Search

We propose a Kalman filter-based observer utilizing noisy remote compartment insulin measurements to estimate plasma insulin concentration. The design considers plant-model mismatch, sensor noise, as well as both uniform sampling intervals, mimicking infrequent continuous measurements, and non-uniform sampling intervals, mimicking infrequent on-demand measurements. The performance of the observer is demonstrated on ten in-silico subjects from the UVA/Padova simulator using real-life scenarios, including variability in sensor noise and variability in insulin pharmacokinetics. The proposed observer provides insight into the future use of insulin measurements for diabetes management.

Artificial pancreas (AP) systems are designed to automate glucose management for patients with type 1 diabetes. In this work, we propose an adaptive disturbance rejection control approach for AP systems to achieve safe and effective glucose regulation. The controller is built within the framework of active disturbance rejection control, but incorporates safety operation constraints, and glucose- and velocity-dependent parameter adaptation modules for the key control parameters. In silico performance comparison between the proposed controller and an adaptive zone model predictive controller (MPC) (Shi, Dassau, and Doyle III, 2019a) is conducted on the 10-adult cohort of the FDA-accepted UVA/Padova T1DM simulator. For both announced and unannounced meals, the controller achieves comparable glucose regulation performance in terms of mean glucose (134.9 mg/dL vs. 135.4 mg/dL, p

A novel Model Predictive Control (MPC) law for the closed-loop operation of an Artificial Pancreas (AP) to treat type 1 diabetes is proposed. The contribution of this paper is to simultaneously enhance both the safety and performance of an AP, by reducing the incidence of controller-induced hypoglycemia, and by promoting assertive hyperglycemia correction. This is achieved by integrating two MPC features separately introduced by the authors previously to independently improve the control performance with respect to these two coupled issues. Velocity-weighting MPC reduces the occurrence of controller-induced hypoglycemia. Velocity-penalty MPC yields more effective hyperglycemia correction. Benefits of the proposed MPC law over the MPC strategy deployed in the authors’ previous clinical trial campaign are demonstrated via a comprehensive in-silico analysis. The proposed MPC law was deployed in four distinct US Food & Drug Administration approved clinical trial campaigns, the most extensive of which involved 29 subjects each spending three months in closed-loop. The paper includes implementation details, an explanation of the state-dependent cost functions required for velocity-weighting and penalties, a discussion of the resulting nonlinear optimization problem, a description of the four clinical trial campaigns, and control-related trial highlights.

We address an optimal experimental design (OED) procedure for the online selection of type-1-diabetes (T1D) mellitus glucose metabolistic models. A fully observable reduced-order nonlinear dynamic model is presented and subsequently parameterised for Gottingen Minipigs and patients, that were both subject to an automatic insulin delivery. A bank of continuous–discrete unscented Kalman filters (CDUKF) is designed and parameterised for Gottingen Minipigs and patients. Based on this filter bank of CDUKF, a novel online OED design procedure is developed, that is used to identify the correct parameter set out of several available sets for measured blood glucose concentrations. The procedure utilises forward model simulations to calculate optimal system inputs. This leads to the identification of the correct parameter set under arbitrary conditions. Results are presented for both subgroups.

Automated insulin delivery (AID) systems have proven safe and effective in improving glycemic outcomes in individuals with type 1 diabetes (T1D). Clinical evaluation of this technology has progressed to large randomized, controlled outpatient studies and recent commercial approval of AID systems for children and adults. However, several challenges remain in improving these systems for different subpopulations (e.g., young children, athletes, pregnant women, seniors and those with hypoglycemia unawareness). In this review, we highlight the requirements and challenges in AID design for selected subpopulations, and discuss current advances from recent clinical studies.

Clinical trials have demonstrated that zone model predictive control is an effective closed-loop blood glucose regulation method for people with type 1 diabetes (T1D). This paper presents a universal model-free optimization method to seek an optimal zone for T1D patients individually. A clinical glycemic risk index named relative regularized glycemic penalty index (rrGPI) is used as the cost function. The proposed method is based on extremum seeking control that uses only the rrGPI index, calculated from measurements by a continuous glucose monitor, to update a controller’s blood glucose target zone’s upper bound and lower bound simultaneously. The method proposed uses a decaying feedback gain and a vanishing dither signal to improve the extremum seeking controller’s robustness against various uncertainties. In silico trials suggest that the proposed method is able to converge to the personalized optimal zone in less than a week of adaptation. In a 30-day in silico trial, the time spent in the range [70,180] mg/dL is increased by about 3% and 2% for unannounced 60 gCHO (grams of carbohydrates) and 90 gCHO meals, respectively, compared to the zone [80,140] mg/dL employed in the authors’ current zone controller.

