Research On Laser Ultrasonic Surface Crack Detection Method Based On CNN

Under complex working conditions,such as high temperature and humidity,impact,vibration,and heavy load,metal parts can easily produce various tiny cracks on the surface of the parts.If it is not discovered in time and allowed to expand,it will cause serious safety hazards and even serious safety accidents.As an emerging ultrasonic non-destructive testing technology,laser ultrasound has shown good application prospects in the field of nondestructive testing under its unique advantages.In the current mainstream detection methods,the identification of surface cracks mainly depends on using a variety of signal processing methods to extract relevant time-frequency domain information from the reflected and transmitted waves of the ultrasonic wave this process is very important to the relevant personnel.The professional skill level puts forward higher requirements;besides,the signal processing process mainly relies on manual operation,which makes the signal judgment result easily interfered with by human factors,leading to unstable detection results.To solve this problem,based on the deep learning theory,this paper takes surface crack depth detection as the starting point and proposes a laser ultrasonic crack detection method based on a convolutional neural network to realize automatic detection and classification of surface crack depth.The main work content is as follows:(1)The finite element method is used to simulate the propagation process of laser ultrasound in the substrate-coating structure,to explore the influence of surface coating and surface cracks on the propagation of surface acoustic waves;to analyze the surface acoustic wave signal in the time domain and time-frequency domain to determine surface cracks The position of the crack and the correlation between the depth of the crack and the frequency characteristics of the reflected wave signal,select the appropriate time-frequency domain analysis method,and prepare for the next data preprocessing before training the convolutional neural network model.(2)Aiming at the problem of manually extracting characteristic signals for discrimination in the process of signal processing,a surface crack depth classification method based on a convolutional neural network is proposed.This method uses the wavelet timefrequency diagram of the surface acoustic wave signal to train the convolutional neural network,and the convolutional neural network realizes the extraction and classification of crack depth features.The proposed convolutional neural network is further optimized by methods such as deepening the network structure and adjusting hyperparameters to improve the model's classification accuracy and speed up the training speed model.The test results show that the proposed convolutional neural network model can deal with cracks.In-depth and accurate classification.(3)Aiming at the problem that it is difficult to obtain a large number of labeled samples in an industrial environment,a surface crack depth detection model based on transfer learning theory is proposed.Use the open-source data set to pre-train the convolutional neural network model,and then use a small number of target domain samples to migrate the model to the target domain to solve the problem of too little data in the target domain.The fine-tuning method is used to fine-tune the classification layer parameters of the model to improve the generalization ability of the model.