Data-driven design of fault detection and isolation systems subject to Hammerstein nonlinearity

This paper is concerned with data-driven design of fault detection and isolation (FDI) systems subject to Hammerstein nonlinearity, a static nonlinearity in the front of inputs. Specifically, the design of residual generation is then formulated as to solve a convex optimization problem by combining ideas from the over-parameterization and least squares support vector machines (LS-SVMs), and thus provides residual signals directly from process data. To solve the multiply-outputs (MOs) problem, a modified approach is proposed by means of the so-called mixed block Hankel matrices. Sufficient conditions for the existence of a parity space are established and proved. A benchmark example is given to show the effectiveness of the proposed approach.