A time domain approach for noise source identification in periodic systems

Hong-Yuh Lin, Purdue University

Abstract

A simple and novel approach to help identify internal source mechanisms for the machinery noise problem has been developed and applied to a small high speed rotary compressor. Because it utilizes a time domain feature extraction and comparison procedure, the approach is capable of dealing with fast time-varying random signal features and possibly high coherence among measured internal process responses frequently encountered in the high speed rotary machinery. The feature extraction relies on the second moment estimation for a nonstationary random signal by the ensemble averaging. The technique was implemented on an HP5451C two-channel FFT analyzer for on-line feature extraction. The feature comparison is based on a linear multiple regression model to calibrate the level of likeness between the external noise signal and each internal process response signal on the event basis. Several simulation cases were introduced to test a computer program written for the feature comparison. Experimental studies were conducted to test robustness of the parameter estimation in the regression analysis when different criterion values required in the analysis were chosen for feature comparison using real data measured from a specially-designed bolted case compressor. In addition, data at two higher noise levels were also studied to examine the consistency analysis results. As a result, for the selected noise band, 3-5 kHz, two major source concepts were formulated: (1) mechanical excitation due to vane and cylinder interaction and (2) flow excitation during the gas discharge period. Also, for the rotary compressor under study, there were two major events which contributed to the 3-5 kHz band noise. In an attempt to verify the proposed source concepts, operating and design parameters of the compressor were varied to test any significant noise sensitivity. It was found that the proposed source concepts were supported by positive trends obtained from these designed tests.

Degree

Ph.D.

Advisors

Sherman, Purdue University.

Subject Area

Mechanical engineering

COinS