An Architecture for Online Diagnosis of Gas Turbines

Abstract

Diagnosis systems are becoming an important requirement in these days given the complexity of industrial systems. This article presents an architecture for online diagnosis based on probabilistic reasoning. Probabilistic reasoning utilizes a model of the system that expresses the probabilistic relationship between the main variables. Thus, the values of some variables are utilized as evidence and the propagation provides an inferred value of other variables. Comparing the inferred value with the real one, an abnormal condition can be detected. Next, an isolation phase is executed in order to find the root cause of the abnormal behavior. This article presents the design of an architecture that performs online diagnosis of gas turbines of combined cycle power plants. The architecture was designed utilizing some of the classes of the Spanish elvira project as a double experiment: (i) to test a general purpose, probabilistic reasoning package elvira in a real application in a real time environment and (ii) to test a previously developed theory for diagnosis in a gas turbine.