Application of Neuro?Fuzzy Technique in Mine Support System for Ground Control

In underground activities ground control is a challenging problem. It affects safety, production, and efficiency. As per statistics of accident data, ?fall of roof/sides? is one of the major causes of mine accidents. In many cases, our experiences and understanding of soil and rock behavior still fall short of being able to predict how the ground will behave. Presently, empirical approaches to design are widely used in estimating mine support parameters. Under these circumstances, expert judgment plays an important role, and such accidents can be obviated with accurate measurement and optimization of data and analysis using the neuro-fuzzy technique of artificial intelligence. Recently, a neuro-fuzzy hybrid approach has become one of the major areas of interest in engineering fields because it gets the benefits of neural networks as well as those of fuzzy logic systems, and it removes the individual disadvantages by combining them on the common features. Neural network and fuzzy logic technologies have common features, such as a distributed representation of knowledge, the ability to handle data with uncertainty and imprecision, etc., like in ground control. Fuzzy logic technique has tolerance for imprecision of data, while neural networks have tolerance for noisy data. In this paper we have focused an intelligent technique, i.e., neuro-fuzzy technique, to approximate the setting load given to the standing support (props) erected for the purpose of supporting freshly exposed roof during underground mining. We have used twelve input variables of rock parameters, and after having trained using a neural network with a sigmoidal function as an activation function, simulation was done to find the output parameter, i.e., the pre-load (setting load)to be applied on props. Outputs of the neural network were again fed to the fuzzy system together with the twelve parameters, incorporating five triangular membership functions for each parameter. Final optimum output was approximated using the MATLAB program, which was found to be satisfactory. Neuro-fuzzy technique has better performance over individual neural networks or fuzzy logic technique.