Knowledge discovery from comparisons of specialized and conventional metrics determined from acoustic fields and measurements

Abstract

Once acoustic signals (information streams) are recorded, there are myriad possible means for extracting and representing the embedded scientific or engineering knowledge that awaits discovery. Given the rich history of acoustic investigations, such knowledge extraction commonly involves comparisons based on conventional or specialized metrics chosen or devised to indicate critical underlying phenomena. Such comparisons may rely entirely on simulation results, experimental results, or a combination of the two. However, a suitable metric must be used in each case. This presentation provides four examples where conventional or specialized metric comparisons indicated new knowledge: (1) performing Monte-Carlo simulations and machine-learning predictions of the probability density function (PDF) of acoustic transmission loss (TL) in uncertain ocean sound channels to attain knowledge of how to best represent PDF(TL) when predicting it; (2) Using simulated acoustic-radiation data from vibrating structures and specialized dilation cross correlation metrics to remotely generate knowledge of experimental damage type and severity; (3) Comparing the measured horizontal coherence of acoustic fields and autoproducts to extract knowledge of their relative coherence lengths and bandwidths; and (4) Scaling of high-Reynolds-number hydrofoil-trailing-edge velocimetry and surface-pressure fluctuations measurements to attain knowledge of the near-wake conditions that lead to tonal noise. [Sponsored by ONR and NAVSEA.]