Stability assessment of homogeneous slopes loaded with mobile tracked cranes—An artificial neural network approach

Abstract

The assessment of stability of homogeneous slopes found as part of embankments, approach ramps, in bridge construction or flood protection levees could be a complex task. Either during construction or at a point in the operating life of the earth structure it can be subjected to loads from the equipment operating on it. Mobile tracked cranes used in heavy lifting or dredging operations can apply loads due to their substantial self-weight combined with the load carried by them. It is important to be able to determine the minimum factor of safety for such slopes. However due to the combination of soil parameters, slope geometry and the variable nature of loading imposed, a substantial number (measured perhaps in hundreds of combinations) slope stability analyses would be required to find the minimum factor of safety. One approach to reduce the number of analyses needed is to develop an Artificial Neural Network, train it using a representative dataset of stability analyses, and rely on its predicting capabilities to determine the minimum factor of safety for the slope for any combination of model parameters. Artificial Neural Networks can simulate the central nervous system of a human brain, by training them using the input data and target data one can build a neural network and use them for the factor of safety prediction. Since this thesis considers the case of homogeneous constructed slopes, thus the slope stability analysis was performed using Bishop Simplified Method, and the load distribution due to mobile tracked cranes was represented by an equivalent triangular distribution acting on the slope surface. The slope stability analysis was performed using Slide (from rocscience Inc.) to obtain the training dataset and MATLAB was used to develop and train the artificial neural network. A detailed investigation to assess and improve the network accuracy was carried out, and it was established that by increasing the neuron numbers and hidden layers, the ultimate average error in predicting the factor of safety for an independent test data set was 0.677%. This error, considering the inherent uncertainty of soil properties, instils confidence in using the Artificial Neural Network for predicting the factor of safety of homogeneous slopes loaded by mobile cranes.