Machine learning-based prediction of machine hours and fuel consumption: a case study in Aruwakkalu limestone quarry, Sri Lanka

Abstract

The machine hours are of paramount importance in the mining sector as they directly impact production levels, operational costs, and overall efficiency. Accurate prediction of machine hours and fuel consumption using machine learning techniques relies on the availability of a comprehensive historical database. This prediction study focuses on a site-specific context and is specifically applicable to large-scale open pit limestone mines, such as the renowned Aruwakkalu Limestone Quarry in Sri Lanka. The scientific objective involves analyzing the accuracy of algorithms and utilizing a highly precise model to predict machine hours and fuel consumption based on monthly tonnage of limestone. This scientific study utilizes four machine learning algorithms: decision tree, linear regression, LGBM (Light Gradient Boosting Machine) regressor, and random forest. The assessment utilizes monthly report data from the quarry spanning a three-year period. Subsequently, the chosen model is applied to predict machine hours and fuel consumption based on tonnage. Findings reveal that the decision tree algorithm demonstrates remarkable accuracy for dump trucks compared to other methods, while the LGBM regressor performs better for excavators and dozers. For predicting fuel consumption, LGBM outperforms dump trucks; decision tree excels in excavators; random forest achieves dozer accuracy.