The project deals with the ambitious objective of developing a Machine Learnt model able to predict the erosion/corrosion mechanism in compressor blades subjected to online water-washing techniques for dirt and salt deposit removal. The project will develop along two main research lines: a) the assessment of an innovative mathematical model aiming at predicting the combined effect of erosion/corrosion (pitting) on metal surfaces where deposit fouling can be present due to operating conditions (e.g. due to salt scaling); the model will be implemented on a commercial code (e.g. Ansys Fluent) through an UDF that will be made available on request to any researcher. b) development and validation of a machine learnt approach based on variational auto-encoders and generative adversarian networks able to correctly predict pitting when feeded with the near-wall flow field and droplet characteristics and the blade material.
It is important to point out that these two activities are relevant `per se¿ and will allow obtaining important research outcomes even when they are considered separately.
The development of the project will require a continuous interaction between the two research lines. These will eventually lead to the implementation of one comprehensive tool of the built model that will be able to generalize. In other words, the model will be applied to cases not included in the database, and even in these cases the algorithm is expected to provide accurate predictions. Once developed and validated, this code will represent a powerful tool for supporting companies producing turbomachinery in the definition of the operating conditions (i.e. setting a proper water washing procedure able to guarantee deposit removal) as well as the time lag between successive stops for blading inspection and overhauling.
