Sliding-mode disturbance observers for an artificial pancreas without meal announcement. (June 2019)

• Record Type: Journal Article
• Authors: Sala-Mira, Iván; Díez, José-Luis; Ricarte, Beatriz; Bondia, Jorge
• Is Part Of: Journal of process control. Volume 78(2019)
• Page Start: 68
• Page End: 77
• Publication Date: 2019-06
• Subjects: AUC Area Under the Curve -- CF correction factor -- FA meal-announcement-free automatic system -- FOSMO first order sliding mode observer -- FP false positive -- FN false negative -- GESO Generalized Extended State Observer -- ICR insulin-to-carbohydrate ratio -- IFB Insulin Feedback -- IB main controller with Ideal Bolus -- IOB Insulin on board -- IVP Identifiable virtual patient -- MB main controller with Misestimated Bolus -- NB main controller with No Bolus -- RA meal rate of glucose appearance -- RMSE root mean squared error -- ST super-twisting -- SMRC Sliding Mode Reference Conditioning -- TDI Total Daily Insulin -- TP true positive; unannounced artificial pancreas -- sliding mode observer -- meal detection -- postprandial control; Process control -- Periodicals; Fabrication -- Contrôle -- Périodiques; Process control; Periodicals; Electronic journals; 660.281
• DOI: 10.1016/j.jprocont.2019.03.008
• Languages: English
• ISSNs: 0959-1524

Abstract

Highlights: The proposed observer enhances linear methods in estimating glucose appearance rate. Best tradeoff between sensitivity and early meal detection: 92.4%, 28.5 min. Non-significant mean glucose is found regarding hybrid system with carbs miscounting. Abstract: Carbohydrate counting is not only a burden for patients with type 1 diabetes, but estimation errors in meal announcement could also degrade the outcomes of the current hybrid closed-loop systems. Therefore, removing meal announcement is desirable. A novel control system is addressed here to face postprandial control without meal announcement. The proposed system grounds on two applications of the sliding mode observers in dealing with disturbances: first, the equivalent output technique is used to reconstruct the meal rate of glucose appearance via a first order sliding mode observer; second, a super-twisting-based residual generator is used to detect the meals. Subsequently, a bolusing algorithm uses the information of the two observers to trigger a series of boluses based on a proportional-derivative-like strategy. An in silico validation with 30 patients in a 30-day scenario reveals that the meal detector algorithm achieves a low rate of false positives per day (0.1 (0.1), mean (SD)) and a detection time of 28.5(6.2) min. Additionally, the bolusing algorithm fulfills a non-statistically different mean glucose than the hybrid counterpart with bolus misestimation (146.69 (12.20) mg/dL vs. 144.28 (11.01) mg/dL, p >0.05), without increasing hypoglycemia (0.029 (0.077) vs. 0.004 (0.014)%, p > 0, 05), although at the expense of a slightly higher time in hyperglycemia (22.51(8.72) % vs. 18.65 (7.89)%, p < 0.05 ) due to the conservative tuning of the bolusing algorithm for the sake of safety.