Research And Application On Corrosion Monitoring Of Rebar Based On Piezoelectric Impedance And Deep Learning

The steel bars are generally in a harsh working environment for a long time,which is likely to cause a series of safety problems due to corrosion,thereby affecting the service life of the structure.Therefore,the implementation of corrosion monitoring on key parts of the structure is of great significance.A large number of scholars have carried out related research using technologies including fiber Bragg grating,electrochemical methods,and acoustic emission.Among them,piezoelectric materials have gradually become a research hotspot due to their advantages of low cost and stable signal.The structural health monitoring technology based on piezoelectric impedance has been widely used in the fields of machinery,aerospace and civil engineering.In this paper,a corrosion sensor based on a piezoelectric-metal composite plate has been deeply researched on its signal characteristics when pitting corrosion occurs.A convolutional neural network is introduced to predict the mass loss rate of the sensor under different corrosion conditions.A finite element model of sensor random pitting corrosion considering the probability distribution of pit depth was established,and pitting corrosion simulation under arbitrary mass loss rate was realized.The study found that the location,radius and depth distribution of the pits can change the sensor’s peak admittance frequency under the same mass loss rate.Therefore,a one-dimensional convolutional neural network model was constructed to realize the classification and recognition of the sensor mass loss rate,and its prediction accuracy on the four types of piezoelectric impedance data test sets was above90%.The mechanical drilling to simulate pitting test and the accelerated to simulate uniform corrosion test of the sensor were designed,and the corresponding piezoelectric impedance data set was established as input.The results show that the convolutional neural network can measure the sensor corrosion mass loss rate under these two types of corrosion conditions.Visualization research was carried out,through t-SNE dimensionality reduction and outputting the middle layer activation channel feature map,studied the processing of the input data by the convolutional neural network.At the same time,the classification effect of the deep neural network and the recurrent neural network on the mass loss rate using the same data set was compared.The results show that the convolutional neural network has the advantages of better generalization,higher classification accuracy,and lower computational cost.Based on the actual situation,a research on the monitoring of steel corrosion in concrete was carried out.The influence of packaging materials and temperature changes on the piezoelectric impedance response of the sensor was systematically analyzed.The comparison summarizes the change trend of the conductivity curve of the sensor in the concrete environment.The convolutional neural network was used to accurately predict the corrosion mass loss rate of the sensor in the concrete environment.The above research provides technical support for further realizing quantitative corrosion monitoring of steel bars in coastal concrete structures.