G. Gilles

Modeling of a distributed parameter process with a variable boundary: application to its control

Abstract Study of an industrial chemical tubular continuous reactor led to a nonclassical distributed parameter model involving a boundary point whose position is time-dependent and an implicit limit condition. This is due to a phase changing phenomenon (evaporation) within the reactor which occurs at a precise point for a given time. This paper deals with a method of solving the hyperbolic partial differential equations with Dirichlet-type limit conditions, one of them being implicit and time-dependent. Then a control model is proposed, keeping this physical aspect in its structure. This control model is used with a finite-dimensional multivariable optimal control law, both in a simulation and in applications to an industrial reactor. The results allow the proposed methods to be generalized for similar chemical or physical processes.

An Example of Computer Control Systems Practical Teaching

Abstract The practical aspects are important in teaching sampled-data control systems. This paper presents a practical illustration of sampled-data systems main concepts in a French University education system. This teaching is progressive, for the illustrated methods as for the apparatus used. Thus, we begin with experiments where a digital minicomputer controls a simple process simulated on an analog computer. Then, this minicomputer is connected to a pilot plant, allowing to consider more concrete problems, and pointing out the classical methods limits. The minicomputer involves a EASIC software admitting calls for subroutines written in an assembler language, what permits to introduce easily real time programming.

Linearizing Control of a Class of Non-Linear Continuous Processes

Abstract Most of the continuous industrial processes working inside a large operating range can be modelled by non-linear state equations that are linear in control. A way for controlling those processes can be a closed loop linearization. It is shown that, when the solution exists, we get an explicit non-linear state feedback control law. The linearizability condition is demonstrated and it is shown that the control law may involve some singularities. Two practical applications on chemical and hydraulic pilot plants are presented.

Identification of Different Discrete Models of Continuous Non-Linear Systems. Application to Two Industrial Pilot Plants.

The continuous industrial plants dynamically working inside a large operating range involve non-linear phenomena that mostly cannot be suitably approached by linear models. In order to describe their behaviour, it is then necessary to use non linear models and/or variable parameter models. The control of such systems can be managed from adaptive methods or multi-model techniques but it may be preferable to try to find a global non-linear model correctly describing the system behaviour in all its operating conditions and a unique control law. Moreover, the goal being the control by means of one or several digital processors, it is necessary to build a discrete non-linear and/or time-varying parameter model.

Identification and Predictive Control of a Glass Feeder

Abstract This paper presents an identification and predictive control strategy for glass. feeder used in the glass containers manufacture.The combustion air action and the ventilation action .are introduced simultaneously in order to obtain a more homogeneous glass prior to forming machine in place of only one control action for the former control system. Based on the zone model estimated by Prediction Error method, the two control actions are computed With OPC algorithm proposed by CLARKE et al . The experimental results show that. the new control strategy has the superiority in the rejection of the measurable disturbance and In the adaptation when the glass type is changed in the feeder as compared with the former Industrial(PID type)controller(TCS)

Dynamic Modelling and Simulation of an Industrial Amidation Reactor

Abstract In order to get a better control of an amidation industrial process, a dynamic modelling study of the plug-flow reactor has been carried out. An infinitesimal phenomenological balance leads to a set of coupled partial differential equations. The chemical static part of the model is got from an analogy with respect to a batch reactor kinetics. Identifiability results of this non linear model are presented. From the non linear distributed parameter dynamic knowledge model, a static model is drawn out and leads to the writing of a static simulation package in order to find the static profiles and to choose the operating points.

Non-Linear Discrete Time Modelling and Control of a Pilot Neutralization Chemical Reactor

Abstract Many continuous industrial processes working inside a large operating range can be modelled by variable parameter and/or bilinear continuous state equations. In the case of processes with variable parameters, the state affine models allow to get a discrete approximated representation. An original direct identification method, involving a variable parameter filter, is processed inside the whole operating range in order to estimate the model coefficients. A dual approach leads to state affine control algorithms by gathering a family of linear controllers into a unique control law. In the case of bilinear continuous processes, their discrete model is of linear time-varying type. A way for digitally controlling them can be a closed loop linearization which leads to an implicit control law. An improvement can be obtained by adding integrators. Experimental results got, in each operating case (variable parameter or bilinear system), from a pilot neutralization reactor show the applicability of the methods.

Digital Control of Bilinear Continuous Processes. Application to a Chemical Pilot Plant.

Abstract Many continuous industrial processes working inside a large operating range can be modelled by bilinear continuous state equations. Their corresponding discrete model is of linear time–varying type. A way for digitally controlling those processes can be a closed loop linearization. The control Law being implicit, the solution needs, during each sampling period, the use of a numerical iterative algorithm. This non–linear state feedback control law also requires observing the state of the process ; it is shown how a discrete non-linear observer is derived from the discrete model. Simulation results are given concerning a monovariable process (chemical pilot plant) and a multi–variable system (D.C. motor). The robustness of the control law is pointed out and it is shown how to get a decoupling control. Finally, experimental results from an application to the digital control of a chemical pilot plant are presented and discussed.

Use of Transinformation in System Characterization and Identification

Abstract Information theory can bring interesting contribution to System Analysis : first, before and during analytical modellisation, then, after this modeliisation, in allowing characterization and sometimes identification of some systems classes. After recalling this theory basic concepts used for our purpose, their application to Identification is progressively proposed to continuous linear systems, to some sampled-data systems, then to a static non-linear systems family. Considering Gaussian inputs, parameter estimation relationships are setup. In a sampled-data systems class, without assumption on the inputs character, setting up simple relationships between generalized transinformations leads to an original procedure in these systems order determination.

A Methodology for Structural Analysis and Partition of Dynamic Complex Systemso Application to Two Macroeconomic Systems

Abstract Man is destined to analyse and command increasingly large and complex systems. The development of sufficiently general analysis methods becomes necessary because Man has to use data and information that cannot be utilized in its entirety. Information by which the system is considered then must be structurised. The methods that are described take into account the quantitative importance of the relationships between variables and lead to the proposing of partitions into slightly coupled subsystems. The first method exposed here uses the transinformation concept in order to measure the intensity of the relationships. Some concepts of the Fuzzy Sets Theory lead to the second method. After an account of ways of building dissimilarity indexes, we propose a method of partitions similar to a classing procedure. The structural analysis and complex systems partition process mainly presents an iterative feature. A set of organigram^ and general rules is then pointed out. The proposed applications belong to the macroeconomic systems domain (study of Forrester’s world dynamics, study of the French monetary system) where we show the limits and the contributions of these methods.