Identification of Pile Defect Based on Wavelet Transform and Neural Network

To improve the accuracy of the analysis of pile low strain testing signal, the metho(1) d which combines the wavelet analysis and artificial neural network is adopted. Abundant time-history velocity response signals of pile can be acquired by the low strain integrity testing of full-scale sound model piles and defective model piles with different types of defects. The time-history velocity response signal of pile can be decomposed by db5 wavelet. The power spectrum mean value reflecting the energy distribution can be extracted from every spectrum range as the feature value. These feature values from one signal makes up the feature vectors representing this signal. Using the feature vectors as input data, the BP artificial neural network can be designed to establish the non-linear mapping relationship between feature vectors of pile and pile defect type. The results show that the combined neural network model achieved high accuracy rates and can identify efficiently and intelligently pile defect type according to the feature vectors of the test signals.