ABSTRACT

Many methods to calculate the consolidation coefficient of soil depend on judgment of testing curves of consolidation and the calculation result is influenced by artificial factors. In this work, based on the main principle of back propagation neural network, a neural network model to determine the consolidation coefficient is established. The essence of the method is to simulate a serial of compression ratio and time factor curves because the neural network is able to process the nonlinear problems. It is demonstrated that this BP model has high precision and fast convergence. Such method avoids artificial influence factor successfully and is adapted to computer processing.

INTRODUCTION

Settlement of soil is generally caused by load, although other causes such as lowering of the water table, vibrations and so on may also cause settlement. At the meantime, the soil volume is decreased, so consolidation occurs. Consolidation is a time-related process involving compression, stress transfer, and water drainage. A general consolidation theory of soils should take into consideration stress and strain conditions in three dimensions. Accounting for the wide variations In properties of soil, the result of this problem is mathematically impossible. Indeed, three-dimensional consolidation analysis focuses on numerical or finite-element procedures. In 1923, Karl Terzaghi advanced a one-dimensional consolidation theory and simplified the calculation. In fact, the vertical compression or consolidation is frequently the largest and the most important component.

(Equation in full paper)

The average degree of consolidation is a very important parameter to predict the settlement at a given time. The key to calculate the parameter U is to determine the consolidation coefficient CV, which can be obtained through laboratory test. TV is proportional to time t and U is directly relevant to testing reading R in a one-dimensional consolidation test. Thus, the relation between R and t can be obtained to determine CV. Many methods to calculate CV have been put forward by scholars at home and abroad. The methods described in Ref.[1] include the logarithm-of-time fitting method, the square-root-of-time fitting method, the trial calculation method, the inflection point method, the three-point method and Scott method. Each has its particular properties. The logarithm-of-time fitting method and square-root-of-time fitting method are the most commonly used to calculate CV, nevertheless they bring about great errors in the results. The trial calculation method has a high precision, and the process of calculation is much too troublesome. The reverse turned point method needs to judge the reverse turned point and the result precision cannot be ensured. The three-point method is recommendable in that it is simple and the error is acceptable. The Scott method needs to consult its diagram. However, all these methods except three-point must be combined with chart or diagram. The result is often influenced by artificial factor and the calculations are not fit to apply to computer program. According to the nonlinear map property of artificial neural network (ANN), the authors put forward in this work an ANN method to determine the consolidation coefficient based on the principle of Scott method.

This content is only available via PDF.