A fuzzy-neuro architecture for modular fault isolation in complex systems

Faulty behavior of complex industrial systems is usually detected by deviations from expected values of performance parameters, regarding the system´s utilities. When the isolation is performed by human diagnosticians, they use shallow knowledge on faulty cases, and deep knowledge on laws governing the installation running, both imprecise in variable values and in their relations. The paper presents a fuzzy-neuro architecture for a computational model, handling the imprecision of human-like thinking on faults and behavior of the conductive flow systems in industry. The primary effects isolation from the secondary effects proceeds in a modular fashion, based on four transport anomalies detected on the flow power disturbance in each (bond graph) junction of interconnected components. The manifestations take a fuzzy representation and the multiple links between manifestations and faults is represented and processed by productive neural networks. The modular structure of the diagnosis system reflects the modular structure of the target system, hence the faults-effects mapping is simpler and the isolation of faults is more precise