An Improved Spectral Subtraction Method for Eliminating Additive Noise in Condition Monitoring System Using Fiber Bragg Grating Sensors

Abstract

The additive noise in the condition monitoring system using fiber Bragg grating (FBG) sensors, including white Gaussian noise and multifrequency interference, has a significantly negative influence on the fault diagnosis of rotating machinery. Spectral subtraction (SS) is an effective method for handling white Gaussian noise. However, the SS method exhibits poor performance in eliminating multifrequency interference because estimating the noise spectrum accurately is difficult, and it significantly weakens the useful information components in measured signals. In this study, an improved spectral subtraction (ISS) method is proposed to enhance its denoising performance. In the ISS method, a reference noise signal measured by the same sensing system without working loads is considered the estimated noise, the same sliding window is used to divide the power spectrums of the measured and reference noise signals into multiple frequency bands, and the formula of spectral subtraction in the standard SS method is modified. A simulation analysis and an experiment are executed by using simulated signals and establishing a vibration test rig based on the FBG sensor, respectively. The statistical results demonstrate the effectiveness and feasibility of the ISS method in simultaneously eliminating white Gaussian noise and multifrequency interference while well maintaining the useful information components.

Keywords

• condition monitoring
• signal processing
• spectral subtraction
• additive noise
• fiber Bragg grating sensor