A New Fault Detection Method From Data-Based Dissipativity Theory

In this paper, a novel fault detection scheme for linear process systems using data-based dissipativity theory is developed. The data-based dissipativity is learned from the available input/output trajectories. Dissipativity of a process is an input/output property, which may not be valid when a fault occurs. In light of this fact, in the proposed approach, the dissipativity of a process is tailored such that it is fault-sensitive, i.e., no longer valid when faults occur. By adopting the storage functions and supply rates in the quadratic difference form, the dissipativity conditions are represented as quadratic functions of the input/output trajectories of the process, which captures much more detailed dynamical features compared to traditional dissipativity. These dissipativity properties are determined offline by solving an optimization problem with linear matrix inequality constraints. The online diagnosis algorithm involves checking of inequalities on input/output trajectories only, which makes the design much simpler. The proposed approach is illustrated using a case study of fault detection of a heat exchanger.