Detection and Spatial Identification of Fault for Parabolic Distributed Parameter Systems

In this paper, a novel method is developed to detect fault and identify its spatial location for a class of parabolic distributed parameter systems (DPSs) with limited sensors. The normal situation of the DPSs is first modeled under limited sensors. Then, the spatio–temporal dynamics of DPSs is decoupled under time/space separation. After the temporal coefficients are further decomposed using the independent component analysis method, the dominant temporal components are taken and then the spatial residual errors are used to form two monitoring statistics. Using the kernel density function, the confidence bounds of these two statistics (fault free) can be established as the reference signals. Unlike model-based fault detection methods that require explicit mathematical models of the processes, the proposed method is data driven and only utilizes the separable characteristic of parabolic DPSs. Experiments on a typical parabolic process and a snap curing oven are used to verify the effectiveness of the proposed method.