Accurate performance prediction across a wide operating range remains challenging for high-performance power turbine blades. To address this issue, a graph-attention network model, GAT-5s, was developed using small-scale experimental data to predict aerodynamic parameters of rotor blade profiles over a wide-incidence range. On the validation profile, GAT-5s achieved mean relative errors (MREs) of 1.6% for the total pressure loss coefficient, 0.2% for the outlet Mach number, and 2.7% for the deviation angle. These errors were lower than those of the back-propagation neural network baseline and the GAT-5s-CFD model trained on RANS under the same tested conditions. However, direct use of GAT-5s for surface pressure prediction produced training and testing losses above 1 & times; 10(-3), indicating limited capability to describe spatial pressure distributions. To address this limitation, GAT-5G was developed by integrating pressure-point position information and a bidirectional gated recurrent unit (BIGRU) module into GAT-5s. The model reduced the training and testing MSE losses to below 1.0 & times; 10(-4), and the prediction errors for both the test and validation profiles remained within 10%. Although GAT-5G did not always yield the minimum local error among all models, it provided balanced agreement with experimental pressure distributions across the tested incidence range. For wake-loss prediction on the validation profile, the MRE also remained below 10% for the tested incidence cases. These results indicate that GAT can serve as useful surrogate models for aerodynamic prediction of wide-incidence rotor blade profiles.
The application of casing treatments is a widely recognized method for improving the stall margin of fans and compressors. However, dimensional constraints in modern aero-engines, particularly in designs with S-duct or boundary-layer ingesting inlets, limit the use of treatments requiring bulky auxiliary structures. This article investigates the effectiveness of a circumferentially partial foam metal casing treatment (PFMCT) combined with an adjacent circumferential cavity. The study is conducted experimentally on a low-speed, single-stage axial flow fan under both uniform and distorted inlet conditions. The effects of varying the circumferential coverage of the foam metal and the influence of the relative angular position between the PFMCT and a 180-degree inlet distortion are systematically examined. Results indicate that PFMCT provides a significant stall margin improvement, which is further enhanced by the presence of the circumferential cavity, especially under uniform inflow. Unlike full-annulus casing treatments, especially the ones with annulus back-chamber, the partial configurations do not alter the inherent spike-type stall inception of the fan. The stabilization mechanism is attributed to two distinct functions: the foam metal section is primarily responsible for dissipating unsteady disturbance energy and redistributing rotor loading, while the circumferential cavity contributes by achieving a more uniform steady-state circumferential loading distribution. Furthermore, stability enhancement is shown to be highly sensitive to the relative circumferential position of the casing treatment and the inlet distortion zone. An optimal alignment is identified where the high-loading region induced by the distortion directly interacts with the cavity, maximizing the load-averaging effect and thereby achieving the greatest improvement in stall margin.
Computational fluid dynamics (CFD) simulations are crucial for optimizing engineering designs, but are often computationally expensive and time-consuming. This article introduces a data-driven artificial intelligence (AI) surrogate model for axial fans that provides rapid approximations of 3D CFD results, significantly reducing the computational resources required for 3D flow solutions. The AI model leverages existing CFD data to learn the complex relationships between geometry, boundary conditions, and flow solutions, allowing it to predict flow fields for new, unseen geometries. This research extends the application of AI surrogate modeling to unstructured data for axial fans in industrial settings, based on previous work with structured data for aircraft compressors. It demonstrates the model's ability to predict low Mach number and incompressible flows relevant to industrial axial fans, expanding the scope of AI in turbomachinery beyond traditional high-speed aerospace use cases. The flexible model architecture accommodates both structured and unstructured CFD data from a variety of turbomachinery flows. Trained on a design of experiments (DOE) database using XYZ coordinates of geometric surface points, rotational speed, and boundary conditions, the model accurately predicts flow variables such as velocities, pressure, and density. Validation against optimized fan designs shows promising agreement with CFD results, confirming the model's effectiveness in capturing essential flow features for designs not included in the training.
Abstract Rotor forced response is a major aeromechanical problem causing blade high-cycle fatigue, and mistuning can significantly increase the maximum blade vibrational amplitude due to mode localization. Most previous mistuned forced response experimental studies were conducted at slow rotational speed sweep rates through the resonance and analyzed using the steady-state assumption. However, in practice, engine sweep rate can be significantly high such that the classical steady-state forced response analysis becomes invalid. To study the transient forced response in a realistic environment, a comprehensive experimental study was conducted in the Purdue three-stage axial research compressor by measuring rotor blade response at four different sweep rates in both acceleration and deceleration. An overall peak amplitude reduction and peak frequency shift were observed at higher sweep rates, along with a transient beating occurring at the highest sweep rate. Initially, an analytical transient analysis for a single-degree-of-freedom (SDOF) system was conducted. While it explains the major features of the transient response well, it fails to accurately predict the amplitude change at different sweep rates. The transient effect differences measured between acceleration and deceleration are also completely missed by the SDOF analysis. Therefore, a transient mistuned rotor analysis was performed utilizing the fundamental mistuning model (FMM). Through a mode superposition approach, the transient mistuned rotor analysis shows that the difference measured in the acceleration and deceleration cases was caused by the relative phase change of the dominant modes contributing to the total resonant response.
Abstract During operation in sand-laden environments, aero-engine compressor blades suffer severe localized erosion, leading to progressive degradation of compressor stage performance over the engine service life. This issue is particularly critical for high-speed transonic compressor rotors. In this study, the erosion behavior of a transonic axial-flow compressor rotor is investigated using a CFD-based erosion prediction framework. Five representative particle diameters are selected based on the particle size distribution measured at the outlet of the upstream inlet duct. Three engineering-relevant particle mass loadings are prescribed through an accelerated loading approach. Under identical operating conditions, blade erosion patterns and surface aerodynamic responses are compared between clean and sand-laden inflow cases. The results show that the erosion level on the suction surface is approximately one order of magnitude lower than that on the pressure surface, and the suction-surface erosion appears as scattered spot-like patches. Within the considered loading range, smaller particles produce lower erosion-rate densities but cause stronger flow-field disturbances, whereas larger particles show the opposite behavior. As particle concentration increases, the contribution of particles to the radial variation of leading-edge erosion rate density increases by several times. Large particles mainly promote localized material removal and blade geometric degradation. Overall, these findings clarify the erosion–aerodynamic coupling characteristics of a transonic compressor rotor under different particle sizes and concentrations, and provide practical guidance for compressor design and protection in sand-laden environments.
