Digraph and fault tree technique for online process diagnosis

Available diagnostic systems and existing diagnostic techniques illustrate the need for a comprehensive, model-based theory for on-line process diagnosis. This thesis classifies the diagnostic task into two functions, problem recognition and fault detection. The thesis hypothesizes the need for goal-directed, causal models to perform these functions in chemical process systems.; A diagnostic theory is developed comprising goal-directed, causal models and a method to use these models to perform the diagnostic functions. Digraph and fault trees form the causal model base. The fault tree's primal events are verified with available on-line indications using probabilistic theory and the digraph's causal structure. Problem recognition is disclosed through monitoring the on-line unreliability of the process goal. Fault detection is accomplished through ranking the failure rates of potential process failures using on-line indications and design data.; The theory provides a basis for performing comprehensive, robust system diagnosis. Examples illustrate the technique's ability to diagnose single and multiple system faults including equipment failures, disturbances, and indicator failures in linear and control loop systems. Limitations to application of the theory are discussed.