Design And Implementation Of Quantitative Evaluation Algorithms For Pulse Infrared Nondestructive Testing

Pulsed infrared nondestructive testing technology,applying a pulsed heat flow to a material under test,makes the discontinuity of heat conduction in the internal defects of the material be expressed by the difference in surface temperature of the material under test.Surface temperature data at different times in the cooling process of the material are collected by an infrared thermal imager.After analyzing and processing of these data via quantitative assessment algorithms,internal defects could be detected.In the way of defect detection methods,it contains three close information processing processes,the first is pretreatment based on gray transformation and sequence features,the second defect segmentation of processed infrared heat wave image data,and the last is defect recognition of segmented data.In the way of experimental materials,there is no information pretreatment method that could be applied to all aerospace composites,because of differences in physical characteristics of aerospace composites.Besides,new research is required,contributed by the emergence of new aeronautical composites and its high performance.In the way of the development of detection algorithms,it is the basis for the landing of defect detection algorithms and the power for detection algorithm that developing professional processing software for pulsed infrared nondestructive testing which covers application requirements.Based on the above analysis,the research of the thesis is based on the improvement of defect detection algorithms,with them being deployed in software platform as a supplement.In terms of defect segmentation,the region partition method of two-dimensional Otsu segmentation algorithms is improved based on an improved two-dimensional histogram,through which the theoretical error of two-dimensional Otsu segmentation algorithms is suppressed.Combined with the relative threshold idea of local threshold segmentation methods,an Otsu defect segmentation algorithm with stronger robustness is proposed.Appling to infrared thermal wave image data of new aerospace composites,this algorithm is experimentally verified and compared with the classical two-dimensional Otsu segmentation algorithm.Experiments show that for one side,this algorithm,introducing neighborhood mean and neighborhood total gradient as important parameters for characterizing the category and space state of pixel points and adopting a statistical adjustment model based on hybrid dimension to dynamically adjust gray values of defect zones and defect-free zones of image data,could solve the problem of uneven distribution of residual background temperature values and improve the robustness of Otsu algorithm.For the other side,this algorithm,adopting the improved two-dimensional histogram region partition method based on gray-neighbor deviation and automatically selecting neighborhood side lengths for finding an optimal threshold,could avoid the inappropriate selection of neighborhood lengths and improve defect segmentation accuracy.In terms of defect identification,based on the defect segmentation,the machine learning method is introduced for researching on intelligent extraction of defect parameters.Firstly,analyzing influences of different initial category coefficients on the defect recognition effect of K-means clustering algorithm,determining a reasonable preset coefficient interval and traversing the presupposed class coefficients,cloud offer a foundation for stable recognition results.Secondly,designing optimization method for rationally compressed amount of data,according to the characteristics of cluster core distribution,could reduce the influence of redundant cluster cores on the operation efficiency.Thirdly,considering about defect segmentation data features,calculate position coordinates based on the idea of connected graphs,so that it could ensure a high recognition accuracy rate.Lastly,determining the input data dimension of BP neural networks by the analysis of heat conduction theory,to establish a defect depth identification networks with relatively small error,could effectively identify defects that are affected by temperature radiation unevenness,deep depth,and small size.In terms of software platform,special processing software for pulsed infrared nondestructive testing is developed.In the process of integrating preprocessing algorithms,to determine algorithms integrated in this software system,a plurality of sequence enhancement algorithms are used for experimental analysis,for the infrared thermal wave image data of the new aerospace composite materials.In the design process of this software,cover users' application requirements for processing and analyzing pulsed infrared nondestructive testing data as much as possible,and take into account the lightweight and scalability as well.In the process of developing and deploying the processing modules,adopting union programming method of C# and MATLAB could provide powerful support for the subsequent scientific research work.The contributions of the thesis are summarized below: Firstly,the two-dimensional vector of pixel point gray value and neighboring gray deviation value is uesd to optimize the two-dimensional Otsu division method,which suppresses the threshold calculation error of classical two-dimensional Otsu algorithm in the infrared nondestructive detection process.Secondly,Otsu defect segmentation algorithm with stronger robustness is proposed,which could solve the problem of uneven distribution of residual background temperature value,combined with the adjusted threshold search method,finally improving the defect segmentation accuracy of pulsed infrared nondestructive testing.