Failure Prediction of Highly Requested Complex Technical Systems: Application to W18v50df Engines

The work done in this paper has focused on the prediction of failures of a complex and highly stressed technical system for energy production, namely the W18V50DF engine powering an MW gas-fired power plant. The aim of this work is to highlight the prediction curves, a priori and a posteriori, of the evolution (probabilistic) of the state of these W18V50DF engines in order to anticipate the appearance of failure and to put a human-machine interface to facilitate the knowledge of a possible event and to allow a remote action. To do this, a hybrid method has been employed in the field of data-oriented modeling which highlights the neural network modeling used to determine the state of the components of the system studied by classification. Coupled with Bayesian network modeling, also known as probabilistic graphical models used to predict the state of the system. The neural model and the HMI have been built respectively through the ntools library and via the GUIDE library of the MATLAB software, while the probabilistic graphical model has been built using the BayesianLab software. The work carried out has shown that the W18V50DF engine and its components are degraded as their lifetimes evolve. In addition, because of its complexity and the criticality of some of its components, the degradation of the W18V50DF engine will be accelerated as each of them will be. In addition to this, the financial evaluation revealed that this work, beyond its multiple technical challenges, would allow the user to make significant financial gains.