Subsurface soil-geology interpolation using fuzzy neural network

Soil geology plays an important role in selection of core soil for constructing rock-fill darns and in geotechnical evaluation while constructing major structures. Inferring the geology formations in the region between one borehole and another (cross-borehole region) is a human-intensive process of only moderate reliability. Improved operation planning and better geological assessment contributing to cost reduction can be achieved if reliability of inference can be improved. Cross-borehole interpolation using neural networks, such as the multilayer perceptron (MLP), is a relatively recent development and offers many advantages in dealing with the nonlinearity inherent in such a problem. However, neural networks alone are not sufficient to accommodate the fuzzy nature of the geological information. Cross-borehole soil-geology interpolation was investigated using a fuzzy-MLP neural network and is summarized in this paper. To train this network, data from borehole investigations were supplemented with artificial data created using human knowledge, which we term "data-based knowledge incorporation." The fuzzy-MLP neural network takes advantage of MLP neural networks and fuzzy set theory. Because of this, fuzzy-MLP not only interpolates but also provides an indication about the interpolation accuracy.