Adaptive Learning-Based Resilient Automation for Switched Systems Under FDI Attacks and Actuator Faults

Abstract

Automated systems operating across multiple modes increasingly rely on networked sensing and feedback, which makes them vulnerable to abrupt actuator faults and false-data-injection (FDI) attacks. This paper studies resilient estimation and control for continuous-time switched automation systems under these two coupled threats. A switching description that combines sojourn-time and sojourn-probability information is first introduced to characterize stochastic mode transitions without requiring explicit transition probabilities. To reconstruct abrupt and possibly nonsmooth actuator faults, a proportional-derivative learning observer with an adaptive learning interval is developed. The learning interval is adjusted online according to the current estimation behavior, so that fast transient reconstruction and steady-state robustness can be balanced within a unified design. To cope with stealthy FDI attacks, a two-layer resilient architecture is further established. A sensor coding layer is first used to transform the effective attack channel and provide a private encoding interface before the corrupted data enter the estimator. Based on the decoded measurements, the proposed learning observer and a fault-tolerant controller are then coordinated to estimate and compensate both attack effects and actuator faults. Sufficient conditions are derived to guarantee bounded estimation errors and closed-loop signals under average dwell-time switching. Two examples, including an RLC circuit, are provided to demonstrate the effectiveness of the proposed method and to show its potential for resilient automation applications subject to sensing and actuation anomalies. Note to Practitioners—This paper is motivated by automated systems that operate under changing modes and rely on networked sensing, such as reconfigurable industrial units, power-electronic processes, and other cyber-physical automation platforms. In these systems, two practical problems often arise at the same time: abrupt actuator faults and malicious manipulation of sensor data. The method developed in this paper addresses both problems within one framework. A sensor coding step is first used to transform the effective attack channel and provide an additional private encoding layer before the corrupted measurement enters the estimator. An adaptive learning observer is then used to reconstruct abrupt actuator faults without relying on a fixed learning window, which helps balance responsiveness and noise robustness during operation. The resulting fault estimate is used by a fault-tolerant controller to maintain closed-loop performance. The proposed method is most suitable for applications where mode changes, sampled measurements, and network security issues coexist. Before deployment, practitioners need a process model for each operating mode, an encoding/decoding mechanism shared between the sensing and control sides, and parameter tuning for the adaptive learning interval based on the expected fault variation rate and measurement noise level.