Study On Defect Recognition And Location Of Road And Bridge Foundation Piles Under Layered Soil Conditions

During the pile engineering detection technique,the layered soil around the pile will have an impact on the inspector’s assessment of the integrity of the foundation pile by causing the time velocity curve to exhibit asymptotic shrinkage or asymptotic expansion curve features.To increase the precision of pile defect recognition and placement in layered soil conditions,multi-feature extraction,and recurrent neural network are integrated in this study for the first time.Here is the primary piece of work:(1)In terms of pile-layered soil,it is a recognized three-dimensional finite element model.Different layered soil conditions are constructed around the entire pile,the enlarged pile,the reduced pile,the segregation pile,and the fractured pile.The dynamic response curve of the complete pile and the defective pile is then evaluated,and the variation law of the soil layer’s elastic modulus is investigated.By adjusting the defect location,defect size,and soil border location,the temporal velocity curves were obtained.The reliability of the finite element modeling findings was confirmed,and they can serve as data sources for later Recurrent Neural Network model training.(2)Extraction of multi-dimensional features.The mean,mean square,variance,skewness factor,and kurtosis factor are extracted from the time domain dimension;the mean frequency,frequency center,root mean square frequency,and frequency standard deviation are extracted from the frequency domain dimension;the time-frequency domain dimension is decomposed using an improved complete ensemble empirical modal decomposition with adaptive noise(ICEEMDAN)method to decompose the signal,using the correlation coefficient criterion to filter the components and sample entropy and information entropy features are extracted from each component.The 17 Vetter levy set was formed as the input to the neural network.(3)Build a Recurrent Neural Network model.From the perspective of model construction,the pile defect recognition effects of LSTM(Long Short-Term Memory Networks)and BiLSTM(Bi-Directional Long Short-Term Memory Networks)are compared and analyzed.The findings demonstrate that,in layered soil conditions,the BiLSTM model’s fault recognition accuracy is 7.99% greater than the LSTM model’s.The recognition effects of stacked LSTM and stacked BiLSTM are compared and analyzed from the perspective of the model level.It is found that the recognition accuracy of the 3-layer LSTM model is 77.08 %,which is 4.86 % and 17.04 % higher than that of the 2-layer LSTM model and 4-layer LSTM model,respectively.Similarly,the recognition accuracy of the 3-layer BiLSTM model is 95.49 %,which is 19.1 % and21.99 % higher than that of the 2-layer BiLSTM model and the 4-layer BiLSTM model,respectively.Compared with the three-layer LSTM model,it increased by 18.41 %.(4)On the basis that the neural network can accurately identify the types of pile defects in layered soil,the research on the location of pile defects is further completed.The defects at different positions from the top of the pile are set up by using the finite element software.The simulation curve of the defective pile is decomposed and reconstructed by a wavelet packet to find the peak value.The results show that the average error of the wavelet packet-local peak finding method in the study of pile defect location is 2.92 %,which is 1.79 % lower than that of the local peak method.(5)The experiment is created and integrated with the engineering example to find and identify the foundation pile defect,which demonstrates the accuracy of the findings of the numerical simulation and the usefulness of the real-world application.To assure the engineering quality,it is more accurate to determine the type of defect and the location of the foundation pile under the situation of layered soil.