Ultrasonic Characterization Model Of Material Microstructures Based On Hierarchical Bayesian Approach

Microstructure is a key factor,which governs material properties.Differences in material microstructure can affect mechanical properties such as fatigue strength,yield strength,creep properties,etc.Therefore,it is crucial to characterize material microstructure accurately.Ultrasonic nondestructive evaluations have been the focus of research because of nondestructive and high efficiency.In this thesis,considering about the difficulty of realizing theoretical models,the uncertainty of experimental conditions and the unstable sensitivity of traditional ultrasonic nondestructive methods,an ultrasonic characterization model of material microstructure based on hierarchical Bayesian approach is studied.Firstly,the detection method of ultrasonic longitudinal wave velocity and longitudinal attenuation coefficient is given to establish the traditional ultrasonic characterization model of material microstructure,and the linear regression model parameters are solved by the least squares method.Then the Bayesian updating are introduced,and the analysis process of the Bayesian method for the ultrasonic model of material microstructure characterization is described.Finally,numerical simulations are performed with a set of experimental data to compare the model parameter estimation results of the traditional method with the Bayesian method,and the effectivity of the model is evaluated by validation samples,laying the foundation of the framework for the subsequent chapters.Then the influence of the variability of the inspection environment,inspectors,and experimental setup of the ultrasonic benchmark data on the model is considered,and the observation is nested hierarchically to construct a hierarchical Bayesian model.Due to the increase of complex parameters in the hierarchical model structure,which cannot be solved by the grid approximation and Gaussian approximation aforementioned.Therefore,stochastic simulation is introduced to simulate the posterior distribution of model parameters using Monte Carlo sampling.Finally,numerical simulations are performed by RSTAN and model evaluation is achieved by using the information law.The above hierarchical model is further validated by two sets of experiments.The first one is the characterization of Al-matrix composites with different reinforcing fiber contents.The longitudinal wave velocity and longitudinal attenuation coefficient were obtained by setting up contact ultrasonic inspection with different experimental conditions,and the hierarchical model was constructed to evaluate the volume fraction of reinforcing fibers in the validation specimens.The other one was to characterize the nodularity of ductile iron.The microstructure characteristics of ductile iron were obtained through metallographic testing.Meanwhile,ultrasonic characterizations were obtained by setting longterm grouping experiments through water immersion ultrasonic inspection system,and ultrasonic longitudinal wave velocity and transverse wave velocity were calculated as explanatory variables to construct ultrasonic hierarchical model of nodularity characterization and evaluate the validation specimens.Finally,MATLAB is used for software design to realize the basic functions of data loading and updating,model definition,calling and evaluation,parameter estimation and verification,and variable prediction through the user interface,laying the foundation for the practical development of industrial software.