The Use of Artificial Intelligence for Cooling Failure Detection in High-Pressure Turbine Blades

Abstract Modern High-Pressure Turbine (HPT) stages in aircraft engines are designed to operate at very high temperatures to maximize the engine performance and efficiency. Blade coatings and sophisticated cooling systems are the main strategies to safeguard the integrity of HPT blades from the hot combustion gases at temperatures even over 2000 K. The cooling flows, extracted from the compressor flow path, are channeled in a complex layout of pipes within the engine casing and the HPT blades. The engine control system must ensure continuous delivery of cooling gas because any interruptions could seriously compromise the blade integrity. Innovative techniques using Machine Learning combined with the Computational Fluid Dynamics (CFD) simulations can support the development of extended monitoring systems or control systems. With the HPT of the Energy Efficient Engine (E3) as reference case of a 3D cooled gas turbine blade, a database of CFD simulations (using the open-source turbomachinery software MULTALL) is performed by varying the cooling mass flow across different ejection zones in a single blade row, simulating real-world scenarios such as partial or full clogging. The use of the open-source platform allows the simulation of a large number of cases in parallel that would be prohibitively expensive for an academic research activity if based on commercial licenses. From the resulting CFD dataset, a large number of global parameters, that can be measured during engine operation, are extracted, together with the local spanwise temperature distributions near the leading and the trailing edge of the inspected row. The resulting database is used to train two Neural Networks models: the first one can reconstruct the temperature distributions from the few measurable parameters, the second one uses the reconstructed temperature profiles to understand if the cooling system has some malfunctioning and where this occurs (Leading Edge, Suction Side and Trailing Edge or Endwalls). In the paper the potential of the proposed approach for the setup of extended engine monitoring tools is demonstrated.