Design Of Electromagnetic Ultrasonic Nondestructive Testing System Based On Deep Belief Network

In the paper,electromagnetic ultrasonic nondestructive testing system is designed with the superiority of the local mean decomposition(LMD)processing nonlinear and non-stationary signal,the strong feature extraction ability and learning ability of deep belief network(DBN)and the superiority of the LabVIEW programming software to meet the need of intelligent nondestructive testing technology.The main research contents are as follows:1)A novel method of electromagnetic ultrasonic defect recognition based on LMD time-frequency feature and DBN is proposed for the problem with low recognition rate because of the nonlinear and non-stationary characteristics of electromagnetic ultrasonic signals.The experimental results show that the proposed method can effectively improve the defect recognition rate through two aspects: the input features and the classifier.2)According to the complexity,incompletion and uncertainty for human paticipating of signal feature extraction will increase the difficulty of identifying defects and reduce the intelligence of machine learning,the defect recognition method of original data based on deep belief networks is used to electromagnetic ultrasonic defect recognition.By reconstructing the feature of hidden layer signals,it is shown that DBN have strong ability for extracting feature and learning.Experiments show that this method can effectively avoid the process of artificial feature extraction and optimization,and effectively identify defects,and the recognition rate is higher than that based on SVM and BP neural networks.3)The electromagnetic ultrasonic nondestructive testing system is built,and the real-time data acquisition,online training and online defect recognition experiments are carried out.The experimental results verify the reliability and the effectiveness of the system.