The engineering classification of rock masses is one of the basic problems in rock mechanics. Based on Geomechanic Classification System, a neural network method for classification of rock masses is proposed in this paper. The mapping relationship between rock mass classes and the parameters used in this approach is constructed. A number of engineering cases have verified the validity of this method.
La classification de traveaux de corps de rocheux est un des problèms essentiels dans le domaine de mechanique rocheuse. A base du systeme de la classification geomecanique, une methode de reseau nerveux pour la classification de corps rocheux est proposee dans le present document. La relation employee dans ce procede entre le corps rocheux et les parametres est construite. Une grande quantite des exemples de traveaux ont verifie la validite de cette methode.
Die ingenieurwessens klassifizierung der felsmassen ist ein der grundlegenden probleme in der felsmechanik. Auf der basis der geomechanischen klassifizierung systems wird ein neuronetzverfahren zur klassifizierung der felsmassen in diesem artikel vorgelegt. Das verhafnis zwischen feismassenklassen ist errichtet und die im verfahren benutzten parameter sind festgelegt. Die korrektheit des verfahrens wird von vielen lebendigen beispielen bestaetigt.
The engineering classification of rock masses is one of the basic problems in rock mechanics. The main purpose of the engineering classification of rock masses is to evaluate the behavior of rock masses comprehensively, to divide the rock masses into several proper classes and to offer quantitative data and guidelines for site choice, engineering design and construction. Therefore rock mass classification is very important for engineering practice, which make the classification an attractive subject for the investigators of rock mechanics.
Because the behavior of rock masses is uncertain and Complex, the classification methods now available for assessing rock masses are in most degree empirical, which are usually deduced subjectively by the investigators. However the ability to learn complex relationships and associations from examples makes the neural network an attractive alternative to the problems with uncertainty and complexity. In this paper, a neural network method for engineering classification of rock masses is proposed. Based on Bieniawski's Geomechanic Classification of Rock Masses, a three-layer back-propagation network was set up and trained by 100 manually-fitted examples and engineering cases to identify the rock mass classes. The training process is quite time-consurning, however, when new data of rock masses are fed to the trained network, the network will instantaneously give out the corresponding rock sort, which can do well enough to meet the requirements of engineering practice. Also the neural network can be continually modified and improved by learning from new engineering cases.
Artificial neural networks are massively parallel adaptive distributed information systems, which simulate the intelligence behavior and structure of human brain. A neural network contains large numbers of simple processing units and arrangements of dense connections. Its information is distributed on the connections among those simple processing units. As an engineering technique, neural networks can increase the speed of computation, because of their massively parallel nature.
