Sviluppo di tecniche numeriche di simulazione e di Intelligenza Artificiale per l’analisi di palettature di Turbine a gas

The work of this PhD thesis investigates the using of different CFD codes for simulation of axial gas Turbine, with particular focus on the High-Pressure part in the aeronautic application. The CFD can be an effective tool, not only for the design new geometries or Validation phase of experimental cases, but also for coupling with “real time” monitoring with using of Artificial Intelligence (AI) algorithms. As first step, a comparative study was conducted in a Meanline Code, implemented in collaboration with a partner company; on Multall,an open-source project CFD code and a commercial CFD code CFX, most widely used code in academic and industrial field. The study was carried out using CFX as a reference, which is the most robust of three codes and it was used to validate the other two codes, which have the advantage of being computationally faster and lighter and which allow databases to be implemented for different geometries and operating conditions. The Validation was carried out on an experimental case design form the late 1970s: Energy Efficient Engine(E3), an aircraft engine of General Electric, focusing on the High Pressure Turbine (HPT). In this work, the geometry of the HPT was redesigned, using the Multall code, in order to obtain an equivalent machine to the reference one that maintains the same overall performance at the design point, but with a different geometry of the blades. The are two designs of the HPT of the E3: a configuration with a single stage and a configuration with two stages, coupled with the blade cooling technique. The first one is implemented to validate the CFD and Meanline codes, the second one is used to perform simulations with blade cooling, which could be implemented in all three coded, in this case the potentials of Multall code are exploited. Those simulations become the reference data for a Machine Learning algorithm application with the Neural Network for monitoring and detecting faults in the turbine cooling system. It was therefore analysed how these types of algorithms can help in the experimental field of the turbomachinery monitoring, highlighting the limitation and potential developments in the AI field, which in recent years has become very popular in many sectors, not only in engineering.