Development and Performance Evaluation of an ANFIS-Based Intelligent Controller for Flotation Aeration Systems

Flotation is a nonlinear and multivariable mineral-processing operation in which variations in pulp conditions and aeration can significantly affect process stability. This study presents a simulation-based intelligent control approach using an Adaptive Neuro-Fuzzy Inference System (ANFIS) for regulating the aeration subsystem of a flotation process and compares its dynamic performance with that of a conventional Proportional–Integral–Derivative (PID) controller. A dataset of 200 synthetically generated samples was constructed within predefined operating ranges representative of the investigated flotation system, with simulated measurement noise incorporated to account for sensor uncertainty. Pulp level and froth intensity were initially considered as process-state variables, while air-valve opening was defined as the manipulated variable. The ANFIS model employed a Sugeno-type inference architecture with Gaussian membership functions and was trained using the Adam optimization algorithm. Using an 80/20 training–testing split, the model achieved RMSE, MAE, and R² values of 1.985, 1.502, and 0.967 on the training set and 2.810, 2.309, and 0.963 on the test set, respectively. For closed-loop evaluation, an ANFIS controller based on the control error and its variation was implemented and compared with a PID controller using a dynamic simulation of the flotation aeration subsystem. At a froth-intensity setpoint of 75 kg/min, the ANFIS controller reduced overshoot from 3.95% obtained with PID control to 1.93%, corresponding to an approximately 51% reduction. The settling time decreased modestly from 16 s to 15 s, while the ANFIS controller also produced smoother air-valve control actions under the simulated operating conditions. These results indicate that ANFIS can provide improved dynamic regulation, particularly in terms of overshoot suppression, within the investigated simulation framework. As this study is simulation-based, the results represent a proof of concept. Experimental validation and integration with other flotation control loops are required for practical implementation.