Abstract:
Froth flotation is a cornerstone process in the mining industry for separating valuable minerals like
gold from gangue, yet its efficiency is often compromised by the complex, non-linear interactions
among various operational parameters (e.g., reagent dosage, particle size, pH). Traditional
optimization methods heavily rely on subjective operator experience, leading to suboptimal and
inconsistent gold recovery rates. This research addressed this challenge by developing and
validating data-driven predictive models and an optimization framework to enhance gold recovery
from the froth flotation process at Wagagai Mining (U) Ltd. Using 1,000 historical operational
records, predictive models based on Multiple Linear Regression (MLR), Random Forest (RF), and
Artificial Neural Networks (ANN) were developed after relevant feature selection (Modifier,
Activator, Collector 1, Collector 2, Frother, and Residence Time). Model performance was
evaluated using R-squared (R²) and Root Mean Squared Error (RMSE). The ANN model
demonstrated superior predictive performance (Testing R² = 0.879) compared to RF (Testing R² =
0.819) and MLR (Testing R² = 0.556), effectively capturing the complex relationships within the
data. The best-performing ANN model was then coupled with a Genetic Algorithm (GA) to
identify optimal process parameter combinations for maximizing gold recovery, constrained by
operational limits. The GA optimization identified an optimal reagent dosage configuration that
the ANN predicted could yield a maximum gold recovery of 99.06%. Sensitivity analysis on the
optimized model revealed that Modifier (Na₂CO₃, 38%), Activator (CuSO₄, 25%), and Collector 1
(HB-33A, 20%) were the most influential reagents affecting recovery. A comparison with Wagagai
Mining's existing practices showed historical monthly gold recovery fluctuating between 87.00%
and 94.01%, significantly lower than the predicted optimum. The company's current approach
involved inconsistent reagent dosing, often leading to over- or underdosing relative to the
identified optimal levels. Implementing the ANN-GA optimized strategy is estimated to yield an
annual reagent cost saving of $657.01 while achieving substantially higher recovery. This study
underscores the significant potential of integrating machine learning and genetic algorithms for
optimizing mineral processing operations, offering a data-driven approach to improve gold
recovery efficiency, reduce operational costs, and support more sustainable mining practices at
Wagagai Mining and potentially similar operations.
