Research On Algorithms And Applications Of Control System Performance Assessment

Industrial process controllers can't achieve the design performance and even seriously deteriorate after a long time run. The reasons include changing (time variant) process operating conditions, process nonlinearities, unknown external disturbances and changes/problems associated with instrumentation and equipment. If the routine monitoring and maintenance is not carried out both effectively and efficiently, the economic benefits of automated process controllers will not become true. Due to the industrial value, the research of control performance assessment has attracted increasing attention in recent years. However, there still exist unsolved problems that restrict the industrial applications of monitoring technology. Based on pioneer's fruits, some problems of controller performance assessment are researched in this dissertation, and presents some innovation solutions.(1) The review of the history & current research status in control performance assessment(CPA) technology and industrial applications is presented. The main problems in CPA are synthetically analyzed.(2) The property of univariate feedback controller-invariance for minimum variance term(MVT) is first proved, and the MVT is only relative to time-delays . An improved linear regression method(ILR) with forgetting factor is proposed, which is employed to deal with performance assessment problem with time-variant disturbance dynamics. The parameters of time-variant disturbance model are perfectly estimated and tracked by introducing forgetting factor, and the corresponding minimum variance term can also track the time-varying disturbance model, and consequently, the stability and accuracy of performance assessment index are improved. Simulation results and industrial applications illustrate that the ILR method holds better stability and effectiveness for the time-variant disturbance dynamics, compared to traditional CPA methods.(3) The property of multivariate feedback controller-invariance for minimum variance term is first proved, and the conclusion that the minimum variance of original output is equal to the one of unitary interactor-filter output is further derived and analyzed. Next, a performance assessment and diagnose(PAD) technology for multivariable feedback controllers is developed, considering the lack of performance diagnose tools. The multivariable filtering and correlation analysis(FCOR) algorithm is combined with relative normalized gain array(RNGA) in PAD system. With PAD method, the performance for multivariable feedback control system can be correctly evaluated with FCOR algorithm, what's more, the controllers that need parameter-tuning are determined by RNGA. As a result, PAD technology can provide important guidance to the maintenance of multivariable controllers through simulation results and industrial applications.(4) This paper formulates a multivariable model predictive control strategy to achieve the goal of automatic variable-loading for the control limits in the variable-loading process of air separation unit. The technology structure, technological control schemes and system implementation process for the control strategy are described in detail. Regarding model predictive controller(MPC) as the core , the problems of coupling in the various technological parameters are solved. The control system can ensure perfect operation character and rate in the automatic variable-loading conditions and steady state, and produce significant economic benefits.