| The pile foundation has gained widespread applications that plays an important and fundamental role in the construction and other projects.The safety and stability of engineering is significantly related to the quality of pile foundation.However,the quality of the pile foundation is difficult to be guaranteed in because of the poor construction conditions of pile foundation,complex construction process,and the fact they are buried deep underground with a certain degree of concealment.The detection and monitoring of the pile foundation is vital for ensuring the quality of the pile foundation.Therefore,the non-destructive testing process of the pile foundation is an indispensable part of the foundation pile project.The detection environment normally contains complex noise during the non-destructive testing of pile foundations.Consequently,the detection signal has a relatively large loss in the propagation process,resulting in a small absolute amplitude of the received signal and a lower signal-to-noise ratio.The quality and cost of non-destructive testing is affected by the extraction of feature vectors of pile defect signals and the identification of defects with denoising and focus amplification of pile defect signals.These steps are crucial for non-destructive testing of stress wave reflection methods of pile foundation.Therefore,this paper focuses on several aspects such as signal denoising and feature vector extraction and recognition.Considering complex detection environment and complex noise of pile foundations,in this paper,we first introduce the empirical modal algorithm using with the kurtosis criterion and the signal autocorrelation analysis to improve the traditional empirical mode decomposition denoising method,where the signal is denoised partly.The simulation experiment results verified that the algorithm achieved certain denoising effects on the premise of retaining the signal feature information.Moreover,the wavelet threshold is introduced based on the initial denoising.The EMD wavelet threshold method is used to further denoise the signal after the initial denoising.Therefore,the denoising process of this part is completed through two denoising processes.The simulation results shows that the denoising of the defect signal based on EMD achieves a clear denoising effect.The lower recognition rate of the hidden features of defects and existing methods for feature extraction with higher number of eigenvector elements resulting in affecting recognition efficiency,where the signal characteristics cannot be truly displayed.This paper uses the empirical modal decomposition method to decompose the signal.The development of the mean-value feature vector of pile-based defect signals based on the information entropy for achieving the extraction of the feature vectors.The fuzzy clustering algorithm is developed containing mean-valued feature vector is reconstructed in the phase space.The reconstructed mean-valued feature vector is used in the fuzzy clustering algorithm to reduce the dimension for considering the selection of feature vector.The simulation experimental results shows that the feature selection based on fuzzy clustering achieves good recognition rate and low time consumption.Additionally,the effect of clustering number on recognition training is discussed with the help of simulation experiments where the selection of the data for phase space reconstruction and effects of the experimental recognition rate are discussed and analyzed.From the numerical simulation to the denoising of signal data to the final feature selection and defect recognition,the paper completes the processing and recognition of the defect signal of pile foundation,and has achieved good experimental results. |
