A.J. Niemi

Residence time distributions of variable flow processes

Abstract Tracer response functions can be interpreted and used under conditions of variable flow, if they are presented in terms of an integrated flow variable and if the flow patterns are not affected by changes of flow rate. Several flow models are analyzed and they show an agreement with the presented theory. Application of the method to arbitrary non-analytic response and distribution functions is outlined. Use of the distribution functions of presented type for control of variable flow processes is similar to the control of stationary systems. The functions are also applicable to time domain methods for control of variable flow chemical reactors with linear kinetics.

Model predictive control for grinding systems

Abstract Grinding systems are often modelled by multivariable impulse response or weighting function matrices. These kind of models are very suitable for design of model predictive controllers. Two predictive control algorithms have been developed and tested with two simulated models of grinding systems. Their design is based on decoupling of the steady state interactions. The dynamic interactions cannot be eliminated but their strengths are reduced by dynamic compensation.

On characterization of pulp and froth in cells of flotation plant

Abstract Dynamic models of pulps in industrial flotation cells are presented with emphasis on the modelling of their chemical conditioning. They are reduced to steady state models that can be combined to describe banks and networks of cells. Kinetic characterization of minerals in the plant is outlined on the basis of samples from appropriate points, their analysis and simulation between such points, in order to model the solids transfer from pulp to froth in the cells. Models of the froth layers and studies of their surface views are reviewed. Images of froths have been recorded in two plants and results of their analysis are presented. Scene analysis is considered a valuable means in acquisition of new information of processes in flotation cells.

Role of kinetics in modelling and control of flotation plants

Abstract The kinetics of industrial flotation are studied with special reference to grain size, chemical conditioning and liberation. Related models are derived for batch flotation in the laboratory and for continuous flow flotation in a plant. For practical testing of their relationships, specimens of regular mineral and samples from the output of an industrial conditioner are floated in a laboratory cell and radioactively labelled specimens of the same mineral in a plant cell. Timed samples of concentrates and tailings are collected and classified, and kinetic coefficients are evaluated in each case for different grain size and conditioning. Use of these data in theoretical models provides consistent results. The quantitative relationships of the kinetics in laboratory and plant delivered by the method have a central role in the scale-up of flotation from the laboratory to a plant cell, including that of the effects of various control inputs. The dynamic flotation models obtained are reduced to steady state and can then be used for model construction and control design of larger flotation plants.

Simulation and control of flotation circuits

An industrial flotation circuit with successive stages of flotation and recycling material is studied in preparation for digital control. After considering the steady state and dynamics of the circuit, a study of the feedback control, based on a linearized model, has been carried out using standard graphical and analytical methods. Good inherent process stability is observed. A model of the circuit with two floatable particle types, controlled by feed-forward and feedback methods, was created using a CSMP digital/analog simulator. Feed-forward gives a better compensation for feed concentration, which is considered the main disturbance, but feedback is needed for several other disturbances, non-linearities, etc. A combined method is being developed and will be used for practical computer control. At the same time values of process parameters are being determined by data collection and statistical analysis, in order to expand the model.

Variable parameter model of the continuous flow vessel

Models of continuous flow vessels with variable volume and flow are studied. It is shown that the residence distributions and weighting functions of many types of continuous flow vessels which are time-variable under such conditions, are invariant with regard to new, composite variables which are introduced. In addition to the simple, basic flow patterns, the Gaussian distribution of liquid velocity and the laminar flow are in agreement with the presented theory. The approach is extended to arbitrary, non-analytic response functions obtained by measurements. Application of dispersion models to time-variable hydrodynamic processes appears to be subject to limitations. In the case of variable volume, the basic variable of perfect mixer differs distinctly from that of other vessels. Applications and uses of the derived models of variable parameter processes are discussed.

A linear-quadratic-Gaussian control algorithm for sulphide ore grinding

Abstract The paper suggests an algorithm for the multivariable control of an industrial sulphide ore grinding plant. The algorithm is based on the linear-quadratic-Gaussian control theory with a moving time horizon. The dynamic process model with two inputs and two outputs is obtained by fitting simple low order linear time-invariant models to measured input-output pairs. Characteristic to the model are long time delays. The model is brought to the state space form, in which the time delays are described by additional state components. The inputs are replaced by input increments for the purpose of avoiding imbalances of stationary values and of assigning quadratic costs to input changes. Because of disturbances both in the process and in the measuring equipment, a Kalman filter is used for state estimation. The control algorithm was tested both by simulation runs and by implementation in the control of the plant. The test runs showed that the algorithm can be tuned rather easily, and satisfactory results were obtained. The steady-state interactions were almost completely compensated and the transient interactions were within acceptable limits.

Composition of Conflict-Free Petri Net Models for Control of Flexible Manufacturing Systems

Abstract An approach using Petri nets to modelling and control of flexible manufacturing systems (FMS) is worked out in the paper. A class of Petri nets suitable for the modelling of FMS and for the design of their control is formulated. In the Petri net models not only conflicts due to mutual exclusions of the shared system resources but conflicts of branching operations are analyzed and solved, as well. Models of FMS suitable for system performance analysis and control design are built up of a set of elementary building blocks proposed in the paper. Analysis, modelling method and control design is developed on the background of a pilot experimental FMS developed at the Helsinki University of Technology.

Image analysis and vision systems for processing plants

Material flowing in a processing plant may be visually observable, but its characterization by physical and computational means of analysis can prove difficult. Problems are encountered in practice both at optical imaging and at extraction of features that characterize the material or its stage of processing, and at related control of the process. Intelligent analysis of the vast amount of data provided by process vision systems is discussed in the paper, in the light of two continuous flow processes, i.e. the froth flotation of minerals and the wet end control of a paper machine.

Control of grinding circuits using phenomenological models

Abstract Dynamic models are obtained as combinations of phenomenological models of the grinding system components, and the tuning of multivariable controllers for new operating conditions of such systems is discussed by means of models. The total model is first reduced to the steady state and its parameters evaluated by experimental and computational techniques. By means of known dependences of the parameters on operating conditions, any new steady state, following e.g. a change in feed or grain size distribution is then reproduced by simulation with the model. Without further experimentation, the original model is next adjusted to agree with the new point of operation, and the number of its grain size classes is reduced to two, in order to obtain its dynamic responses by simulation. These are converted to frequency responses, and appropriate multivariable controllers are designed by means of Nyquist arrays.