Browsing by Author "Swartz, C L E"

Now showing 1 - 4 of 4
  • Thumbnail Image
    Item
    Open Access
    The application of fuzzy control to fed-batch fermentation
    (1995) Hadley, Trevor Deon; Swartz, C L E
    Fermentation processes are highly nonlinear and subject to variability. The fermentation's states are not readily available on-line and therefore the application of closed loop control schemes have been hindered. It was decided to investigate fuzzy control as it is able to deal with systems whose operation does not easily fit into the mathematical framework of traditional control approaches such as fermentations where the systems are highly nonlinear. The fermentation of lysine is an emergent industry in South Africa and it was therefore decided to focus on this fermentation. The control of penicillin fermentation was also investigated as it closely resembles the fermentation of lysine. A review of the types of control and estimation techniques used in the literature for biosystems was done to assess state of art in biocontrol. This covered optimal control techniques, neural networks, fuzzy controllers and adaptive control techniques. The operation and properties of fuzzy controllers were investigated. A specific form of fuzzy controller, presented in the literature, which was shown to correspond to a PI controller with a nonlinear gain was discussed. The effect of the number of output sampling points was analysed and it was found that the number of output sampling points used has an effect on the output and input response. It was also found that a higher number of sampling points results in a nonlinear integral constant and a non linear gain which has more resolution. The fuzzy controller's output response equations were found to be of a PI form with a possible bias term irrespective of the number of sampling points. The fuzzy controller was shown to yield better output and input response to that of an equivalently tuned linear PI controller for a first, second and third order system because it is able to take advantage of its nonlinear form. It was also shown that it is possible to obtain less severe input action for relatively the same value of SSE (sum of squared errors) when a higher number of sampling points is used for a first order system with dead time.
  • Thumbnail Image
    Item
    Open Access
    Dynamic matrix control of milling circuits
    (1993) Mkhize, Sikhumbuzo Cazwell; Swartz, C L E
    The main aim of the study was to investigate the suitability of DMC for milling circuit control. This was conducted through simulation studies using models of two different milling circuits. A generalised software package was developed for the application and analysis of DMC. DMC was developed in the United States of America by Shell Oil Company (Cutler and Ramaker, 1979; Prett and Gillete, 1979). It falls under the class of controllers which is termed Model Predictive Control (MPC). The control algorithms falling into this category are multivariable and model-based, and as such are expected to improve control of processes which exhibit strong interactions, non-minimum phase behavior, and operate at constraints. Other control schemes falling into this category are Model Algorithmic Control (MAC) and Internal Model Control (IMC). Details of MAC are largely proprietary, while the frequency-based IMC method does not permit direct handling of constraints. Thus the focus of this project was on the DMC algorithm and its variants; Linear Dynamic Matrix Control (LDMC) and Quadratic Dynamic Matrix Control (QDMC).
  • Thumbnail Image
    Item
    Open Access
    Dynamic operability assessment : a mathematical programming approach based on Q-parametrization
    (1997) Ross, Roderick; Swartz, C L E
    The ability of a process plant to guarantee high product quality, in terms of low variability, is emerging as a defining feature when distinguishing between alternative suppliers. The extent to which this can be achieved is termed a plant's dynamic operability and is a function of both the plant design and the control system design. In the limit, however, the closedloop performance is determined by the properties inherent in the plant. This realization of the interrelationship between a plant design and its achievable closed-loop performance has motivated research toward systematic techniques for screening inherently inferior designs. Pioneering research in the early 1980's identified right-half-plane transmission zeros, time delays, input constraints and model uncertainty as factors that limit the achievable closedloop performance of a process. Quantifying the performance-limiting effect of combinations of these factors has proven to be a challenging problem, as reflected in the literature. It is the aim of this thesis to develop a systematic procedure for dynamic operability assessment in the presence of combinations of performance-limiting factors. The approach adopted in this thesis is based on the Q-parametrization of stabilizing linear feedback controllers and involves posing dynamic operability assessment as a mathematical programming problet? In the proposed formulation, a convex objective function, reflecting a measure of closed-loop performance, is optimized over all stable Q, subject. to a set of constraints on the closed-loop behavior, which for many specifications of interest is convex. A discrete-time formulation is chosen so as to allow for the convenient hand.ling of time delays and time-domain constraints. An important feature of the approach is that, due to the convexity, global optimality is guaranteed. Furthermore, the fact that Q parametrizes all stabilizing linear feedback controllers implies that the performance at the optimum represents the best possible performance for any such controller. The results are thus not biased by controller type or tuning, apart from the requirement that the controller be linear.
  • Thumbnail Image
    Item
    Open Access
    Extensions to the data reconciliation procedure
    (1996) Seager, Mark Thomas; Swartz, C L E
    Data reconciliation is a method of improving the quality of data obtained from automated measurements in chemical plants. All measuring instruments are subject to error. These measurement errors degrade the quality of the data, resulting in inconsistencies in the material and energy balance calculations. Since important decisions are based on the measurements it is essential that the most accurate data possible, be presented. Data reconciliation attempts to minimize these measurement errors by fitting all the measurements to a least-squares model, constrained by the material and energy balance equations. The resulting set of reconciled measurements do not cause any inconsistencies in the balance equations and contain minimum measurement error. Two types of measurement error can occur; random noise and gross errors. If gross errors exist in the measurements they must be identified and removed before data reconciliation is applied to the system. The presence of gross errors invalidates the statistical basis of data reconciliation and corrupts the results obtained. Gross error detection is traditionally performed using statistical tests coupled with serial elimination search algorithms. The statistical 'tests are based on either the measurement adjustment performed by data reconciliation or the balance equations' residuals. A by-product of data reconciliation, obtained with very little additional effort, is the classification of the system variables. Unmeasured variables may be classified as either observable or unobservable. An unmeasured variable is said to be unobservable if a feasible change in its value is possible without being detected by the measurement instruments. Unmeasured variables which are not unobservable are observable. Measured variables may be classified as either redundant, nonredundant or having a specified degree of redundancy. Nonredundant variables are those which upon deletion of the corresponding measurements, become unobservable. The remaining measured variables are redundant. Measured variables with a degree of redundancy equal to one, are redundant variables that retain their redundancy in the event of a failure in any one of the remaining measurement instruments.