Nonlinear autoregressive-moving average-L2 (NARMA-L2) controller for multivariable ball mill plant

Abstract

Mineral processing facilities concern an enormous amount of dynamically complex unit operations (due to nonlinearities), for instance ball mill system. Normally, these processes need multivariable controllers to smooth actions by designing for plant constraints such as deadtimes and dynamics interactions. The present work presents a comparison between a classical PI and nonlinear moving average autoregressive-linearization level 2 (NARMA-L2) controllers based on artificial neural network (ANN) for a ball mill system. The manipulated variables of this plant are the rotation velocity (Vr) and the feeding weight (Wf), while the controlled parameters are the hold up (HU) and the mass fraction under 45 μm (P45). The simulation was built in the MATLAB software (Simulink), comparing the actions of PI and NARMA-L2 controllers in the face of operational changes in specific regions (constraints). The performance of proposed controllers was verified by the integral of absolute error (IAE), integral of squared error (ISE), or the integral of time-weighted absolute error (ITAE). The results of simulation showed the validity of the model obtained and the control technique proposed in this paper, which contributes to studies of multivariate controller designs for ball mills with significant applications. Additionally, this paper brings a first hybrid approach (PI/NARMA-L2) with successful implementation described in the literature.