Integrated and intelligent design framework for cemented paste backfill: A combination of robust machine learning modelling and multi-objective optimization

Abstract

Modern mining industry thrives for energy-efficient, clean and sustainable mining processes. The cemented paste backfill (CPB) technology, which may constitute 25–30% of the total mining cost, is no exception. One of the major bottlenecks for the current CPB design is that different steps were considered separately. No integrated design frameworks have been proposed, hindering the selection of the optimal CPB processing parameters. Towards this end, this study introduces an integrated and intelligent design framework for CPB (IIDF_CPB). The efficiency and accuracy of the proposed IIDF_CPB rely on two important parts. For one thing, robust machine learning (ML) modelling from constituent materials/processing parameters to performance indicators is established. Accurate ML modelling can save lots of time and substantially reduce the number of lab experiments. For another, IIDF_CPB is inherently a multi-objective optimization problem where two or more objectives need to be optimized simultaneously. The methodology of IIDF_CPB is presented and its feasibility is validated using a comprehensive case study. In the case study, ML modelling is conducted using a hybrid method that combines gradient boosting regression tree (GBRT) and particle swarm optimization (PSO). The non-dominated sorting genetic algorithm II (NSGA-II) is employed to maximize two conflicting performance indicators, namely slump and unconfined compressive strength at 28 days (28-UCS). The case study shows that the GBRT-PSO is robust in the slump and 28-UCS predictions. The average correlation coefficient between experimental and predicted outputs is 0.970 for slump and 0.991 for UCS. NSGA-II is effective in the concurrent optimization of slump and 28-UCS, which determines the Pareto front and maintains the diversity of non-dominated points.