APPLICATION OF ARTIFICIAL NEURAL NETWORKS OF THE MULTI LAYER PERCEPTRONS TYPE (MLP) IN THE PROPOSAL OF A NEW MODEL FOR STUDY OF SLOPE RESISTANCE

Application of artificial neural networks of the Multi Layeerr Perceptrons Type (MLP) in the proposal of a new model for study of slope resistance

Abstract

The events after Mariana and Brumadinho demonstrated to Brazilian society the urgency and the need for more in-depth studies on the stability of slopes, including large busbars such as those used in mining. The present work sought to build a safe and faster method of obtaining the Safety Factor (FS) to guarantee the stability of slopes, using neural networks. For comparison purposes, a traditional calculation method known as Fellenius (1936) was chosen to obtain the Safety Factor (FS) for different characteristics of soil and slopes with or without the water table applied to the training of the neural network. A cluster was built in which the same input parameters were given: specific weight (KN/m3), cohesion (KPa), friction (0) and slope (L / H) with the dry FS as the outlet. After training the network, 30 sets of data were tested using the traditional method and neural networks, which obtained 95% adjustment to the values ​​obtained by the method of Fellenius (1936). Thus, it has been shown that neural networks, after the improvement and understanding of the method, can be used with numerous advantages over the traditional Fellenius method for calculating slope stability.