Study On Damage Detection Of Steel Truss Bolted Connection Based On Computer Auditory And Visual Technology

Bolt connection is widely used in steel truss structure.When the steel truss structure is subjected to vibration,impact and other external effects,the bolt connection state of each node will often change,such as loosening,falling off or fatigue fracture,which will damage the integrity of the structure and endanger the reliability and safety of the structure.The traditional method of bolt looseness detection based on contact measurement usually requires the sensor to be fixed near the position of the bolt to be tested,which is not easy to implement for a steel truss structure with a very large number of nodes and bolts.At present,although the commonly used methods of artificial hearing and visual observation are effective,they are time-consuming,laborious,inefficient and have great security risks.In recent years,with the development of artificial intelligence and UAV technology,it provides a new research idea to solve the problem of on-line identification of steel truss bolt looseness by teaching the computer to "listen" and "see" to identify structural damage.In this paper,aiming at the problems and shortcomings of traditional detection methods of bolt looseness and missing,the damage identification of bolt connection in complex environment is studied by using array signal processing,information fusion,machine learning,deep learning and other theories,based on advanced technologies such as computer hearing,computer vision and UAV,Finally,the damage detection scheme of bolted connection of steel truss structure is constructed to realize on-line damage detection of bolted connection of steel truss structure under complex working conditions.The main work and results are as follows:(1)Based on the sound signal,the early warning method of Steel Truss Bolt Looseness damage is studied,and the sound signal early warning method of Steel Truss Bolt Looseness damage based on support vector machine and genetic algorithm is proposed.Firstly,the time-domain features and wavelet packet frequency band energy features of the sound signals under the pre-set damage and non damage conditions are extracted.Then,the support vector machine recursive feature elimination method is used to select the sensitive features,which are input into the support vector machine model for training.At the same time,the genetic algorithm is used to optimize the parameters,and finally the optimal recognition model is obtained.Using this model,it can quickly and accurately distinguish whether the detection signal is the bolt loose damage signal.The experimental study on bolt looseness of steel truss structure shows that this method has strong practicability and can achieve better recognition effect with less training data sets.(2)The deep learning image classification technology in computer vision is used to detect the sound events of bolt looseness damage,and a bolt looseness recognition method based on wavelet time-frequency image and lightweight convolutional neural network Mobile Netv2 is proposed.Through preprocessing and continuous wavelet transform of the sound signals collected from 16 preset working conditions,the wavelet timefrequency diagram is obtained.Taking the wavelet time-frequency diagram as a sample,the lightweight convolutional neural network Mobile Netv2 is trained to obtain a robust recognition model,and then the model can be used to realize the accurate recognition of the sound signal of bolt loosening damage.The experimental results of steel truss show that this method can give full play to the advantages of wavelet time-frequency analysis in processing non-stationary signals and the powerful image classification ability of Mobile Netv2,and can realize the accurate classification and recognition of bolt loosening sound signals with different environmental noise,different damage location and damage degree.(3)The sound signal-based bolt loosening damage location research of steel truss structure is carried out,and the ODB-SRP-PHAT fast sound source location method and the multi-frame target signal analysis result fusion decision method based on evidence theory are proposed.The idea of ODB-SRP-PHAT method is that before on-line localization,the Peak density clustering method is used to accurately find out all possible sound source locations to build an off-line database ODB of sound source location information.In the search stage,only the locations in ODB are searched,thus greatly reducing the amount of search calculation.Simulation test and field measurement show that this method can effectively overcome the shortcomings of traditional methods,which are too much calculation and not conducive to online positioning.The decision-making method of multi frame target signal analysis results fusion based on evidence theory refers to the combination of ODB-SRP-PHAT method and evidence theory to fuse the results of different frame signals analyzed by fast source localization method.The experimental study on bolt looseness of steel truss structure shows that this method can not only significantly reduce the amount of calculation and save the calculation cost to the maximum,but also has strong anti-noise and robustness,and can accurately identify the damage location even in the case of low signal-to-noise ratio.(4)The computer vision target detection technology is applied to bolt missing detection,and an intelligent bolt missing detection method based on UAV and improved YOLOv4 target detection technology is proposed.The original YOLOv4 target detection method is improved by taking measures such as lightweight network structure,data enhancement and prior frame adjustment,which greatly reduces the calculation parameters of the original network structure,speeds up the calculation speed,reduces the cost of model deployment,and is conducive to the realization of real-time engineering structure bolt missing target detection by combining the model with UAV.Through the experimental study of bolt missing in steel truss structure,it shows that this method has the characteristics of high accuracy,fast recognition speed and strong generalization ability,and can realize the intelligent detection of bolt missing connection damage in steel truss structure model.