Business Challenges 

Diesel fuel is a major energy source and cost driver in surface mining. Haul trucks consume a large portion of this energy, and their consumption is affected by gross vehicle weight, truck speed, total resistance, road profile, rolling resistance, grade, operating behaviour, and environmental conditions. Because the relationship between these parameters and fuel consumption is nonlinear, a conventional deterministic model was insufficient for accurate prediction and optimisation.

Expanded Technical Solution

The project developed a Multilayer Perceptron Artificial Neural Network (MLP-ANN) to predict haul-truck fuel consumption. The model used gross vehicle weight, truck speed, and total resistance as core input variables and produced fuel consumption as the output. The selected architecture included three input variables, one hidden layer with 15 cells, and one output layer. Sensitivity analysis was used to evaluate the influence of each input variable and to support operational decisions about payload, speed management, and haul-road condition improvement.

Technical Work Packages/Methods

• Collected and prepared mine-site haulage data including payload/gross vehicle weight, speed, total resistance, and fuel-consumption observations.

• Normalised data, removed outliers, and separated training, validation, and testing datasets.

• Developed, trained, and tested an MLP-ANN architecture with three input variables, 15 hidden neurons, and one fuel-consumption output.

• Used model validation and sensitivity analysis to evaluate predictive performance and variable influence.

• Converted the trained ANN into a practical fuel-consumption fitness function for operational analysis and optimisation.