Event Reconstructing Adaptive Spectral Evaluation (ERASE) Approach to Removing Noise in Structural Acceleration Signals

Abstract

Floor vibrations for event localization has gained attention recently for its human-related applications such as footstep tracking. However, noise can corrupt signals, reduce signal-to-noise ratios (SNR), and lead to imprecise estimations of the event’s amplitude and force. Techniques to remove noise have been developed such as bandpass filters, which eliminate noise without regard to overlapping event frequency components. These methods can corrupt the signal, removing important information about the event. The authors propose adapting a common speech processing technique, called spectral subtraction using half wave rectification, to remove only the noise’s contribution. The Event Reconstructing Adaptive Spectral Evaluation (ERASE) approach is compared to unfiltered and Butterworth-filtered data in impact localization and force estimation through the Force Estimation and Event Localization (FEEL) Algorithm. A total of 810 impacts from ball drops of five different heights and impulse hammers across eighteen locations were utilized for testing. Signals were corrupted by noise from different sources. ERASE demonstrated 93.9% average impact localization accuracy and -2.40% ± 1.85% force magnitude error on a 99% confidence interval, improving the SNR verse the other filtering techniques.