Online discrete-time LQR controller design with integral action for bulk Bucket Wheel Reclaimer operational processes via Action-Dependent Heuristic Dynamic Programming

José Pinheiro de Moura; Patrícia Helena Moraes Rego; João Viana da Fonseca Neto

doi:10.1016/j.isatra.2019.01.010

Online discrete-time LQR controller design with integral action for bulk Bucket Wheel Reclaimer operational processes via Action-Dependent Heuristic Dynamic Programming

ISA Trans. 2019 Jul:90:294-310. doi: 10.1016/j.isatra.2019.01.010. Epub 2019 Jan 30.

Authors

José Pinheiro de Moura¹, Patrícia Helena Moraes Rego², João Viana da Fonseca Neto³

Affiliations

¹ UEMA, Brazil. Electronic address: josepinheiro@professor.uema.br.
² UEMA, Brazil.
³ UFMA, Brazil.

PMID: 30732992
DOI: 10.1016/j.isatra.2019.01.010

Abstract

In this paper, a novel approach for online design of optimal control systems applied to the bulk resumption process by bucket wheel reclaimer (BWR) is presented. This approach is based on reinforcement learning paradigms, more specifically Action Dependent Heuristic Dynamic Programming (ADHDP), that learn online in real-time the Discrete Linear Quadratic Regulator (DLQR) optimal control solution with integral action. Due to the geometric irregularities of the storage yard stacks and variation in physical and chemical characteristics of the stacked material, the flow control of solid bulks by bucket wheel reclaimer requires methods that are suitable with the high degree of imprecision of process variables and environment uncertainties. The resumption of bulk solids is carried out by dividing the stack into layers, each layer is approximately 4 m high, and the layers are divided into workbenches up to 12 m in length. To take up a workbench several translation steps are required (penetration in the stack), with the translation step varying from 0 to 1 m. In order to maintain the desired ore flow throughout the process, the BWR lance speed must be periodically adjusted. The main advantage of the proposed control method is that besides the decision rule is fully independent of plant model, the gains of the resulting controller are self-adjustable. The control system was designed in such a way that the ADHDP-based DLQR controller with integral action would act in real-time in the plant control, using only the input and output signals and states measured along the system trajectory.

Keywords: Action dependent heuristic dynamic programming; Bucket Wheel Reclaimer; Discrete Linear Quadratic Regulator; Integral action; Online design; Optimal control systems.