On the Application of Genetically Optimized Artificial Neural Networks in Geomechanical Classification Schemes

摘要

The primary purpose of the Rock Mass Classification methodologies is to classify geologically different rock masses into behaviorally and mechanically homogeneous categories. A variety of classification methods such as Barton's (Q) and Bieniawski's Rock Mass Rating (RMR) have been used. This paper uses a genetically optimized feed-forward back-propagation-type Artificial Neural Network (ANN), applied to geological/geotechnical data from engineering projects in Greece. Results show that with the use of the Q classification variables the ANN can place a rock mass in the aforementioned RMR classification rating very quickly and with very high accuracy. Further, the engineering bias often present in the traditional ways of achieving these ratings is reduced.