Introduction of an Objective Matrix Coding Method for Rock Engineering Systems through Self Organising Maps

Determination of the nonlinear behavior of multivariate dynamic systems often presents a challenging and demanding task. Therefore, complex geomechanics systems are an ideal area for the application of intelligent methods. Often the physical problem is controlled by nonlinear equations, conveying to the use of neural networks. Hudson, (1992a), aiming at efficient engineering modelling and design, introduced and developed Rock Engineering Systems (RES) approach, through the implementation of generic interaction matrix GIM. This approach has been used for a wide variety of rock engineering and other problems, resulting in many case studies available, which illustrate the application of RES to rock engineering problems. Considering the spectrum of the coding methods for the Generic Interaction Matrix, we suggest an objective coding approach. Among the five proposed coding GIM methods, the coding method according to the slope of Pi and Pj relation is relatively accurate, but it might be difficult or impossible to establish or estimate the exact relation of Pi vs Pj for most of the variables involved within a rock engineering system. Self organising maps, are a special type of unsupervised artificial neural networks, which can learn from complex, multidimensional data, and transform them to visually decipherable clusters. The proposed, coding of the GIM, is performed through the combined use of scatter plots produced through, (SOM), and the coding according Pi vs Pj slope. Naghadehi et al., (2013) constructed a database of worldwide experience, in open-pit mine slope stability, the selection of the principal factors was based on Hudson (1992a) who proposed an ‘atlas’ of factors affecting the stability of rock slopes. Slope classification was performed according to status of stability. The rating of the GIM matrix was completed, coupling scatter matrix produced by SOM training, and Pi vs Pj slope plot. Results refer to the identification of clusters, according to status of stability within the original multivariate data set, and the rating of the principal factors according to the dominance and interaction intensity within the open pit mines RES.