PREDICTING MINING MACHINE CUTTING TOOL WEAR USING NEURAL NETWORKS

Abstract

To improve safety and economic productivity, the mining industry has been striving towards completely unmanned underground operations. The potash sector has participated in this effort, and has succeeded in automating or remote operating the continuous mining machines used in these mines. However, the detection of worn cutting tools on these machines has remained a manual function performed by experienced operators. In this thesis, research into a method of automatically scheduling cutting tool replacement outages is described. Recurrent neural networks were used to identify the dynamic process of mining machine revenue generation with tool wear. A genetic algorithm technique was employed to train the neural network on line. The trained neural network was used to predict the productivity of the machine following a postulated outage, thereby allowing an informed decision as to whether an outage would be beneficial. The results of this project show that an on line dynamic neural network system can be employed to schedule outages for a continuous mining machine for worn cutting tool replacement. The method approached the productivity realized with current outage scheduling. However, the potential for fully automating mining operations may be facilitated by this method.