A Novel Approach in Predicting Dump Truck Tyre Life in a Mine Based on Multilayer Perceptron Neural Network Optimised with Particle Swarm Optimisation

Abstract

Tyre hours/life deficit is a major operational challenge facing the mining industry which adversely affects materials production and costs. An accurate forecast of the tyre life is key in addressing this menace. This study for the first time employed the hybrid intelligent technique by utilising three metaheuristic optimisation algorithms, including particle swarm optimisation (PSO), genetic algorithm (GA), and whale optimisation algorithm (WOA), as trainers for the parametric weights and biases to optimise multilayer perceptron neural network (MLPNN) for enhancing prediction of on-site dump truck tyre life in the mine. Four hybrid models known as PSO-MLPNN, WOA-MLPNN, GA-MLPNN, and BP-MLPNN were developed using a total of 157 tyre dataset records obtained from a surface mine in Ghana. In assessing the prediction performances for the models developed, five statistical performance metrics of variance accounted for (VAF), Nash–Sutcliffe efficiency index (NASH), coefficient of determination (r²), mean absolute percentage error (MAPE), and correlation coefficient (r) were utilised. Moreover, ranking, uncertainty analysis and Bayesian information criterion (BIC) techniques were utilised to establish the most effective hybrid model, whereas sensitivity analysis was conducted on the input parameters. Results achieved showed that PSO-MLPNN was the best for prediction because it had the least MAPE value of 1.196% and relatively high values of VAF (99.642%), NASH (0.996), r² (0.996), and r (0.998). Besides, PSO-MLPNN had the best selection criteria values of 6, 7.1725, and 444.834 for the ranking, uncertainty analysis and BIC respectively. Hence, PSO-MLPNN is recommended for the prediction of on-site dump truck tyre life for the studied mine.