Research On Acoustic Emission Signal Source Location Based On Neural Network

At present,the construction of infrastructure in China is gradually improving.China’s technological strength in transportation,engineering construction and aerospace has rapidly upgraded.The construction and renewal of large facilities in various fields has accelerated.The safety of various large structures are more stringent,but usually large structures are more expensive to maintain.Structural health monitoring can detect damage and hidden dangers in large structures as early as possible,maintain structural safety and prevent damage caused by structural damage.Traditional non-destructive testing is very time-consuming.In terms of detection means,there is no real-time online large-scale monitoring function yet.And due to the complex structure of some equipment,internal damage or minor damage that is difficult to find on the surface cannot be detected and repaired in time,leaving hidden dangers to the structure.These problems are in urgent need of developing online,real-time,and efficient solutions.Structural health monitoring technology has the advantages of online,real-time,and efficient,which can better solve these problems.It is widely used in aerospace,civil engineering structures,transportation,and industrial equipment.Acoustic emission signal is a transient elastic wave generated when a structure is deformed,impacted,or fractured.Acoustic emission technology is an effective means of non-destructive testing.Compared with other NDT methods,acoustic emission technology can be monitored dynamically and passively in real time.Traditional acoustic emission signal localization methods need to calculate the wave velocity of acoustic waves,which is not suitable for some complex structures.The neural network can satisfy the localization of acoustic emission signal source in complex real production situation,and can obtain high network approximation speed.To address the above points,this paper investigates the following aspects:(1)The current status of acoustic emission signal processing research in China and abroad and the basic concepts of neural networks are described respectively.Acoustic emission signal processing is widely used in a variety of fields,and in recent years,the field of neural network-based acoustic emission signal processing has been gradually expanded.Research on pattern recognition of acoustic emission signals using neural networks has appeared and has been applied to structural health monitoring.In this paper,we analyze the prospects and advantages of neural network in acoustic emission signal processing,discuss the application of neural network model in acoustic emission signal detection,and try to use neural network for acoustic emission signal localization.The acoustic emission system consists of an emission source,a data acquisition device,and a host computer.The collected data are imported into the neural network,and this paper discusses the theory and structure related to neural networks.The specific structure of the unit structure neurons of the neural network and the weights,thresholds and activation functions required to construct the neurons are analyzed.(2)An acoustic emission signal processing system using piezoelectric sensors is built,and the acquisition experiments and model training process are described.The acquisition system acquires the acoustic emission signals in real time and uploads them to the host computer through a network acoustic emission meter.The network acoustic emission meter can directly display the waveform and characteristic parameters of the captured signal.Many feature parameters are filtered and pre-processed,and the pre-processed data are used as input parameters for the neural network.A BP neural network model is built and trained to find the parameters with better training results and obtain the calculated coordinates on the x-axis and y-axis,respectively,and the final localization results are obtained by inverse normalization.After comparing the final calculated results with the actual results,it was confirmed that the deviation values were between 0.5 cm and 2.3 cm.The conclusion confirmed the possibility and superiority of using neural networks for acoustic emission signal localization,and identified the shortcomings and deficiencies of the system.(3)The acoustic emission signal acquisition system based on FBG sensors is built.Experimental procedure and model training are described.Since FBG is resistant to high temperature and corrosion,it can be used in more practical production fields,and FBG sensor is used as the sensor device in the system.An array of waveguide gratings is used as the optical filter device to separate the reflected beam of FBG according to the center wavelength,then distinguishing the waveforms acquired by different FBG sensors.A data collector is used to collect the amplitude and arrival time of each channel waveform and preprocess the data.The preprocessed data are used as the input parameters of the neural network and introduced into the radial basis function neural network model for training,adjusting the parameters of the neural network model and obtaining better localization results.The localization results are compared with the localization accuracy of the BP neural network-based acoustic emission signal localization system.Advantages and disadvantages of the two systems are compared.