A novel Model Predictive Control (MPC) strategy for an artificial pancreas to treat people with type 1 diabetes mellitus is proposed and tested in-silico. The proposed controller has been designed to handle a large spectrum of meal sizes more effectively than the authors’ current strategy Hyperglycemia due to small meals is treated more assertively leading to increased time in euglycemia, whereas hyperglycemia due to large meals is treated more cautiously resulting in a markedly reduced risk of controller-induced hypoglycemia. This is achieved by replacing the purely quadratic cost function employed within the MPC optimization by a mixed linear-quadratic cost function. This simple modification results in significant leeway to design a controller that responds more effectively and safely to the dietary habits of the AP user, while maintaining the quadratic program structure of the resulting optimization problem. The proposed MPC strategy’s benefits are demonstrated by in-silico analysis using the Universities of Virginia/Padova US Food & Drug Administration accepted metabolic simulator.

People with type 1 diabetes are at risk for early- and late-onset hypoglycemia following exercise. Reducing this risk may be possible with strategic modifications in carbohydrate intake and insulin use. We examined the exercise preparations and management techniques used by individuals with type 1 diabetes before and after physical activity and sought to determine whether use of differing diabetes technologies affects these health-related behaviours.We studied 502 adults from the Type 1 Diabetes Exchange's online patient community, Glu, who had completed an online survey focused on diabetes self-management and exercise.Many respondents reported increasing carbohydrate intake before (79%) and after (66%) exercise as well as decreasing their meal boluses before (53%) and after (46%) exercise. Most reported adhering to a target glucose level before starting exercise (77%). Despite these accommodations, the majority reported low blood glucose (BG) levels after exercise (70%). The majority of users of both insulin pump therapy (CSII) and continuous glucose monitoring (CGM) (Combined) reported reducing basal insulin around exercise (55%), with fewer participants adjusting basal insulin when using other devices (SMBG only = 20%; CGM = 34%; CSII = 42%; p0.001). However, CSII and Combined users reported that exercise makes their BG levels harder to control (p0.05) and makes them feel less able to predict their BG levels while exercising (p0.001); they show agreement that fear of low BG levels keeps them from exercising (p0.01).These findings highlight the need for exercise-management strategies tailored to individuals' overall diabetes management, for despite making exercise-specific adjustments for care, many people with type 1 diabetes still report significant difficulties with BG control when it comes to exercise.

A novel Model Predictive Control (MPC) law for an Artificial Pancreas (AP) to automatically deliver insulin to people with type 1 diabetes is proposed. The MPC law is an enhancement of the authors’ zone-MPC approach that has successfully been trialled in-clinic, and targets the safe outpatient deployment of an AP. The MPC law controls blood-glucose levels to a diurnally time-dependent zone, and enforces diurnal, hard input constraints. The main algorithmic novelty is the use of asymmetric input costs in the MPC problem’s objective function. This improves safety by facilitating the independent design of the controller’s responses to hyperglycemia and hypoglycemia. The proposed controller performs predictive pump-suspension in the face of impending hypoglycemia, and subsequent predictive pump-resumption, based only on clinical needs and feedback. The proposed MPC strategy’s benefits are demonstrated by in-silico studies as well as highlights from a US Food and Drug Administration approved clinical trial in which 32 subjects each completed two 25 hour closed-loop sessions employing the proposed MPC law.

The properties of an objective function are fundamental to the functionality of a model predictive controller (MPC). In automated glucose control, avoidance of hypo- and hyperglycemia introduces significant challenges in the design of this cost function. We present a new formulation of the cost function for an MPC based on clinical requirements, and validate the algorithm under in silico and advisory mode assessments. The proposed formulation exhibits significant improvements in avoiding hypoglycemia, compared to clinically validated controllers, and can also mitigate hyperglycemia, across a wide range of in silico scenarios as well as glucose responses seen in actual clinical settings.

