Model based controllers for blood glucose regulation in type I diabetics

Abstract

Diabetes is a chronic disease affecting millions of people in the world. Regular insulin injection therapy is now practiced for maintaining blood glucose level within normoglycemic range (70-100 mg/dl) in Type I diabetics having insulin dependent diabetes mellitus. Controllers for automatic monitoring and regulation of blood glucose in diabetics have been investigated. In this study, several model-based controllers including the ubiquitous proportional-integral-derivative (PID) controllers are designed for specifying insulin dosage in Type I diabetics. The study employs a recently reported and detailed physiological model of a diabetic along with a meal disturbance model. The performance and robustness of designed controllers are evaluated on 577 diabetic patient models generated by considering i??40% variation in the significant parameters of the physiological model.The detailed physiological model of the diabetic is successfully implemented and validated for use in evaluating the designed controllers. An internal model controller (IMC) is designed based on a first order plus time delay (FOPTD) model approximation of the detailed physiological model of the nominal diabetic. Enhanced internal model controller (EIMC) is then developed due to its simple structure, better disturbance attenuation and uncertainty reduction. Both these controllers are assessed for their ability to track the normoglycemic set point of 81.1 mg/dl for blood glucose while rejecting meal disturbances both in the nominal patient case and 577 perturbed patient models. The results show that EIMC performs better than IMC as well as the robust Hi?Y controller (Parker et al., 2000) for blood glucose regulation in Type I diabetics.Noting that the ubiquitous PID controllers have not been tested on the detailed physiological model employed in this study, several PID controllers are designed using classical and recent tuning techniques. A secondary objective for this part of the study is to analyze the effectiveness of the recent tuning techniques for PID controllers for challenging biomedical applications such as diabetes control. Detailed results of testing the PID controllers designed on the perturbed patient models for meal disturbance rejection, show that the PID tuning by Shen (2002) is the best among the four tuning techniques tested. It is able to maintain the glucose concentration above the hypoglycemic range (hypoglycemia occurs when blood glucose concentration is less than 60 mg/dl) in 95% of all the 577 patient models considered while rejecting both single and multiple meal disturbances in a day.A nonlinear internal model controller (NIMC) using input-output linearization is developed for a Type I diabetic. Although this controller showed promising results for rejecting meal disturbances in the case of nominal patient model, spikes in the controlled variable (i.e., insulin injected) made it impossible to test NIMC for all the 577 perturbed patient models. The reason for spikes seems to be numerical errors, and further investigation is needed to confirm this.In summary, several model based controllers (namely, IMC, EIMC, PID and NIMC) are designed and evaluated for blood regulation in Type I diabetic. Among these, EIMC and PID controller tuned by Shen's technique have performed better than the robust Hi?Y controller which itself was shown to be better than the computationally-intensive model predictive controller (Parker et al., 2000). Considering this and the simplicity of EIMC and PID controller, it is concluded that these a??simplera?? controllers may be an attractive alternative over the more complex controllers for glucose regulation in Type I diabetics.