PRANK: A Singular Value Based Noise Filtering of Multiple Response Datasets for Experimental Dynamics

Abstract High quality measurements are paramount to a successful application of experimental techniques in structural dynamics. The presence of noise and disturbances can significantly distort the information stored in the data and, if not adequately treated, may result in erroneous findings and misleading predictions. A common technique to filter out noise relies on decomposing the dataset into singular components sorted by their degree of significance. Discarding low-value contributions helps to clean the data and remove spuriousness. This paper presents PRANK, a novel singular value-based reconstruction approach for multiple-response vibration datasets. PRANK integrates the effect of Principal Response Functions and Hankel filtering actions, resulting in an improved data reconstruction for both system poles and zeros. The proposed formulation is tested on both analytical and numerical examples, showcasing its robustness, efficiency and versatility. PRANK operates with both time- and frequency-based data. Applied to noisy full-field camera measurements, the filter delivered excellent performance, indicating its potential for various identification tasks and applications in vibration analysis.