Vehicle Noise Identification In Bridge Acoustic Emission Damage Monitoring

At the end of the "Twelfth Five-Year Plan",the "five vertical and five horizontal" pattern of Chinese comprehensive transportation network has gradually become clear,and the national rapid railway network and the national expressway network,which are suitable for the modern passenger and freight transportation,have basically taken shape.With the continuous breakthrough of technology,while the scale of transportation infrastructure is constantly refreshing,the transportation network structure has also undergone major adjustments.The traffic lines have continuously broken through the constraints of natural terrain,and the proportion of bridges and tunnels is increasing year by year.Considering the importance of bridge safety,in the technical system of structural health monitoring and evaluation,modern non-destructive testing(NDT)plays an important role and is one of the fastest growing and most branched technologies in the world.As a new type of non-destructive testing technology,acoustic emission technology has an intuitive dynamic capture capability for material and structural damage evolution.Increasing the signal-to-noise ratio and reducing the interference of redundant signals,invalid signals,and heterogeneous signals is a common concern of all non-destructive testing technologies including acoustic emission.For bridges,the traffic noise(vehicle noise and fleet noise)in the operation process is the most important excitation source of acoustic emission signals and the most common interference source of acoustic emission monitoring.The automatic identification and processing of vehicle noise is the most convenient way to take advantage of the dynamic and real-time technical advantages of acoustic emission.It is the premise to identify the difference between the acoustic emission characteristics of the vehicle noise and the damage signal and establish the comparison database to realize the automatic identification of the vehicle noise.It is also an effective technical path to verify the front-end denoising effect.In this paper,using the self-built program to extract the acoustic emission characteristic parameters,the "source" is distinguished by the external label of the vehicle signal,the fleet signal and the destruction signal,and then the neural network training is performed.The obtained network has a high recognition rate,and its key research The content is as follows:1.The noise simulation test is carried out in the laboratory to verify that the frontend denoising can be used to eliminate most of the noise during the acoustic emission monitoring process,and the method has poor effect on eliminating the noise of the vehicle.The indoor test and the on-site test were carried out separately,and the acoustic emission signals of the different load-section beams and the vehicle noise signals during the operation of the bridge were collected.Then the database was established to form a control library source for the identification of the vehicle noise during the bridge health monitoring process.2.The lead-breaking test is carried out before the formal loading,and the leadbreaking signal data is analyzed and processed.Then two data screening criteria are extracted: 1)integrity criterion,2)logical sequence(distance inverse ratio,proportional)criterion based on the lead-breaking test,and the acoustic emission signal of the beam damage is selected to obtain the database of the damage signal.The denoising analysis of the vehicle signal is carried out by using the characteristic parameter correlation graph analysis and the waveform graph feature,and the characteristic parameters are checked by the experience graph analysis to obtain a more accurate vehicle signal database.The normal distribution fitting is used to statistically distribute the characteristic parameters of the two types of signals in the database,and the distribution differences of each parameter are explored.3.Based on the literature,BP neural network was constructed initially,and the optimal hidden layer nodes,activation function,training function and learning function were determined by error correlation(MSE)analysis using the control variable method to optimize the network performance.The beam destruction acoustic emission signal and the vehicle noise signal database are respectively labeled,and then the identification effect is verified by taking the characteristic parameter as an input and using the signal source label as an output.The results show that after 3000 cycles,the correct rate of single recognition and the total recognition accuracy of BP neural network are above 95%.4.The time domain superposition model based on energy attenuation relationship is used to simulate and synthesize the fleet waveform signal data.The training and recognition are performed in the optimized BP neural network by taking the characteristic parameters extracted by the self-built program and the characteristic parameters of the beam-damaged acoustic emission signal as an input and using the signal source label as an output.In the BP neural network,the recognition accuracy rate is above 95%,which indicates that the BP neural network constructed in this paper can better distinguish the beam damage signal from the vehicle noise signal.