Abstract:
The variation in soil layer has been regarded as one of the problems frequentlyencountered in foundation engineering discipline. Existing soil data, if available, can normallybe used as information for the design of foundation of a new facility. This is particularly true forthe project that occupies a large construction area where a number of bore hole is limited due toeconomic reasons. A tool that utilizes the data obtained from soil investigations, and capable ofpredicting soil layers or properties at other locations within the project area may prove useful forsoil engineering practice.The present study investigates the possibility of using Artificial Neural Networks(ANN) for soil layer prediction. ANN simulates the operations of a human-brain, making use ofexperience in solving the problem at-hand. Back propagation technique was used to develop asuitable weight matrix. A computer program called "NeuroSOIL" has been developed for theprediction of four soil parameters including Shear strength, Unit weights, Standard penetrationtest, and Soil type. Soil data obtained fiom 10 construction sites were used to train the neuralnetwork model. Each site contains data of at least five boreholes. The trained modules were thenused to predict the four soil parameters at locations where relevant data was not used in thetraining process. The prediction results obtained from NeuroSOIL were found to be in-line withexisting data with a 95 percent level of confidence. It is also found that level of accuracy for thepredictions depends mainly on the design of ANN architecture, the default value of Learningrate, Momentum, Number of iteration, and level of threshold.