Noise source detection in mechanisms by sound localization

The primary objective of this study is to investigate and implement methods capable of accurately determining the positions of close-range sound sources using a limited number of microphones. For this purpose, the four-microphone array of the Microsoft Xbox Kinect was utilized. It was identified that Time Difference of Arrival (TDOA) approaches based on the GCC-PHAT method, along with iterative nonlinear solution techniques, are particularly effective in this context. In the experimental setup, a linear microphone array consisting of four microphones was employed. The sound samples were taken in the form of singular sound impulses generated by metal-on-metal impacts, sampled at a frequency of 16 kHz. Subsequently, the GCC-PHAT algorithm, implemented in Python, was used to compute the time delays (TDOA) between microphone pairs. These delay values were then fed into an iterative linear equation model solved using symbolic matrix methods and the least squares approach to estimate the source location. In conclusion, the developed Python implementation and selected methods could not demonstrated effective localization performance at close range using a limited number of microphones but it was observed that this system can give an idea of general location of the sound source. A microphone array with more microphones which has a higher rate of frequency, the system can deliver more robust results. Future work will involve experimental setups incorporating z-axis variation, assessment of method performance with different types of sound sources, and analysis under increased noise conditions.