Research On Infrared Thermal Wave Scanning Detection Of Defects In Large-size CFRP/Al Hneycomb Structure

CFRP/Al honeycomb structure is an advanced composite material with high strength,high rigidity,corrosion resistance,and lightweight characteristics.In recent years,it has gradually become one of the important choices to replace metal components,and is widely used in many fields such as spacecraft manufacturing,nuclear power,rail transit,and so on.The special laminated structure of CFRP/Al honeycomb structures is prone to defects such as interlayer debonding,water accumulation in the core,oil accumulation,and ice accumulation that are difficult to detect.Therefore,it is crucial to inspect and evaluate the defects of CFRP/Al honeycomb structures at various stages of manufacturing,use,and maintenance.However,due to the lack of appropriate collection tools and data processing methods,the current non-destructive testing of CFRP/Al honeycomb structural composites can only be limited to a small range,and cannot be conducted in a single large area.Firstly,based on the description of the heat distribution of a moving Gaussian heat source and the explanation of the principle of infrared thermal wave nondestructive testing,this paper completed the transient heat conduction analysis of infrared thermal wave scanning testing of large size CFRP/Al honeycomb structures;Using the finite element simulation method,a simulation study on infrared thermal wave scanning detection of defects in large size CFRP/Al honeycomb structures was conducted.The effects of simulation parameters on infrared thermal wave scanning detection results were analyzed by setting different scanning speeds,excitation powers,defect sizes,and types.Secondly,based on theoretical analysis and simulation results,an infrared thermal wave scanning detection method for detecting large size CFRP/Al honeycomb structures used in actual aviation is proposed;A special platform for installing excitation heat sources and infrared thermal imagers was designed,which was combined with industrial robot technology.A robot based infrared thermal wave scanning and detection system was built,and the system communication scheme was determined and arranged;The preparation of large size CFRP/Al honeycomb specimens containing debonding,water accumulation,oil accumulation,and ice accumulation defects was completed;Using this detection system,the efficient and reliable detection of defects in large size CFRP/Al honeycomb structures has been achieved.Then,a false static matrix reconstruction(FSMR)algorithm is used to pre process and convert the obtained dynamic infrared thermal wave scanning image sequence,and PPT,THD,LDA,and QDA algorithms are used to post process the converted standard static sequence;The processing effects of different algorithms on the converted matrix are discussed,and the best processing scheme for the image sequence obtained by dynamic infrared thermal wave scanning method is obtained through comparison.The experimental results show that the LDA algorithm has a good effect on improving the signal to noise ratio of debonding defects;The QDA algorithm can effectively improve the signal-to-noise ratio of water,oil,and ice accumulation defects,and increase the detectability of defects;It is shown that the infrared thermal wave scanning inspection method is competent for the nondestructive testing of defects in large size CFRP/Al honeycomb structures,and can be effectively and practically applied in industrial sites.Finally,the dynamic infrared thermal wave scanning detection method is combined with neural network technology to achieve automatic detection and classification of defects in CFRP/Al honeycomb structures.Through a large number of experiments,infrared image datasets of different types of defects in CFRP/Al honeycomb structures were collected and established,and the basic datasets were effectively expanded.By constructing Darknet19,Resnet18,and Squeezenet convolutional neural networks,the effects of different solvers and network structures on recognition speed and accuracy were explored.The experimental results show that the above three network structures can achieve a recognition rate of over 80%,but the SGDM solver exhibits the smoothest training curve and can achieve the best fitting effect;Squeezenet takes the shortest time to train in the SGDM solver environment,indicating that the lightweight network structure can effectively perform the automatic detection and classification of defects in CFRP/Al cellular structures.The final comparison results show that when considering the training duration,the optimal convolutional neural network model and optimizer combination is "SqueezeNet+Sgdm".After using GPU for acceleration,the training time is only 8.6 minutes,and the accuracy can reach 99.86%;When only considering optimal accuracy and ignoring machine performance,the optimal convolutional neural network model and optimizer are combined as "DarkNet19+Sgdm",with an accuracy of 99.97%.