Virtual Reality and Neural Networks for Exploiting Geotechnical Data

Abstract

Design of foundations for large-scale civil works naturally involves soil characterization over considerable volumes. The 3D-interpretation of properties where only scarce geotechnical data is available is crucial for deriving effective and safe engineering decisions. Because of the ever-increasing cost of site investigation, it is neither practical nor economical to acquire geotechnical data at each point of interest for a complete definition of soils behavior. This situation makes it necessary to explore spatial-variability modeling alternatives that can manage limited geo-information. In this paper, a dynamic-neural procedure is developed for describing spatial relations between a set of geo-parameters easy-to-obtain. Once the network is finished, this topology is used to expand the small initial set of values into millions of computer-generated measurements. The massive database is incorporated into a Virtual Reality engine that facilitates the intuitively visual understanding of geo-information, permits to present all relevant data in a comprehensible format for decision making and provides a way to reduce very complex and diverse data sets into the essential elements without loss of data quality.