Secure adaptive fault-tolerant control of networked T-S fuzzy glucose-insulin system against pump fault and attack under dynamic adaptive event-trigger and dynamic observer

Abstract

Nowadays, the closed-loop glucose-insulin system or artificial pancreas system is becoming an effective and popular insulin therapy for diabetic patients. However, with the integration of extra intelligent devices like communication networks and insulin infusion pumps, some practical constraints on these devices will degrade the therapy effect and even cause serious health problems. In order to improve its practice and reliability, this research is concerned with the secure adaptive fault-tolerant control of networked T-S fuzzy glucose-insulin systems in the presence of insulin pump fault and attack by using an event-triggered protocol and a state observer. First, a T-S fuzzy modeling method is employed to cope with the system nonlinear dynamics and make it easy to be analyzed by using the existing linear system methods. Second, in order to conserve the limited communication bandwidth, an improved dynamic adaptive event-triggered protocol is proposed. This scheme has two merits, the one is that a dynamic triggering threshold is constructed to regulate the releasing frequency along with the system evolution adaptively. The other is that an internal variable related to the traditional triggering condition is introduced to enlarge the event interval. Third, to overcome the state constraint issue and realize the state feedback control, a dynamic state observer by introducing an auxiliary variable related to the integration of the estimation error is designed to get more precise observed state. Fourth, to improve the system reliability and security against the insulin pump with loss of effectiveness fault and malicious attack signal, a secure adaptive fault-tolerant control strategy is constructed based on the observer state, the estimation of the fault factor obtained by an adaptive law and the approximation of the unknown attack signal by an neural network-based attack estimator. Then, resorting to Lyapunov approach, several sufficient co-design conditions of the controller and observer gains and the triggering matrix are derived in terms of linear matrix inequalities. Finally, some comparison simulations are executed to illustrate the advantages of the proposed strategy over some existing methods.