2aNS3. Acoustic source location using a neural network.

Session: Tuesday Morning, June 17

Author: Tarun Bhatt
Location: Dept. of Mech. Eng., Tennessee Tech. Univ., Cookeville, TN 38505, tb7101@tntech.edu
Author: Esther T. Ososanya
Location: Tennessee Tech. Univ., Cookeville, TN 38505
Author: Corinne M. Darvennes
Location: Tennessee Tech. Univ., Cookeville, TN 38505


This paper presents an artificial neural network (ANN) approach to locate an acoustic source and determine its distance from a microphone. First, two sets of experiments were performed to locate a rattling bolt on a plate using the unwrapped phase method, by comparing the acceleration signal from the vibrating plate to the acoustic signal received by an array of four microphones. The first test used a wide frequency bandwidth random signal and the second test used a short frequency bandwidth signal. Then the neural network was trained using this experimental data. The effectiveness of the ANN at locating a source using four microphones was evaluated, in light of experimental problems associated with frequency resolution and effects of reflecting surfaces. The ANN was also used to locate the same source using only three microphones. The expectation is that fewer microphones will be needed if an ANN is used and still provide the same level of accuracy on source location. This would result in better accuracy, or substantial savings for acoustic source location evaluation.

ASA 133rd meeting - Penn State, June 1997