Research On The Assistant Design Method Of Cement Flyash Gravel Piles Composite Foundation With Neutral Networks

Cement Flyash Gravel piles (for short CFG piles) composite foundation is a type of low strength piles composite foundation, and has been widely used in constructions due to its great advantages such as the economy on materials, the celerity and facility in actualizing and the favorable effect on reinforcing. As the manipulative design parameters of CFG piles composite foundation, the eigenvalue of bearing capacity.fspk and the settlement value S of composite foundation has being an issue concerned by civil engineers. However, the design methods of bearing capacity and settlement of CFG piles composite foundation provided by active criterions are set up on some assumed conditions, and the results, especially settlement value are much different from the fact.Artificial Neural Networks (ANN in shortening) is a rising branch of subject developing rapidly from 1980s, which can find the relation between input and output deduced with the knowledge gained from learning and memory, thereby the result can be attained. It is maybe the most favorable means, especially when the question is a fuzzy system, which is not completely known on the mechanism and law or can not be described with mathematic models.ANN is attempted to be used in CFG piles composite foundation in this paper, and an assistant design model is established and achieved for CFG piles composite foundation with neural networks, which is based on the materials of CFG piles composite foundation from in-situ static load test of several projects in wuhan area. The main jobs and conclusions in this paper hereinafter:1.The active design methods on bearing capacity and settlement of CFG piles composite foundation are discussed, and the learning algorithm of Back Propagation neural networks and which how to be realized with Matlab are studied.2.The several key factors affecting on the bearing character of CFG piles composite foundation have been analyzed, and the following 11 targets are adopted as the input parameters in building the assistant design model with ANN: hammer number of standard penetration test N63.5, natural hole ratio e, hydraulic index IL, compressing modulus Es, compressing coefficient α1-2, friction angle φ, viscous force c, pile diameter d, pile length L, replacement ratio m, construct technics P.3.In the course of setting up the assistant design model with neural networks, a transitional layer has been set between the input layer and output layer of standard Back Propagation neural networks according to the particularity of issue, which makes the networks model have more explicit physical meaning, and a two-class neural networks model has been adopted which is based on the logical relation between bearing capacity and settlement.4. Fuzzy Analytical Hierarchy Process is introduced into the assistant design model with neural networks, and the effect of soil around piles has been evaluated, with which the problems of design model have been solved such as the overfull number of input point and the difficult achievement in logical disposal of input, meanwhile, the design model degenerates into the standard three-layer neuralnetworks model, and the model can be realized easily.5.The design model is proved to be steady by the training testing and prediction appraising, and the speed and precision of which are better than the current experiential methods and numerical methods.6.The following two points are based on mutual compare analysis of the key design parameters of bearing capacity and settlement of the design model: the effective approach of improving bearing capacity of composite foundation is to increase pile diameter and replacement ratio on the condition of the economical and rational pile length, and the effective approach of reducing settlement of composite foundation is to increase pile length on the condition of the rational pile diameter and replacement ratio.