Study On Non-destructive Testing Of Shallow Defects In Steel Bridges Based On Eddy Current Thermography Principle

Steel bridges that have been in service for a long time are susceptible to corrosion and rust due to environmental factors,which can affect the load-bearing capacity and durability of the bridge.Fiber-reinforced polymer(FRP)composites have been widely used as an excellent protective measure.However,due to construction processes,the occurrence of bubbles and debonding in the adhesive layer cannot be avoided,which affects the safety performance of the FRP-steel structure.How to detect corrosion and interface debonding defects without damaging the steel bridge has become an urgent issue in the field of bridge engineering today.In this paper,eddy current thermography(ECT)technology was adopted,and a self-developed ECT detection system was used to conduct a series of experimental studies,including paint peeling and corrosion of steel bridges and interface debonding of FRP-reinforced steel structures.By establishing a multi-physics field coupling model,the process of ECT detection of interface defects in FRP-steel composite structures was theoretically analyzed and simulated.At the same time,based on deep learning algorithms,four sets of steel bridge defect datasets were established,and a deep learning platform based on these algorithms was constructed.Finally,relying on the ECT system and the intelligent detection platform,experimental studies on actual steel bridge detection were simulated,and a series of research achievements were obtained.(1)Eddy current thermography(ECT)non-destructive testing technology performs well in detecting paint peeling on steel bridges and can accurately reflect the true form of defects,especially during the cooling process,where detection results are more pronounced.The size of defects,the thickness of the steel plate,and their relative position to the heating source all affect detection performance,with larger defect areas being detected earlier and more clearly.Due to the edge effect,defects at the edges are heated first and thus detected earlier.(2)Compared to paint peeling,the oxide layer formed by steel bridge corrosion hinders the transfer of heat to the air,resulting in a high-temperature state during ECT detection.In addition to the shape and position of the defect,the thickness of the heated steel bridge has a greater impact on detection performance,and the heating time and power need to be adjusted according to different plate thicknesses.Furthermore,different levels of corrosion lead to different thermal response modes in the steel bridge,and the degree of corrosion can be indirectly evaluated by the |?T|,allowing for timely repairs.(3)This ECT system can detect interface defects and scratches on FRP-steel composite structures,but due to the "fuzzy effect",the shape of defects is difficult to determine.The temperature distribution curve undergoes a sudden change at the defect position,forming a "low-temperature valley",which can be used to identify the defect location.The heating scheme should be selected according to the structure type,and different interface layer defects can be distinguished by ΔT.Defects with a diameter smaller than 10 mm are difficult to detect,and longer heating times should be used to improve detection performance when measuring thick steel plates.(4)The multi-physics coupled model can accurately simulate the change pattern on the surface of test pieces and detect defects in approximately 50 seconds.The heating power has a certain influence on the defect detection effect,and a small current will reduce the heating ability,but it does not have a significant impact on the optimal detection time.The thickness of the epoxy resin layer and the heat transfer coefficient "h" will affect the defect detection effect.Increasing thickness will aggravate the blurring effect and delay the optimal detection time.The larger the heat transfer coefficient,the shorter the defect detectability duration.(5)Based on the YOLOv5 algorithm,after 100 rounds of training,the identification accuracy of steel bridge defects was significantly improved,with an average reliability of over 85%.Different datasets are suitable for different network structures.Adopting the YOLO v5 s model can improve the identification efficiency while ensuring the identification effect.The optimized dataset can improve the average precision m AP@0.5:0.95 and shorten the training time.By modifying the configuration file parameters,automatic counting of defects can be achieved.A visual intelligent detection platform has been established to improve the efficiency of bridge disease detection.(6)This ECT system can accurately detect defects in the simulated steel bridge,and the deep learning algorithm can accurately identify defects contained in both infrared and ordinary images.However,different heating methods will seriously affect the defect detection effect.The paint detachment and corrosion defects of the steel bridge are suitable for surface heating,and the interface defects of FRP-steel structures are suitable for flange heating.Smaller interface defects are difficult to detect,but naturally occurring empty defects can still be accurately detected.Overall,the system has high reliability and accuracy.