Artificial intelligence approach to calculation of hydraulic excavator cycle time and output

Abstract

Accurate prediction of the cycle time and output of tracked hydraulic excavators is notoriously difficult, not least because such data are freely available to the mining practitioner from only a limited number of plant manufacturers. Previous research attempted to rectify this problem through the development of ESTIVATE. ESTIVATE utilized a multiple regression (MR) equation to predict machine cycle time and subsequently, on the basis of this, to estimate machine output and excavation costs. However, with a coefficient of determination (R²) of 0.88 and a mean absolute percentage error (MAPE) of 20%, the MR equation failed to provide an adequately robust predictor of machine cycle time. Improvement to ESTIVATE's predictive capacity was sought through the use of a feed-forward artificial neural network with back-propagation training. With a sum square error of 0.194 and a MAPE of 7% (that is, a 14% reduction on the equivalent MR equation) the feed-forward network provides a significant improvement over the MR equation. An aim in future work will be to expand the capability of the model to include long-reach machines.