Fast reconstruction of velocity field in inlet for turbine-based combined cycle engine (TBCC) based on sensor-layout optimization

The inlets of turbine-based combined cycle engines confront a complex flight environment with wide-flight speeds and dual-inlet operation. Their internal flow field is affected by various factors such as the incoming Mach number, variable geometry channel, and backpressure of the outlet. Conventional end-to-end models and reduced-order models have achieved significant results when confronted with regular flow fields or uniform grids. However, problems such as the considerable amount of data, high cost, and low accuracy in the reconstruction of irregular flow fields exist. A reconstruction method (GPOD-MLP, combining gappy proper orthogonal decomposition and a multilayer perceptron) of the velocity field for irregular flow fields is proposed. It uses wall‑pressure information to reconstruct the velocity field of a turbine‑based combined cycle engine inlet. First, simulation datasets were constructed for different incoming Mach numbers, variable geometry channels (rotation angle of the splitter plate), and different backpressures. Then, improved particle swarm optimization was used to optimize the parameters on the wall, and the sensor layouts were optimized based on actual engineering requirements. Finally, Gappy POD was used to rapidly extract the velocity-field characteristics of the TBCC inlet, and MLP was used to combine coordinate information to obtain more wave structures and flowfield details. This further completed the high-density and high-precision reconstruction of irregular flow fields. The results indicated that sensor-layout optimization achieved significant results by considering practical engineering requirements while improving accuracy and reducing costs. The reconstruction accuracy of GPOD-MLP was higher than that of the other methods; the reconstruction error reduced by 70%–90% (e.g., 89.9% for the combined-sensor layout under unstart condition) compared with GPOD-CNN. This study successfully captured the separation region of the boundary layer and shock train leading edge (STLE). It ensured good reconstruction capability even when the inlet was in an unstarted state. This is crucial for identifying multiple working modes and monitoring hazardous states, specifically inlet unstart.