On-line diagnosis of a power generation process using probabilistic models

Diagnosis has been applied in several approaches in all human activities. The general approach is the construction of a model that predicts the behavior of the system in order to compare it with the observed behavior. Sometimes, additional models of the process are constructed in the presence of certain failures with the aim of identify these failures. This paper presents the utilization of a previous work developed for sensor validation, to diagnose a complete process and not only the sensors. The main advantage of this approach is the construction of a model when the process is working properly. Only one model is necessary. This is done using historical data and machine learning algorithms for Bayesian networks. Having a model of the correct process, the early detection of any deviation of the normal behavior is possible. A case study of a steam generator (boiler) of a power plant is presented.