Applying Machine Learning to Jet Noise Prediction

Dowdall, Jonny

关键词: DATA MINING;    DATA PROCESSING;    EDUCATION;    ERROR ANALYSIS;    JET AIRCRAFT NOISE;    JET FLOW;    LINEAR SYSTEMS;    MACHINE LEARNING;    NEURAL NETS;    NOISE PREDICTION;    PRODUCT DEVELOPMENT;    PROVING;    PYTHON (PROGRAMMING LANGUAGE);    SURFACE PROPERTIES;   

RP-ID  :  GRC-E-DAA-TN62141

学科分类：声学和超声波

美国|英语

来源: NASA Technical Reports Server

【 摘 要 】

This presentation summarizes the application of machine learning to jet noise data in an effort to predict the resulting noise from the interaction between a jet and a hard surface. The Aero-Acoustic Propulsion Laboratory at the NASA Glenn Research Center has acquired the noise resulting from the interaction between a jet and metal plate over a range of surface placements (e.g. plate lengths and positions) and a range of jet flow configurations. For each configuration, the noise was measured at 24 observer locations via a microphone array centered around the jet nozzle. An artificial neural network developed with Keras and TensorFlow was trained on the data to predict an 88-band spectrum as a function of surface placement, jet conditions, and observer location. Analysis of the machine learning models provide insight into which experimental parameters contribute more to the noise and which parameters could potentially be removed entirely to simplify future experiments. Preliminary results will be discussed and presented via a live demonstration of the software, which outputs a sound spectrum in real-time with user-inputted jet-surface configurations.