Coding-based abnormal behavior differentiation approach for industrial systems

This paper deals with the problem of sensor faults isolation from overlapping un-stealthy attacks based on coding sensor outputs for industrial systems represented by Lipschitz affine nonlinear models. A novel structure of a coding scheme, a network of three groups of interlinked observers, and an adaptive threshold technique is developed. To detect the abnormal behaviors, the first group is developed. The second group is designed to expose the attack behaviors. To isolate sensor faults, the third group of observers is constructed by exploiting the sensitivity definition. Beyond the novelty of the proposed coding scheme, the introduced approach has a novel structure that provides more degrees of freedom to prove the stability of observers in the nonlinear coordinates. Finally, the effectiveness of this approach is verified using the simulation of benchmark continuous stirred tank reactor (CSTR).