Removing Unwanted Noise from Operational Modal Analysis Data

Abstract

Operational modal analysis data includes the measurement of dynamic signals such as structural vibration data and the corresponding excitation force or pressure. In addition to the desired information, measured structural vibration data can include unwanted electrical noise and vibration energy from adjacent structures. Measured dynamic pressures can contain unwanted signals such as acoustic and vibration induced pressures. In this paper, a noise removal technique is presented in which an unlimited number of unwanted correlated signals can be removed from a set of measured data. In its simplest form, this technique is related to coherent output power (COP). However, unlike COP noise removal, multiple signals can be removed from measured data while retaining the magnitude and phase of the original data required for modal analysis processing. This technique is demonstrated using vibration data and dynamic wall pressure measurements from a thin-walled aluminum cylinder filled with water flowing at 20 ft/s. IntroductionOperational modal analysis involves measuring the vibration of a system during operation either because the desired operational excitation is not appropriately represented by traditional impulsive modal impact testing or because the system cannot be conveniently taken off-line. Transfer function data between vibration and input force are traditionally measured and used as input for an experimental modal analysis. However, when the input force cannot be measured, cross-spectrum measurements between accelerometers can be used in place of the transfer function data. Resonance frequency, damping, and operational modes shapes can all be obtained from cross-spectrum measurements between accelerometers placed at locations required to appropriately resolve the structural mode shapes. Occasionally, measurements of dynamic force or pressure can also be made to help determine modal mass.Measurements of structural vibration or pressure often include unwanted signals not associated with the signals of interest -especially in an operational setting where other machinery or processes are often running. The unwanted signals may be associated with electrical noise, vibration energy from adjacent structures, vibration induced pressures, and acoustic pressures. While various noise removal techniques are available to "clean up" measured data, a conditioned spectral density technique [1] is presented which can remove multiple signals of unwanted noise from measurements of cross-spectra. This technique can be applied to any dynamic signal measurements which meet necessary criteria.Lauchle and Daniels [2] employed a related subtraction technique using multiple sensors measured simultaneously to remove noise signals from their measurements of turbulent boundary layer (TBL) wall pressure. Naguib, Gravante, and Wark [3] discuss the advantages of an optimal filtering approach compared to a difference approach. In their work, the optimal filtering provides a better estimate of the noise than the difference approach. The ...

Output-only Modal Analysis of a Beam Via Frequency Domain Decomposition Method Using Noisy Data

“…After a clean data acquisition, techniques such as filtering and averaging are used to reduce noise effects [10][11][12]. Bonness and Jenkins [13] presented a noise elimination technique wherein it is possible to remove a limitless number of undesirable correlated noise from a collection of experimental data by modifying the statistical correlation relations and spectral functions. Focusing on the study of the Gaussian noise elimination, Feng and Lin [14] introduced numerous noise removal procedures based on wavelet thresholding including global, Maxmin and BayesShrink.…”