<p style="text-align: justify">Metallic thin-walled structures, with their vast applications in fields such as aerospace, automotive, and civil engineering, have a tremendous importance in design of complex engineering systems. Prolonging the life and minimizing the maintenance costs of those systems is highly correlated with an early detection of damage in them. As yet, multiple conventional damage detection techniques have been devised to address this problem. However, one of the main challenges with regard to those techniques is the lack of consistency of performance for utilization in different structures. This inconsistency arises from the incapability of revealing the small differences between the healthy and damaged states of a system by conventional techniques. Machine Learning (ML) algorithms possess a distinguished capability in recognizing and classifying the patterns in a data set. Therefore, their application in damage detection methods has been of great interest in recent studies. In the current paper, by proposing a framework via deployment of several classifiers, the presence of a damage on an aluminum plate is assessed. The required data sets for the classifiers are provided through a hybrid technique comprised of numerical modeling of the structure and statistical analysis of the signals. The signals employed in this study are based on ultrasonic guided waves (UGWs). They can be actuated by using piezoelectric transducers, that are cheap and lightweight. However, in this work, the actuation of the excitation signal is simulated numerically. Commercial finite element package, Abaqus, is used for modeling the wave propagation in the plate. Tone-burst functions at several central frequencies are set as the excitation type, which are applied on the structure as a point force. The propagated wave, resulting from the excitation force, is measured at several locations on the aluminum plate, and the obtained signals are designated as the first raw data for the classification algorithms. The described procedure is repeated for several scenarios in which the central frequency of the excitation function is altered, or the plate has a damage or not. Based on the obtained signals from the numerical simulations, a data set is formed. Measred ultrasonic signals, even for a very short period of time, consist&nbsp;of a high number of samples. Hence, for reducing the dimension of the input space, several feature extraction techniques based on statistical signal condition indicators are employed. Based on the obtained features, a matrix is formed where the columns correspond to different features, and each row correspond to an instance. This study shows the importance of the selection of the proper combination of the features for the classification of the state of the structure. Additionally, the presented results highlight the impact of the duration of the measured ultrasonic signals on the feature vectors and ultimately on the performance of the classifiers. Moreover, the importance of a sensitivity analysis for choosing the optimized&nbsp;hyperparameters of a classifier is outlined. The result of this study contributes to devise a more automated and generalized process for detection of damages in metallic structures.</p>

Feature Generation and Selection for Identification of Damage in Thin-Walled Structures Based on a Statistical Approach

Paper Proceedings: https://asmedigitalcollection.asme.org/SMASIS/SMASIS2021/volume/85499

ASME 2021 Smart Materials, Adaptive Structures and Intelligent Systems (SMASIS2021)

Supporting the need among researchers worldwide for additional online and multimedia content, ASME now offers the ASME Conference Video Collection. This platform will collect and provide long-term access to conference presentation videos, encompassing all technology areas within ASME's technical conference and events program. About 30,000-40,000 engineers across all mechanical engineering and related disciplines participate in ASME conferences each year. This includes authors of approximately 6,000 to 8,000 technical papers published by ASME, the majority of whom also submit video presentations for our virtual conferences. The video collection will be available as an add-on to existing product packages or as a separate subscription.