Multi NNARX model of complex engineering systems for fault detection and diagnosis applied to a fossil fuel electric power plant

In this paper, we present a dynamic Multi-NNARX model to describe the full operation range of complex engineering systems and its application to a Fossil Fuel Electric Power Plant. The proposal is based on a black-box approach under the assumption that man-made systems, designed to satisfy performance specification, are susceptible of being modeled by transposing linear models. The model is a switched Multi Neural Network ARX model where the structure and coefficients of every model are identified for the main operation modes of the plant. We enhance the identification process introducing an exhaustive method of evolving neural-networks designed to avoid over-fitting and bad-generalization of the model, and we amalgamate the signals of the parallel models in a unique output by using the rate of change to maximize the expectation of selecting the right output. The Multi-NNARX model is intended to be used in a Fault Detection and Diagnosis System based on models for complex engineering systems, presented as a companion paper, thus, we illustrate its application identifying the main operation modes of the Steam Generator Subsystem of a Fossil Fuel Electric Power Plant using data from a high performance simulator.