Here we describe the development of a dual electrochemical immunosensor microchip for simultaneous detection of insulin (I) and cortisol (C) biomarkers that can enhance the ability to improve glucose regulation using automated insulin delivery. The successful realization of the simultaneous I and C measurements has been realized by integrating different enzymatically-tagged competitive and sandwich immunoassay formats on a single chip platform. The insulin detection is based on a peroxidase (HRP)-labeled sandwich assay whereas the cortisol detection relies on an alkaline phosphatase (ALP)-labeled competitive immunoassay. The attractive analytical performance of the dual marker immunosensor, with no apparent cross-talk, was achieved through systematic optimization of the incubation and amperometric detection of the different captured enzyme tags. Evaluation of dual biosensor chip in untreated serum samples indicated favorable simultaneous detection of picomolar (pM) insulin and nanomolar (nM) cortisol concentrations in a single microliter sample droplet within less than 25min. The new dual immunosensor chip offers considerable promise for frequent decentralized testing of I and C towards a tighter glycemic control and improved management of diabetes.

The accuracy of continuous glucose monitors (CGMs) plays a critical role in glucose management and the development of artificial pancreas systems. Current CGM performance can be improved by a run-to-run (R2R) strategy based on continuous wear that personalizes sensor calibration parameters using data from previous weeks' use. The proposed strategy determines personalized calibration curve parameter sets (calibration curve slope and intercept, sensitivity drift curve, mean reference error) by tapping into data that is readily available at the end of each week after weekly new sensor reinsertions, minimizing a cost function that penalizes the deviation between expected and actual blood glucose (BG) values. The algorithm was evaluated on 10 in silico subjects within the UVA/Padova metabolic simulator. The enhanced CGM's BG tracking was significantly improved over the standard CGM, decreasing the summed square error by over 20% in the second week and converging to 50% improvement by the 6 th week. The R2R algorithm enhances glucose accuracy week by week based on continuous wear of the device.

Prandial glucose regulation is a major challenge for the artificial pancreas using subcutaneous insulin (without a feedforward bolus) due to insulin's slow absorption-peak (50-60 min). Intraperitoneal insulin, with a fast absorption peak (20-25 min), has been suggested as an alternative to address these limitations. An artificial pancreas using intraperitoneal insulin was designed and evaluated on 100 in silico subjects and compared with two designs using subcutaneous insulin with and without a feedforward bolus, following the three meal (40-70 g-carbohydrates) evaluation protocol. The design using intraperitoneal insulin resulted in a significantly lower postprandial blood glucose peak (38 mg/dL) and longer time in the clinically accepted region (13%) compared to the design using subcutaneous insulin without a feedforward bolus and comparable results to a subcutaneous feedforward bolus design. This superior regulation with minimal user interaction may improve the quality of life for people with type 1 diabetes mellitus.

A multi-parametric model predictive control (mpMPC) algorithm for subcutaneous insulin delivery for individuals with type 1 diabetes mellitus (T1DM) that is computationally efficient, robust to variations in insulin sensitivity, and involves minimal burden for the user is proposed. System identification was achieved through impulse response tests feasible for ambulatory conditions on the UVa/Padova simulator adult subjects with T1DM. An alternative means of system identification using readily available clinical parameters was also investigated. A safety constraint was included explicitly in the algorithm formulation using clinical parameters typical of those available to an attending physician. Closed-loop simulations were carried out with daily consumption of 200 g carbohydrate. Controller robustness was assessed by subject/model mismatch scenarios addressing daily, simultaneous variation in insulin sensitivity and meal size with the addition of Gaussian white noise with a standard deviation of 10%. A second-order-plus-time-delay transfer function model fit the validation data with a mean (coefficient of variation) root-mean-square-error (RMSE) of 26 mg/dL (19%) for a 3 h prediction horizon. The resulting control law maintained a low risk Low Blood Glucose Index without any information about carbohydrate consumption for 90% of the subjects. Low-order linear models with clinically meaningful parameters thus provided sufficient information for a model predictive control algorithm to control glycemia. The use of clinical knowledge as a safety constraint can reduce hypoglycemic events, and this same knowledge can further improve glycemic control when used explicitly as the controller model. The resulting mpMPC algorithm was sufficiently compact to be implemented on a simple electronic device.

For patients in intensive care units (ICUs), control of blood glucose level is an important factor in reducing serious complications and mortality. Standard protocols for glucose control in ICUs have been based on infrequent glucose measurements, look-up tables to determine the appropriate insulin infusion rates, and bedside administration of the insulin infusion by ICU staff. In this paper a new automatic control strategy is proposed based on frequent glucose measurements and a self-tuning control technique. During a short initial time period when manual glucose control is performed using a standard protocol, a simple dynamic model of the glucose–insulin system is identified in real time using recursive least squares. Then an adaptive PID controller is tuned, based on the model parameters, and the controller is turned on. A simulation study based on detailed physiological models of the glucose–insulin dynamics demonstrates that the proposed control strategy performs better than standard protocols for insulin infusion.

