Sound extraction and recognition for running machines
The extraction of target sound source and de-noising is in high demand in applications such as structural health monitoring, non-destructive evaluation with acoustic emission, and the ordinary reduction of background noise in acoustic devices, like cell phones and hearing aids, etc. Specifically, a directly measured acoustic data is a clue to identify the status of the running device and evaluate its performance. For example, analysis of engine noise can be used to monitor the proper inlet/outlet airflow and fuel combustion inside the cylinders. Directly measured acoustic data at the engine includes a high level of background noises from other machine parts near the engine thus the target sound signal is buried. Acoustic extraction algorithm needs to be applied to the directly collected data to get a clear chart for analysis, and a further recognition approach, with the comparison of the collect data to benchmark database, can tell if the engine is working properly.
This project will focus on measurement of sound produced by running machine and apply signal processing algorithms to the directly measured data to extract acoustic signals, followed by analysis of the signals with pattern recognition methods, thus determine the operation status of the machine. With such approach, any abnormality of the running machine can be detected and identified prior to failure. Multiple signal processing techniques, such as fast Fourier transform, short-time Fourier transform, wavelet transform, filtering, windowing, etc. will be applied. Numerical simulation and experimental validation will be applied to various types of target signals, such as transient sound, continuous sound, narrow band, and broadband sound. Extraction and recognition results will be compared and discussed. An optimal solution will be determined for various types of signals, and different factors will be set up.
The student will work on the numerical model, simulation, and experimental validation with real-time sound data collected in the machine shop.
No special requirements
September – October 2017: Build up program code for different functions, such as fast Fourier transform, wavelet transform, filtering, and windowing.
November – December 2017: Build up program code for pattern recognition.
January – February 2018: Numerical simulation for sound extraction and recognition.
March – April 2018: Experimental testing with real data collected in a machine shop.
The student needs to be familiar with Labview and Matlab software and have a fundamental knowledge of coding. Students in computer science and information systems and mechanical engineering can apply. A fundamental knowledge of mechanical systems is required.
- Job Opening ID
- Fall 2017 and Winter 2018
- Work could be done by someone not coming to campus (e.g., online or non-local student)
- What majors can apply?
- Computer Science and Information Systems (MS)
- Mechanical Engineering (MSE)
- Faculty Name
- Na (Linda) Zhu