This paper presents an analysis of the operation of a stage of an aircraft engine gas turbine in terms of generation of flow losses. The energy loss coefficient, the entropy loss coefficient and an additional pressure loss coefficient were adopted to describe the losses quantitatively. Distributions of loss coefficients were presented along the height of the blade channel. All coefficients were determined based on the data from the unsteady flow field and analyzed for different mutual positioning of the stator and rotor blades. The flow calculations were performed using the Ansys CFX commercial software package. The analyses presented in this paper were carried out using the URANS (Unsteady Reynolds-Averaged Navier-Stokes) method and two different turbulence models: the common Shear Stress Transport (SST) model and the Adaptive-Scale Simulation (SAS) turbulence model, which belongs to the group of hybrid models.

The aim of this paper is to assess the impact of the mutual positioning of the turbine stage stator and rotor blades on noise generation. The Ansys CFX commercial software package and the Scale-Adaptive Simulation (SAS) hybrid turbulence model are used for numerical analyses. The paper is focused on an analysis that the pressure wave generation resulting from unsteady flow phenomena. In order to present the problem, the Fast Fourier Transformation (FFT) analysis of pressure fluctuation is carried out at selected points of the turbine stage computational domain. A comparison of values of individual components for subsequent control points allows an approximate determination of the place of generation of pressure waves, the direction of their propagation and the damping rate. Moreover, the numerical analyses make it possible to evaluate the justification for the use of the SAS model, which is rather demanding in terms of equipment, in simulations of unsteady flow fields where generation and propagation of noise waves occur.

CAD models clustering with machine learningSimilarity assessment between 3D models is an important problem in many fields including medicine, biology and industry. As there is no direct method to compare 3D geometries, different model representations (shape signatures) are developed to enable shape description, indexing and clustering. Even though some of those descriptors proved to achieve high classification precision, their application is often limited. In this work, a different approach to similarity assessment of 3D CAD models was presented. Instead of focusing on one specific shape signature, 45 easy-to-extract shape signatures were considered simultaneously. The vector of those features constituted an input for 3 machine learning algorithms: the random forest classifier, the support vector classifier and the fully connected neural network. The usefulness of the proposed approach was evaluated with a dataset consisting of over 1600 CAD models belonging to 9 separate classes. Different values of hyperparameters, as well as neural network configurations, were considered. Retrieval accuracy exceeding 99% was achieved on the test dataset.