Enhancing downstream operation through run-of-mine crusher selection model: an application of edge detection software and soft computing algorithms
Babatunde Adebayo,Ebenezer Oyedele Ajaka,Thomas Busuyi Afeni,Blessing Olamide Taiwo,Ismail Adeniyi Okewale,Abiodun Ismail Lawal,Bidemi Olaoluwa Ogunyemi,Waliu Olalekan Apena,Victor Abioye Akinbinu,John Olurotimi Amigun
DOI: https://doi.org/10.1007/s12517-024-11933-4
2024-03-15
Arabian Journal of Geosciences
Abstract:During quarry design, most of the crushers selected at this stage are not appropriate for the muck-pile size ranges produced after rock fragmentation. Therefore, fragmentation performance, quality rating, and fragment size distribution prediction for suitable crusher selection were investigated to bridge the gap of mismatched crushers in quarries. Bulk density and P-wave velocity were determined in the laboratory. In situ rebound hardness values were obtained for the different case study sites. Also, rock quality designation (RQD) and rock mass rating (RMR) were determined. Field data obtained includes geometrical parameters, crusher specifications and information, a digital image of the muck pile, and crusher data. Particle distribution cumulative curves for 183 digital blast images were utilized to determine the 80% passing size using WipFrag. Models for crusher selection were developed using artificial neural network soft computing techniques with nine parameters (5 uncontrollable parameters and 4 controllable parameters) as inputs, and thereafter, the performance of the models was evaluated. The results of mean bulk density, rebound hardness value, P-wave velocity, RQD, and RMR are 2886.4 kg/m 3 , 46.1, 4153.4 m/s, 64%, and 71, respectively. The muck-pile particle size distribution analysis revealed that 80% passing size and Uniformity Index are 387.2 mm and 1.78, respectively. In addition, the relationship between Uniformity Index and powder factor suggests that a high powder factor can result in an increased Uniformity Index. The result of crusher classification is in the range of 1–5 unique classification identifier (UCI) for large and small crusher gaps. Three models were developed, and it was discovered that the third model, trained with eighteen (18) inputs and fifteen (15) neurons using a hybrid transfer function, has the highest prediction accuracy based on the model correlation coefficient. This model will select a suitable crusher that will match the resulting muck-pile particle distribution, thereby eliminating mismatches in crushers in quarries.
geosciences, multidisciplinary