Data-Driven Operational Optimization Control For Mineral Grinding Processes

In the mineral processing industry,mineral grinding as the subsequent procedure of mineral crushing,further grinds the mineral material on the basis of mineral crushed,transforms the large particles of the run-of-mine ore into desired size of particles to allow for the mineral liberation between useful minerals and lode minerals,and to provide the fresh materials for the subsequent procedures in mineral processing.Due to the characteristics of high energy consumption and its central position among all mineral processing procedures,the mineral product particle size and operation efficiency of mineral grinding have huge effects on the concentrate grade and the productivity of a mineral processing plant,and they are highly related to the comprehensive economical indices of a mineral processing plant.Therefore,studies on the operational optimization and control methods for mineral grinding have been considerably and consistently drawing attentions of the researchers all over the world.Mineral grinding processes are composed of the inner loop containing fresh ore feeding control and sump water flow rate control,and the operational loop containing grinding product particle size and circulating load control.There are complex interactions among multiple ore feeders,the ball mill and the classifier equipment.The mechanisms of mineral grinding are complicated and hard to establish a mathematical model.The variation of operational conditions deads to a dynamic inner closed-loop system.The outputs of the inner loop are constrained.The objective is to regulate grinding product particle size and circulating load to prescribed ranges and to maximize throughput.The aforementioned challenges and difficulties lead to an undesired situation where it is difficult and challenging to apply conventional model-based optimization and feedback control methods in the operational control problem for mineral grinding.This dissertation is focused on the data-driven optimization and control methods for mineral grinding.The study on data-driven optimization control for mineral grinding is of great value and crucial importance,not only on the level of studies of control theories,but also on the level of the practical implementations in process industries.This dissertation is supported by the National Key Basic Research Program of China(973 Program)"Research on integrated control strategy and operational control methods for complex manufacturing of whole process control systems(2009CB320601)".And it is focused on the data-driven optimization and control for mineral grinding.The contributions of the dissertation are summarized as follows:1.The inner loop composed of fresh ore feeding and sump water feeding control is influenced by disturbances,such as that there are fluctuations in the ore feeding process resulting from the switching of multiple conveyor belts.Therefore,the closed-loop inner loop system presents dynamics from the view of the operational loop.For the operational control problem of mineral grinding under the dynamic environment of inner loop closed-loop system,an operational control method is proposed based on adaptive dynamic programming and barrier functions.The dynamics of the inner loop are transformed into the operational level with the sample time in the operational level.Barrier functions are introduced into the performance index of the optimal control problem to deal with the constraints on the inputs and states,ensuring that the constraints will not be violated during the solution of the optimal problem of setpoint following for mineral grinding.Adaptive dynamic programming techniques are used to obtain an approximate solution on line in real time without modeling the mineral grinding processes achieving operational optimization control for mineral grinding.Simulation models are built based on mechanisms analysis and data validation.Simulation results show that on the premise that no mathematical model is needed for operational loop,the proposed method can make the setpoints of both fresh ore feeding rate and sump water flow rate satisfy their respective constraints,and in the meantime,grinding product particle size and circulating load can be regulated to the prescribed ranges assigned by the grinding technology.2.Aiming at the difficult problem that the outputs of the inner loop are saturated by their constraints and thus it can hardly regulate the operational indices to the prescribed optimal target values,an operational control method is proposed based on adaptive dynamic programming and reference governor.A steady output steady input mapping is introduced to build a lookup table to design the reference governor.The reference governor can modify the setpoint to satisfy the input constraints through prediction.An approximate policy iteration algorithm is proposed based on adaptive dynamic programming techniques to regulate the operational indices to the optimal target value or the nearest feasible value.The proposed method is compared with a method without the reference governor through simulation and the simulation results show the effectiveness of the proposed method.3.Aiming at the problem of controlling grinding product particle size and circulating load within the prescribed ranges based on the grinding technology,and maximizing the throughputing rate,a multi-objective online operational optimization method based on extremum seeking is proposed.First,for the steady-state operation of mineral grinding,consider the range control and throughput maximization problem at the same time,and define performance indices and assign different weights,respectively,which formulates a single-objective optimization problem from the original multi-objective optimization.The constraints on fresh ore feeding rate and sump water flow rate are dealt with by an exact penalization-based method.The excessive terms of inputs are penalized by a penalization function.To ensure the steady operation of mineral grinding,dead zone operators are introduced and the control structure is modified accordingly,such that the oscillations caused by the modulating signals do not exist after a steady state is reached.In presence of the dead zone operators,the solutions of the error closed-loop system is uniformly ultimately bounded.In absence of the dead zone operators,the origin is a locally exponentially asymptotically stable equilibrium point of the error closed-loop system.Simulation experiments including a comparison with a barrier-function-based extremum seeking method have been carried out.The simulation results have shown that without system modeling for mineral grinding,the proposed method can accelerate the rate of convergence,and eliminate the oscillations after a steady state is reached,making the trajectories of the operational indices smoother,and the operation steadier which avoids over-grinding and under-grinding.It is guaranteed that fresh ore feeding rate and sump water flow rate can satisfy their respective constraints.The simulation results have also shown the effectiveness of the proposed method.