative processing can effectively eliminate the influence of heating unevenness and shape on the thermal image signal.The long short-term memory recurrent neural network model is more accurate than the artificial feedforward neural network model in dealing with time dependent information,at the same time,it can also realize the analysis of the internal defect depth of the specimen.40 Figures,4 Tables,and 85 References.