A NOVEL METHOD FOR DIAGNOSING PILE INTEGRITY USING BACK-PROPAGATION ARTIFICIAL NEURAL NETWORKS

摘要

Foundation is the utmost important element of a building, especially for high-rise structures. Any defect or damage of the pile segments can seriously destroy the building structure above. However, the satisfactory assessment of such pile is by no means an easy task. It is important to develop the efficient technique to determine the accurate integrity of foundation pile. In this paper the artificial neural network (ANN) program based on the low strain dynamic test is presented for diagnosing pile integrity. The records of pile integrity testing as well as the length, cross-sectional area of pile and wave velocity are identified to be the most reliable and are used as the input data set in the network. The back-propagation learning algorithm is employed to train the networks for extracting knowledge from training examples. In order to determine the level of accuracy of the pile integrity, the novel approach is proposed containing two back-propagation ANN models. The classification of pile integrity can be determined using the first model. Then the second model is responsible for the further investigation of the exact degree of pile defect and corresponding location. The actual piles are used to illustrate the process. After the diagnosis for defect type, a total of thirty-six patterns of neck pile were particularly selected for the training while four others for testing. The results from the testing phase indicate that the neural network was successful in modeling the relationship between foundation pile integrity and the other parameters and generally gave the reasonable predictions.