Study On Type Recognition And Reconstruction For Ultrasonic Detection Of Internal Defects In Revolution Body

The workpieces of revolution body have been widely utilized in many areas such as aerospace,chemistry,etc.However,these workpieces are easily broken,due to internal defects during the designing and manufacturing process,which directly affects the stability and safety of the entire systems.Hence,nondestructive detection of defects is an essential procedure when designing and manufacturing workpieces of revolution body.At present,ultrasonic technology is one of the most common approaches to detect internal defects in revolution body.However,this approach required technicians to observe ultrasonic waves manually,leading to low efficiency.Moreover,the detection accuracy relies on individual's experience,leading to large detection variability.To overcome those limitations,we investigate,in this thesis,the mechanical scanning manners,ultrasonic field computation as well as the feature of echo signals.Based on these outcomes,we proposed approaches of signal denoising,feature extraction,defect recognition and reconstruction technologies.Regarding the concern of noise effects on ultrasonic echo signals,we utilized minimum square adaptive filter to denoise signals,which remarkable elevates the signal-to-noise ratio(SNR).This is a key procedure for detection of internal defects precisely and efficiently.Additionally,a model of multi Gaussian ultrasonic sound beam field was used to analyze the ultrasonic field transmission inside evolution body,and finite element analysis(FEA)model was established to simulate the feature of echo signals when defects appear.The simulation results were then compared with the experimental signals,verifying the accuracy of the proposed analytical and simulation approaches.Those outcomes provide the data supports for the subsequent procedures of internal defect reconstruction.Three typical shapes of defects,including circle,ellipse and rectangle,were extensively investigated in this thesis.Based on these,a novel characterization method for multi-dimensional ultrasonic echo signals and an automatic recognition model were proposed.Specifically,an ensemble empirical mode decomposition(EEMD)algorithm was adopted to decompose the original ultrasonic echo signals into several layers.Moreover,an energy distribution and information entropy distribution were selected as the input features in defect recognition model,wherein the number of decomposition layers was also determined by the physical features.Secondly,an approach of decision tree algorithm was used to establish the defect recognition model.The experimental results showed that the accuracy can reach 88%.In order to improve the efficiency and robustness of defect recognition model,a model of random forests,which consists of many weak classifiers,was established.As the consequences,the recognition accuracy has reached as high as 92%.The proposed random forests model was found to achieve the minimum standard deviation(SD),when compared with conventional methods,such as decision tree,support vector machine(SVM)and backpropagation neural network.The results indicated that the inherent ensemble strategy of random forests algorithm can effectively overcome random influences of single weaker classifier,hence ensure the stability and robustness.For the situation that workpieces have more than two internal defects and are not located at the center of workpieces,we utilized an approach of self-organizing map(SOM)neural network to automatically recognize the defects within near,medium and far ultrasonic field.A defect recognition model was established based on the near ultrasonic field signals.The experimental results showed that the proposed model not only reduced the size of training dataset(reducing by about 4/5),but also remarkably improved the recognition accuracy(90.14% vs.85.32%).We also analyzed,in this thesis,the limitation of time-path based conventional transmission method for defect reconstruction through combining the theories of ultrasonic wave transmission and technologies of B scanning echo signal for cross-section detection.Furthermore,we discovered the trends of the tangent pattern in defect bound and ultrasonic wave front arc.Additionally,a novel transmission time-arc tangent fitting-based defect reconstruction method was proposed and implemented to reconstruct circle,ellipse and rectangle shapes of defects.For the ellipse type of defect that is hard to obtain the closed equations of the tangent between defect bound and ultrasonic wave front arc,a local region tangent between circles method was proposed to replace the original tangent between ellipses.Subsequently,reconstruction was implemented through fitting the bound of ellipse type defect.In order to precisely obtain the transmission time of ultrasonic echo signals within workpieces,a convolution-solving method was adopted to extract the time point of lower surface and defect featured echo signals.The proposed reconstruction method was applied to a real defect workpiece at different cross-sections,and volume rendering approach was adopted to display the defect in three-dimensional visual stations.Experimental results showed that the location and size of the defects reconstructed by proposed reconstruction method was in accordance with the X-ray scanning results.In summary,we extensively investigated,in this thesis,the ultrasonic detection of defects in revolution body through theory analysis,computer simulation as well as signal acquisition experiment.The study indicates significance of the proposed approaches in both theory development and engineering applications.In addition,the proposed approaches are also applicable for the ultrasonic detection of other defects with similar defects.