On-line fault diagnosis using the signed directed graph

Abstract

Fault diagnosis using structural knowledge, namely, the signed directed graph (SDG), is presented. A design procedure is proposed to overcome several problems associated with the SDG : it produces spurious (multiple) interpretations and it may delete the true interpretation when the process variable is going through nonsingle transition (this is frequently encountered in a control loop). The proposed method has the following features : discretize a continuous process response into several states, and different conditions (truth tables) are imposed to check the consistency of fault propagation; find the dominant path of fault propagation using steady-state gains; and express the variable associated with the integrator in the velocity form. The first feature improves the modularity of the diagnostic system, which in term makes the design and maintenance of the diagnostic system easy. Furthermore, improved diagnostic resolution can be achieved by imposing more stringent conditions at different states and by finding the dominant path. The third feature enables the system to handle variables with nonsingle transition in a control loop. A CSTR example is used to illustrate the design procedure. Simulation results show that the proposed approach based on the SDG provides an attractive alternative for process diagnosis