Automated experimental modal analysis with density-based clustering: A benchmark validation through jury testing

Abstract

Modal analysis has evolved into a major technology for the study of structural dynamics in the past several decades. Through it, complex structural dynamics phenomena can be represented in terms of structural resonance characteristics, i.e., the modal parameters: natural frequencies, damping ratios, and mode shapes. These analyses are often performed by analysts who manually select the system’s poles (which represent the modal parameters), from so-called stabilization diagrams, which is a time-consuming and error-prone process. Furthermore, the difficulty of interpreting stabilization diagrams increases with the complexity of the dataset, sometimes requiring an expert with a high level of domain knowledge to interpret it. Therefore, the automation of pole selection in modal analysis is important to accurately process complex datasets without user-dependent interaction and with repeatability. In this work, three Automated Modal Analysis (AMA) methodologies are proposed using three Machine Learning (ML) density-based clustering techniques, combined with domain knowledge of modal parameter selection. The methodologies are benchmarked against the manual selection by multiple engineers with different levels of modal analysis expertise. The benchmark study consists of the analysis of eight industrial datasets, and the further comparison to another industrially used AMA method applied to the same datasets. A comparative overview of the results is described in this paper, along with the advantages of the proposed AMA methodologies.