Home

Short Bio

Mission

Expertise

Research Interests

Teaching

NSERC Type Bio

Professional Links

Personal Links

The classic way to control a system, in a model based framework, is to obtain a model of the system and then to use it for the design of a controller, process that can be executed online by an indirect adaptive controller. This work is devoted to describe the particular structure of such a controller, and also to show how Laguerre orthonormal basis functions can be extended to multivariate systems and used to produce a valid linear process model. Further this model can be used in a constrained multivariable predictive controller, at each time step, to produce a control move that accounts for a good reference tracking in the presence of disturbances and a reduced actuator movement within given constraints. Process industries need a multivariable predictive controller that is low cost, easy to setup and maintains an adaptive behavior, which accounts for plant non-linearities as well as potential mismodeling. Therefore, to answer this request a multi-input/multi-output (MIMO) adaptive model based predictive controller (MBPC) has been researched. This controller is now a real time prototype implemented on a Windows-NT platform to be used in plant trials.

Automatic feedback control of drug administration is ideally suited to anesthetized surgical patients as well as the critically ill since drugs with rapid onset times, short duration of action and small margins of safety are frequently used. The application of an adaptive predictive process control technology to drug administration will assist physicians in avoiding both over-dosages and under-dosages in their patients. Because over-dosing of neuromuscular blocking drugs (NMBs), used during anesthesia, can be safely managed, we have selected the NMBs project as a first step in applying adaptive control. An adaptive controller would avoid over-dosing and under-dosing by compensating for non-linear drug responses as well as inter- and intra-patient variation. This project will then allow us to proceed rapidly to human clinical trials.

Anesthesia is a reversible pharmacological state where safe desired levels of the patient’s muscular relaxation, analgesia and hypnosis are controlled and guaranteed by means of monitored administration of specific drugs. A major challenge for the anesthetist is the proper maintenance of the desired patient state by adjustment of the rate of delivery of anesthetics while monitoring the patient’s vital signs. Although not very frequent, awareness during anesthesia can be very traumatic for patients. The lack of precise methods for monitoring a patient’s state of unconsciousness needs to be addressed. Because in this role, the anesthetist is essentially behaving like a feedback controller, it is natural to investigate the application of control engineering to automate the delivery of anesthetics. While such a system would not replace the anesthetist, it would allow the latter to set the desired effect, leaving to the control system the task to adjust the drug delivery rate. It would free the anesthetist for higher level tasks. Constant monitoring and control would also allow the use of shorter-acting drugs, thus speeding patient recovery. There are however several obstacles that need to be removed. First, better methods for assessing the patient‘s state must be developed, particularly for hypnosis and analgesia. Second, because there is significant variability in individual response to anesthetics, it is crucial for any anesthesia control system to provide guarantees of performance and robustness for the expected variability. Finally, any such system must be easy to setup, to use and override. Although our ultimate goal is the full multivariable control of anesthesia , the proposed work addresses only some of the challenges above, particularly those related to hypnosis. First, a better understanding of the brain mechanisms of anesthetic-induced loss of consciousness is necessary. The development of better, and predictive methods for measuring awareness in an anesthetized patient will be undertaken. Our control methodology will first be demonstrated on neuromuscular blockade, as it is the safest loop to consider. A hypnosis control system that is robust to the intra- and inter-patient variability as well as to the presence of measurement artifacts will then be developed. Finally, using methods inspired from the aerospace industry, we intend to develop a methodology for certifying an anesthesia control system with guaranteed performance and robustness.

One of the main challenges in the process of improving the patient quality of life through better health care is to ensure the dissemination of medical evidence and further the prescription of medical therapies both for practicing health care professionals and patients. All parties involved in delivering health care are currently looking at ways to have better guidance for the diagnose process, based on sufficient scientific evidence obtained through randomized clinical trials. This in turn offers the promise of reduced costs. Through adequate knowledge representation of the medical information, a reasoning process that is continuously updated learning the scientific evidence, could provide the required support for informed and tractable medical diagnosis. A dramatic increase in medical knowledge coupled with rapid advances in medical information technologies are among the significant changes in the way medicine is developing today. This has led us to pursue new directions of research and development in the study of human-computer interaction in the context of health care practice and training, within laboratory and work settings. In conclusion there is a need for a system, which helps with time critical, accurate and consistent decision-making in the medical informational world. Also, to deal with the increasing amounts of data provided, a tool is required to enhance the ability of the physician to access the right data, knowledge and expertise at the right time.

Under the auspices of The Division of Control Systems at the Department of Pharmacology & Therapeutics at the University of British Columbia a multidisciplinary team composed of mathematicians, engineers, computer scientists, an educational researcher, medical clinical specialists, and pharmacologists representing boh academia and industry are pursuing modeling research for pharmaceutical development. The initial project is to develop and test a simulation model for computer controlled drug delivery systems in response to measurable outcomes of the drug action. Animal experiments have been designed to provide the information necessary for the mathematical modeling used in the infusion control system. The models will be translated into a real time software simulator. The validity of the approach will be shown by further animal testing, and compared to existing pharmacokinetic-pharmacodynamic models. The information relating the infusion rate and patient characteristics to the effect (neuromuscular blockade) will be recorded. This database will be available for further pharmacokinetic modeling. Projects in the immediate future include modeling of inhalational anaesthetic drugs, insulin during cardiopulmonary bypass, and short acting antiarrhythmic drugs in patients with implanted defibrillators. Our claim is that mathematical modeling and its expression as computer software will profoundly influence the type and characteristics not only of drug administration, but the of nature of the infused drugs themselves.

For the class of systems characterized by a large number of inputs and outputs, such as the cross direction control of a paper machine or an aerospace smart structure, we require a reduced computational time to produce the controller parameters and further the command. Our solution to this problem is a direct adaptive predictive controller, which operates in the Laguerre shift operator domain and replaces the system identification step together with the calculation of the predictive controller parameters. This task is achieved by: 1) a least squares solution, 2) two simple linear algebra operations (QR and SVD decomposition) of a matrix constructed from input and output measurements of the unknown system, and 3) a quadratic program optimization or another least squares calculation designed to produce the plant command. The modeling step is performed in a subspace identification fashion. The resulting algorithm provides major computational savings due to the reduced dimension of the system matrices together with the absence of a specific state space model. Simulation results are presented for a better evaluation of the proposed methodology.