SUPERVISORY GENERALIZED PREDICTIVE CONTROL AND FAULT-DETECTION FOR MULTIVARIABLE ANESTHESIA

Single-input/single-output generalised predictive control (GPC) has been applied successfully to muscle-relaxant anaesthesia in both simulations and clinical trials. This work was extended later to the multivariable case involving simultaneous control of muscle relaxation through EMG measurements, and unconciousness using blood-pressure monitoring in a series of simulation studies. To achieve even higher degrees of automation, a fault-detection isolation and accommodation layer has been superimposed on top of the multivariable GPC strategy. The performance of the overall supervisory control scheme has been evaluated in a series of simulation runs on a work station. The results are encouraging in the sense that the algorithm has proved very effective in detecting, diagnosing and eventually compensating for various simulated faults normally encountered in the operating theatre. The experiments used two versions of GPC: the multivariable GPC (using a P-canonical form for the process model) and GPC with feedforward, and were conducted in noise-free and noise-corrupted situations since operating theatres can be electrically dirty environments. The study forms the basis for an experiment protocol which will be followed in forthcoming clinical trials.