GPU-accelerated large-eddy simulations of supersonic jets from twin rectangular nozzle

Large eddy simulations of supersonic jets from a twin rectangular nozzle are performed with the GPU-accelerated version of Cascade's "charLES" compressible flow solver. Comparisons of the numerical noise predictions with near-field and far-field microphone measurements show good agreement for the screech tone frequency, broadband spectra and overall sound pressure levels. For the screech tone amplitude, the agreement is further improved when the reflective surfaces upstream of the nozzle exit are modified in the computational domain to account for the thick acoustic foam covering these surfaces in the wind-tunnel. Such modifications and efficient exploration of the nozzle design space are made possible by the increased computational throughput with the GPU-accelerated solver: for the present O(100) million cell mesh and relatively long noise data collection of 2000 acoustic time units, the simulation results are obtained in about 12 hours on 30 standard GPUs.