Special Issue: Adaptive Methods for Resilient Control Systems

Gang Tao, Tansel Yucelen
2021 International Journal of Adaptive Control and Signal Processing  
Resilience literally is the ability to readily recover from illness, depression, or adversity. Control system resilience is the system's ability to accommodate its faults, disturbances, and uncertainties, to recover its desired performance. Resilient control is to increase control system resilience for performance guarantees in the presence of uncertain faults (such as actuator and sensor failures and structure damage) and uncertain disturbances. Desired system performance includes asymptotic
more » ... acking and optimality, in addition to system stability, and their robustness. Resilient control systems need desired resilience which needs to be provided by effective control techniques. Thus, a main feature of resilient control systems is that they can recover desired system performance in the presence of uncertain faults and disturbances, which is a crucial feature needed for performance-critical applications, and for resilient control design. Research in resilient control systems has been developing fast in recent years, reflected by increased activities in either general control conferences or special resilient control systems conferences, and by their resulted publications. New resilient control design methods and techniques are needed for emerging performance-critical applications. Contributing to the foundations of resilient control system techniques, fault detection, and fault-tolerant control has made significant progresses, and so does adaptive control which has its unique capabilities to accommodate system parametric, structural, and environmental uncertainties caused by payload variations or system aging, component failures, and external disturbances. For resilient control systems, fault-tolerant control and adaptive control face new challenges: how to effectively estimate and compensate system uncertainties caused by faults and disturbances. Systematic study of adaptive methods for effective handling of uncertain system faults is an open area of research with the main focus to the guarantee of desired system performance, the most important feature of resilient control. The goal of this special issue is to show the state-of-the-art in recent developments of advanced control methods using adaptive control and fault-tolerant control techniques to deal with uncertain system faults, for the recovery of desired control system performance. This special issue contain six papers that cover both theory, techniques, and applications of resilient control systems. The paper "Robust nonlinear adaptation algorithms for multi-task prediction networks" by Abulikemu Abuduweili and Changliu Liu studies robust online algorithms to adapt nonlinear prediction models, which can lead to more accurate predictions that facilitate subsequent control decisions. There are multiple ways for the prediction results to enter the control loop to make the system more resilient. In a safety critical situation, online adaptation can reduce the uncertainty of the prediction model, hence reducing the safety margin in the safe control law. During human-robot collaboration, online adaptation can improve the accuracy of the human intention prediction, hence allowing the robot to better collaborate with the human by choosing actions that respond to human's needs. Nonetheless, this paper focused solely on adaptable prediction. The authors introduced a robust online adaptation algorithm MEKF-MA-ME by combining exponential moving average and multi-epoch update with the modified EKF algorithm. The proposed methods can effectively adapt complex models encoded in neural networks, which has been validated through numerical studies. The second paper is "An asymptotic decoupling approach for adaptive control with unmeasurable coupled dynamics" by K. Merve Dogan, Tansel Yucelen, and Jonathan A. Muse. Adaptive control methods have the capability to provide resiliency against operational anomalies such as disturbances and system uncertainties. However, their stability can be challenged by coupled dynamics as in, for example, airplanes with high aspect ratio wings, slung load systems, cooperative robot manipulators, and actuated-unactuated dynamical systems. This paper proposes a decoupling approach for model reference adaptive control algorithms. The key feature of the proposed framework is that it guarantees asymptotic convergence between the trajectories of an uncertain dynamical system and a given reference model without relying on any measurements from the coupled dynamics under a tight sufficient stability condition. The paper also provides a generalization to address the uncertainty in the control effectiveness matrix, where the resulting sufficient stability condition in this case relies on linear matrix inequalities. 312
doi:10.1002/acs.3220 fatcat:cg5ok4xgrbagro3igadwth3qfu