Search

An algorithm is introduced to estimate the zero-shift covariance of additive input and output noise in a situation when the process inputs and outputs are measured and the input-output model of the process is known. It is assumed tat the noise to be estimated is uncorrelated in time. The estimates, that are optimal in a least-square sense, are based upon a measurement error vector easy to compute from the actual measurements. Though the algorithm is rather simple, it is prone to numerical difficulties and, if the process parameters are not exactly known or the noises are not really white, to significant estimation errors.

A supervisory scheme is presented, aimed at stabilizing adaptive controllers. The scheme works on the basis of heuristic (expert) rules and includes a monitor, a detuner and a structure identifier. The monitor detects onsetting instability based on the analysis of the amplitude and trend of the error signal. Upon detecting instability, the contoller is detuned, to a safer algorithm, recursively until stability is restored. Simultaneously, new data is gathered and the plant structure is re-identified. Based on the new structure, a new controller is designed and normal adaptation is finally restored.

Automation may facilitate drug infusion by providing a constant or preprogrammed pump rate, by giving physicians advice about the appropriate drug dose to be manually injected, or by the implementation of true closed loop control. The infusion of insulin in diabetic patients and the administration of vasoactive drugs to manipulate arterial blood pressure have been the most common applications of closed loop control. Adaptive control of drug delivery is under development, but the routine clinical use of such controllers depends on the imlementation of appropriate supervisory algorithms. The combined use of analytical techniques and knowledge based rules is likely to be effective in the further development of adaptive drug delivery systems

A stationary estimator is presented that provides estimates both for the outputs and the inputs of linear time-invariant systems. The estimates satisfy the input-output equations and are optimal in a weighted minimum variance sense. It is shown that the true optimum problem is too complicated to allow a general closed-form solution and a sub-optimal approach is suggested that leads to a simple computational algorithm.

A hierarchical real-time computer system has been developed for laboratory automation purposes. It consists of microcomputers for data acquisition and control and a large mini for data processing. The control micros are connected to each other, the mini and an operator’s terminal through a communication microprocessor. This latter is designed in accordance with the APL concept of Shared Variable Processor, taking into account the reliability and throughput considerations steming from the real-time and distributed nature of the system.

Techniques to detect and isolate failures in complex technological systems, such as sensor biases, leaks and equipment deterioration, are discussed. The surveyed methods are based on analytical redundancy afforded by a mathematical model of the system. The main components of such techniques are residual generation using the model, signature generation via statistical testing and signature analysis. Model-structural conditions for failure isolability are introduced together with transformation methods to attain them. Sensitivity and robustness considerations are presented and an artificial intelligence approach proposed based on model redundancy.

The last decade has been marked by a rapid propagation of computer applications to industrial process control. However, only a small part of the achievements of modern control theory have found general utilization in these systems. The aim of this paper is to give a brief overview of the digital techniques of automatic control that, in addition to being theoretically sound, serve real practical needs and are simple enough to program and run effectively. The specific areas covered are modelling, identification, filtering and control.

Five identification methods have been developed for application in closed loop, two of which utilize some special property of the inferential control scheme. All five have been computer tested and compared to the open-loop prediction-error least square approach. The two closed-loop least-square methods, though more complicated, did not yield any real improvement over the accuracy of the reference technique. The open-loop model-error least-square method improved accuracy but showed convergence problems. Finally, the two corrective methods proved far superior in accuracy, at the expense of slightly increased numerical complexity.