Intelligent Recognition And Quantitative Analysis Of Defects In Road And Bridge Pile Foundation Structures

Concrete pile foundation structures are a common form of bearing in road and bridge construction.When problems such as substandard materials,non-compliant construction,loose soil,and inadequate concrete vibration occur during the construction process,different forms of defects such as cracks,separation,expansion neck,and contraction neck will occur,which can damage the integrity of the structure and reduce the bearing capacity and stability of the pile foundation.Currently,the main methods of detecting pile foundations on construction sites are low-strain reflection wave and ultrasonic detection.Ultrasonic detection requires the reservation of an ultrasonic tube in the pile foundation,and the detection process is complex and relatively expensive.Moreover,the analysis and reading of detection data need to be done manually,which requires high professional quality of the inspectors.Low-strain reflection wave is a low-cost,simple,and non-destructive detection method,which generally uses sensors to read the law of the excitation force propagating in the pile body during the detection process.Then,the method of analyzing the data characteristics manually is used for pile foundation detection.However,this method also requires extremely high experience of the inspectors,and the data volume of low-strain reflection wave is usually large,which makes the detection efficiency low and the detection process time-consuming and laborious.With the development of computer technology,using machine learning tools to train intelligent recognition models for pile foundation detection provides a new idea for monitoring pile foundation structures.This thesis aims to address the shortcomings and problems of traditional pile foundation detection methods,and conducts targeted research on low-strain reflection wave,and comprehensively uses signal processing,machine learning,deep learning,software development,numerical simulation,and other theories.Based on computer vision and hearing technology,combined with on-site experiments and software development technology,an intelligent identification scheme for pile foundation structure damage is established to achieve batch detection and quantitative analysis of defect size for intelligent identification of pile foundation structures.The main work and results of this thesis are as follows: The main work and conclusions of this article are as follows:(1)Through literature research,five common types of structural damage in bridge pile foundations were analyzed,and the different manifestations of each type of structural damage in low-strain reflection wave signals were analyzed.The advantages and disadvantages of the current detection methods were analyzed based on the current research status at home and abroad,providing a basic direction for the subsequent research of this article.(2)Finite element simulation and verification of pile foundations were carried out.Using the powerful numerical simulation function of finite element software,common defective pile foundations were numerically simulated,and field experiments and actual engineering verifications were carried out on piles of the same size.Based on the data characteristics of the field experiments,a series of modifications were made to the finite element model,and it was finally determined that the 16-meter-long pile in the field was 15.87 meters in numerical simulation,with an error of only 0.13 meters,thus verifying the accuracy of the numerical simulation.(3)Establishment of pile foundation detection data set.Based on the finite element modeling method in the previous section,the finite element software was developed for secondary development to realize batch generation of different defective pile foundation models.Random defective model data was generated according to different defect sizes,types,and positions,and finally,2000 defective pile foundation data were randomly generated,with 400 data for each type of defect.(4)Denoising and enhancement processing of defective data sets.Using the advantages of wavelet packet tools in signal-to-noise separation,the original data was denoised and enhanced from multiple perspectives such as time domain and frequency domain to improve the readability of the data.Using Python data processing and adding Gaussian noise operations,the original data set was enhanced.The one-dimensional data was dimensionally increased and data augmentation was achieved.(5)A convolutional neural network intelligent model and a convolutional neural network defect quantitative analysis model were built,and the models were trained to achieve intelligent recognition of pile foundation defect types and quantitative analysis of defect sizes.