Neurocontrol of a ball mill grinding circuit using evolutionary reinforcement learning

Expert system based adaptive dynamic matrix control for ball mill grinding circuit

“…Increased fresh ore hardness, for example, can easily lead to a steady increase in circulating load which consequently causes mill overload or a mill circuit overload. In order to counteract the effects of these disturbances, a basic circuit usually has a minimum of two manipulated variables, namely fresh ore feed rate and dilution water flow rate (Conradle & Aldrich, 2001).…”

“…These control methods involve decentralized PID control and model predictive control (MPC) (Pomerleau, Hodouin, Desbiens, & Gagnon, 2000;Ramasamy, Narayanan, & Rao, 2005), multivariable control (Duarte et al, 1999), robust control (Galan, Barton, & Romagndi, 2002), supervisory control (Radhakrishnan, 1999) and neurocontrol (Conradle & Aldrich, 2001). Although numerous advanced control strategies have been reported in control literatures, both at simulation, as well as experimental levels, very few of these currently operate in industry due to the complex process dynamics and severe disturbances in grinding circuits.…”

Supervisory expert control for ball mill grinding circuits

“…Much work has been done to control grinding process variables. These control methods involve decentralized PID control and multivariable predictive control of particle size and circulating load (Pomerleau, Hodouin, Desbiens, & Gagnon, 2000), adaptive control of particle size (Najim, Hodouin, & Desbiens, 1995), model predictive control of particle size and circulating load (Ramasamy, Narayanan, & Rao, 2005) and neurocontrol simulation of particle size (Conradle & Aldrich, 2001). In recent years, advanced control strategies to improve mill capacity have become hot topics.…”

Hardware-in-the-loop simulation platform for supervisory control of mineral grinding process

“…But, these methods are hard to be applied in practical MGPs, as accurate modeling is difficult to achieve or the established models do not accurately describe the actual dynamic processes. Recently, intelligent technologies (i.e., rule based reasoning (RBR) [31], fuzzy logic [32], case based reasoning (CBR) [33], neural network [34], and reinforcement learning [35]) are used or integrated together to realize the supervisory control for the practical MGP.…”