Fault tolerant control concepts applied to anesthesia

The first application of automatic control to a physiological variable during general anesthesia was reported in 1949. Since then a lot of research has been devoted to this topic. Still, today there is no commercial anesthesia system available with the patient in a feedback loop. One reason for this is that research has mainly been focused on controller design. Important aspects of clinical practice such as the treatment of measurement artifacts and faults have only marginally been addressed. In this thesis exactly such problems which are usually tacitly neglected but which are of great relevance are addressed. We are referring to these aspects as supervisory functions. Before addressing these supervisor functions some foundations are established. This involves in the first place a hard- and software research platform that allows to implement and test control algorithms as well as the supervisory functions in the operating theater. It is our view that fault handling must start with the design of a system. Special attention is therefore paid to the selection of the platform components and the software design. The software structure is built with special emphasis on extendibility. A second foundational building block is a mathematical model which describes the dynamic relationship between vaporizer concentrations and surgical stimulations on the input side and the inspiratory and endtidal Isoflurane concentration as well as mean arterial pressure (MAP) on the output side. A thorough review of the physiological background is followed by a step by step development of the model equations. Modeling is finalized with the identification of the system parameters and validation experiments. Since the design of control algorithms is not the main focus, controllers are taken form a thesis by Marco Derighetti. They are refined for broader applicability where necessary. The result of extensive clinical validations of an observer based state feedback (OBSF) controller for the endtidal Isoflurane concentration and an OBSF with endtidal overrides for MAP are presented. In view of the limited ability of this MAP controller to compensate heavy disturbances a control scheme based on disturbance anticipation is suggested. For the supervisor functions first a structure is suggested which allows to allocate all functions postulated in literature. A selection of these is developed in more detail. First, an elegant strategy to handle measurement artifacts in the framework of OBSF controller is proposed. It is based on a nonlinear modification of the output injection gain. The stability of this algorithm is proved and recordings of several successfully suppressed artifacts are shown. A large part of the thesis is then dedicated to fault tolerant control (FTC). Following a sequential design procedure developed by Prof. M. Blanke a strategy for handling the most critical faults in the system is obtained. In this context the concept of recoverability for linear time invariant (LTI) systems is developed. This concept is utilized to analyze to what degree the functionality of a faulty system may be recovered in case of a fault. Finally, a man machine interface (MMI) is designed. Aspects of MMI are important in this context since a well designed MMI increases operational safety. The main contribution of this thesis is to show that mathematical process models are a useful tool in dealing with faults also in a biomedical environment.