An artificial pancreatic β-cell based on multipartite zone model predictive control (Multi-Zone-MPC) is evaluated on the UVa FDA-accepted metabolic simulator. Multi-Zone-MPC provides different tunings for the MPC weights based on four regions of glycemia: hypoglycemia, normoglycemia, elevated glycemia, and hyperglycemia. Defining these four zones provides richer control tunings that result in safe and effective control. The controller predictions are based on an average ARX-model that is developed using data collected from a meal response of ten different in silico adult subjects. A comparison of Multi-Zone-MPC and standard Zone-MPC is conducted on 100 in silico adult subjects following a one meal scenario of 75g. Multi-Zone-MPC outperform Zone-MPC with low blood glucose index (LBGI) of 0.1 versus 0.5, respectively without any significant change of high blood glucose index (HBGI). The use of four control zones in the Multi-Zone-MPC is successfully demonstrated to be a better control strategy that reduces the postprandial peaks and at the same time avoids any hypoglycemia due to efficient insulin administration, hence, providing effective and safe glucose regulation to individuals with T1DM.

Accurate prediction of future glucose concentration for type 1 diabetes mellitus is needed to improve glycemic control, which can produce early and proactive glycemia management before glucose concentrations drift to undesirable levels. This paper assesses the feasibility of data-driven latent variable (LV) based statistical analysis methods to characterize the glycemic variability and serve as the forecasting engine. It illustrates the capability of LV-based multivariate statistical analysis to (1) model the correlation relationship among glucose and exogenous inputs, (2) make reliable predictions of future glucose concentrations, and (3) capture the underlying dynamics of prediction errors for on-line reliability evaluation. The new approach provides an automatic predictive monitoring strategy for glucose management in type 1 diabetes. Its feasibility is successfully assessed using data collected from the Food and Drug Administration (FDA)-accepted University of Virginia (UVA)\University of Padova metabolic simulator.

Recent technological progress in insulin pumps and continuous glucose monitors (CGM) are enabling development of an artificial β-cell that will allow superior glycemic control for patients with type 1 diabetes mellitus (T1DM). A control algorithm that is implemented in such system will need to regulate basal insulin as well as to reject unmeasured disturbances, such as meals. A traditional approach is to combine feed-forward control as a means to overcome meal disturbances, where the user informs the controller on a meal and estimates the size of that together with PID control or Model Predictive Control (MPC) to address the regulation problem. This approach fails with T1DM adolescents and children because they often forget to give a pre-meal bolus and are poor at estimating meal sizes. A novel approach to overcome this problem is suggested in this paper by the combination of a meal library and a meal detection algorithm in the framework of Model Predictive Control (MPC). In this work, the challenging problem of an unannounced mixed meal is being addressed using this novel combination.

In the past few years, rapid thermal processing (RTP) has gained acceptance as mainstream technology for semiconductor manufacturing. This single wafer approach allows for faster wafer processing and better control of process parameters on the wafer. However, as feature sizes become smaller, and wafer uniformity demands become more stringent, there is an increased demand from rapid thermal (RT) equipment manufacturers to improve control, uniformity and repeatability of processes on wafers. In RT processes, the main control problem is that of temperature regulation, which is complicated due to the high non-linearity of the heating process, process parameters that often change significantly during and between the processing of each wafer, and difficulties in measuring temperature and edge effects. This paper summarizes work carried out in cooperation with Steag CVD Systems, in which algorithms for steady state and dynamic temperature uniformity were developed. The steady-state algorithm involves the reverse engineering of the required power distribution, given a history of past distributions and the resulting temperature profile. The algorithm for dynamic temperature uniformity involves the development of a first-principles model of the RTP chamber and wafer, its calibration using experimental data, and the use of the model to develop a controller.

A methodology is presented to define a set of operating conditions to produce a desired product, given a database of historical operating conditions and the product quality that they produced. This approach relies on the generation of a reliable model that can be used to predict the quality variables (the Y block) from the decision variables (the X block). Genetic programming (GP) is used to automatically generate accurate nonlinear models relating latent vectors for the X and Y blocks. The GP has the capability to carry out simultaneous optimization of model relationship structures and parameters, as well as to identify the most important basis functions. Once an adequate model is generated, it is used to predict the required process conditions to meet the new quality target by reverse mapping.