Effective fault detection and isolation using bond graph-based domain decomposition

The problem of fault detection and isolation in complex chemical/biochemical plants can be effectively addressed by a hierarchical strategy involving successive narrowing of the search space of potential faults. A bond graph network is one means of achieving a decomposition based on a separation of the physical domains such as mechanical, electrical, etc. In this work, bond graph theory is used with a three-stage procedure to fulfill the tasks of fault detection and isolation. First, the multivariate statistical method of principal component analysis is used to remove outliers and reduce the data dimensions. Second, the discrete wavelet transform is applied to the resulting scores to abstract the dynamics at different scales. In the third and final step, the Mahalanobis distance is applied to the results found in step two to calculate the confidence level. Based on the degree of violation from the nominal probability level, the detection of a potential fault is concluded to be true. Following a conclusion of true, fault isolation is achieved by comparing the time scale at which the violation of the nominal probability level occurred to the time scale associated with each physical domain. Two examples are presented to demonstrate these concepts.