Research On CFRP Visual Detection System Based On Deep Learning

Carbon fiber reinforced polymer(CFRP)is widely used in various industrial fields,and its production and manufacturing technology has become one of the important pillars of China's strategic emerging industries.However,the research on fatigue damage detection of composite materials is still lack of mature theoretical system.Therefore,the research of CFRP detection method has been widely concerned.The purpose of this paper is to realize full-time,in-situ and multi-point defect detection based on electromagnetic detection method and artificial intelligence theory.Through the study of the electrical characteristics of CFRP,it is found that the distribution of its internal parameters is obviously different along different directions,which is the main obstacle to the detection of structural defects and fatigue damage of CFRP.It is an effective method to obtain the impedance characteristics in different directions by changing the relative angle between excitation and measurement.On this basis,an angle sensitive electromagnetic sensor array is designed.By exciting the coils at different positions in time and time,the defect detection of plain woven CFRP is achieved without moving the position of the coil.The requirement of multi-point and in-situ detection of the defect detection system is solved,and some noise interference caused by moving is avoided.Next,the simulation models of plain woven CFRP sheet and angle sensitive electromagnetic sensor array are constructed.By simulating different kinds of defects in the real sheet,a large number of defect data are collected by using array sensor,and the defect database is established.Then,the idea of deep learning is introduced into the detection system,and the auto-encoder network is trained by using the training set data in the defect database.In the network training,RBM is introduced to train the single-layer network layer by layer,and then the chain rule is used to connect them to form a deep network.In the whole network,the defect features of CFRP are extracted by the encoder,and the data are classified by Softmax classifier.Finally,the trained network is verified by the test set data in the defect database.The experimental results show that the auto-encoder not only has a good ability to identify defects by extracting the data features of CFRP defects,but also can be applied to defect classification with good results.