Pile Defect Diagnosis Based On Wavelet Analysis And Neural Networks

With the extensive application of the pile foundation, problems brought by the quality of piles are becoming more and more prevalent. Therefore, it is important to inspect the integrity of piles. After discussing the shortages of existing analysis methods in this field, an approach based on Wavelet theory and Neural Networks is proposed for pile damage assessment.The main research work is consisted of several parts. It includes the method to diagnose the fault types of piles using multi-resolution analysis and neural networks, to diagnose the fault types of piles using wavelet package analysis and neural networks, to diagnose the fault area in piles using multi-resolution analysis, and the method to diagnose the damage degree of piles using wavelet package analysis. Whatever method needs the work of extracting characteristics from dynamic stress-wave signals obtained by Sonic Integrity Testing method first. Wavelet transformation shows its effectiveness in this step. In this paper, an improved BP (back-propagation) neural network model was used to diagnose the fault types and damage degree of piles. The training and testing of the network were based on a database of several precast reinforced concrete (RC) piles and auger-injected RC piles from the laboratory site. The prescient fault types and damage degree of piles were ttsed as the desired output in training. The results of the wavelet decomposition of a residual obtained from the dynamic stress-wave signals are used to diagnose the fault area in piles.The study shows that crack identification via wavelet analysis is accomplished easily whereas it can hardly be detected by the traditional FFT method and the neural network model can predict the damage situation of piles reasonably well. This approach can improve the reliability of pile integrity inspection and is promising for being an assistant decision-making method for engineering practice.