Search

In situ characterisation of rock is crucial for mine planning and design. Recent developments in machine learning (ML) have enabled the whole learning, reasoning, and decision-making process to be more efficient and accurate. Despite these developments, the application of ML in rock-cutting is at an early stage due to the lack of mining applications of mechanised excavation leading to limited availability of data sets and the lack of the expert knowledge required when fine-tuning models. This study presents a novel approach for rock identification during mechanical mining by applying a self-adaptive artificial neural network (ANN) model to classify the rock types for selective cutting, in which datasets from two novel cutting operations (actuated disc cutting (ADC) and oscillating disc cutting (ODC)) were employed to test and train a model. The model was also configured with the Bayesian optimization algorithm to determine optimal hyperparameters in an automated manner. By comparing the performance of each evaluation, the model was trained to identify the best set of hypermeters at which uncertainty is minimal. Further testing indicated the model is very accurate in classifying rock types for ADC as the accuracy, recall, and precision all equal unity. Some misclassifications occurred for ODC with the accuracy, recall, and precision ranging from 0.68 to 0.99. The promising results proved the model is a robust and scalable tool for classifying the rock types for selective cutting operations enabling the interpretation to be performed more precisely, selectively, and efficiently. Since mechanical cutting requires significant energy, any improvement in matching machine characteristics to the rock mass will increase productivity, and energy efficiency and reduce cost. [ABSTRACT FROM AUTHOR]