Quantitative Evaluation Of Subsurface Defect Using Laser Ultrasonic Technique And GA-BP Neural Network

With the rapid development of industrial technology,metal materials are widely implemented in all walks in life.During their manufacturing and service,it is easy to develop tiny defects inside the material as a result of the impact of harsh environments such as high temperature,high pressure,etc.If tiny defects cannot be identified in time and be prevented by implementing effective methods,hidden damage will gradually develop into a series of large defects unnoticedly,which may ultimately cause fracture or failure of materials,and leading to serious safety accidents.Therefore,an effective nondestructive evaluation method that enables the detection and quantification defects inside the metal materials is crucial and highly demanded.The laser ultrasonic technique is considered as a promising and emerging approach in the field of nondestructive testing as a result of its advantages in non-contact,high precision,and ability to work in harsh environments such as high temperature,high pressure,high radiation,etc.In this thesis,the interaction of laser-generated ultrasonic waves with subsurface defects in metallic alloy materials is studied based on finite element simulation and experimental research,and a systemic approach that combines signals process methods and machine learning algorithm is developed to quantitative evaluation subsurface defects in materials.The specific research contents and innovations of the thesis are as follows:(1)The theoretical principle and the finite element method of laser-generated ultrasound are mainly introduced.The numerical model based on the thermoelastic mechanism is established to simulate the process of laser-generated ultrasonic waves in 2024 aluminum alloy plates and to solve the interaction of laser-ultrasonic with subsurface defects of various dimensions.Simulation results indicate that ultrasonic signals contain subsurface defect information,which reflected that the variance in signal amplitude and phase are significantly related to the width of the subsurface defects.(2)A non-contact laser ultrasonic defect detection experiment system is designed,and its setup and working principle are analyzed in detail.The 2024 aluminum alloy plates with various crack widths are chosen as the experimental specimen,which can be designed and fabricated for investigation the correlation between laser ultrasonic signals and defect sizes.The experimental results of the detected ultrasonic signals in time domain validate the numerical model based on the finite element method and verify that the variance in signal amplitude and phase are significantly related to the subsurface defects,which experimentally indicate that laser ultrasonic signals contain the crack width information(3)A systemic approach that combines signals process methods and machine learning algorithm is developed to quantitative evaluation subsurface defects.The GA-BP neural network model is designed to quantify subsurface defects by combining back propagation(BP)neural network and genetic algorithm(GA).Signals process methods are implemented to analyze detected ultrasonic signals,and extracted sensitive features in time domain and frequency domain to establish dataset for GA-BP neural network training and testing.The relative errors of the prediction results in the GA-BP quantitative evaluation model that are all below 6% and the average error is 2.15%.Such low prediction results suggest the combination of the GA-BP neural network model is a feasible and reliable tool for evaluating the width of subsurface defects using laser ultrasonic technique.