Research On Differentiate Tensor Defect Detection Technology And Experimental Verification Of Infrared Nondestructive Testing

The composites are prone to consist of defects,e.g.,debonding,porosity and delamination in production and use.Thus,nondestructive testing technology is required to evaluate the safety as well as preventing potential hazards.Since vision based nondestructive testing technology has been developing rapidly optical pulsed thermography has been widely used due to its excellent properties such as rapid detection,intuitive detection effect and non-contact.However,it suffers from the strong noise and reflection light spot interference in the optical infrared thermal images in which it is difficult to accurately extract defect features.The performance of the defect detection algorithm(e.g.,defect contrast,detection rate and efficiency)needed to be further improved.In the thesis,we propose a differentiate-based tensor decomposition algorithm,which extracts micro-defects under complex thermal interference.It is applied in the defect detection of surface or sub-surface.The main research work is as follows:1.Using the team build optical infrared thermal imaging experimental system,we collect the infrared thermal image data to construct the test cases.Based on these cases,the limitations of existing detection algorithms are analyzed and summarized.2.To further enhance the effect of defect detection,we propose differentiate-based tensor decomposition algorithm,which is built on a rank model.In this algorithm,the difference information between the structures of the Tucker decomposition approximation model with different ranks is deeply mined to extract the foreground component(defect information)of the image sequence and weak the interference information.Then the proposed probability tensor model is introduced to correct the potential mismatch pattern and accurately extract defect information.This algorithm can enhance defect contrast and suppress noise and spot interference.3.A large number of composite sample experiments were designed to verify the validity and generalization of the proposed method.On the one hand,starting from method design and parameter ablation,the rationality of the proposed method in design is analyzed and verified in fine granularity.On the other hand,compared with other advanced defect detection algorithms and tensor decomposition algorithms horizontally from multiple dimensions,the improved detection performance of the proposed algorithm is analyzed and verified.The proposed algorithm achieves higher performance in F-Score than other algorithms in quantitative analysis.It shows that the algorithm obtains better performance at detecting weak defects.In addition,compared with other algorithms,the SNR score of the proposed algorithm gains a high improvement,which shows that our algorithm has an advantage in enhancing defect contrast.In the qualitative visualization experiments,we can find that the proposed algorithm has a stronger ability to remove interference information.