ABSTRACT

Rock mass classification systems and the neural networks have similar characteristics: both employ a data base for their development and weights are used in the processing. The main rock mass classification systems, Q and RMR, can be written as local neural representations. Distributed neural representations can increase the reliability in geomechanical classifications, but the most important contribution of the work is to open new possibilities of interpretation of the classification process. The paper shows local representations of both Q and RMR systems, and the further development of a neural network to identify wall and roof instability with good results.