Damage Identification Strategy in Time-Varying Dynamic Systems Combining Cepstral and Image-Based Features

Damage identification and performance assessment in structural systems are vital for ensuring their safety, dura-bility, and efficiency. Vibration-based pattern recognition strategies have become popular in the past decades to identify anomalies using the vibration response from the monitored structures. However, conventional methodologies often fail to recognize and classify the different types of damage, particularly in dynamic structural environments where patterns change over time. These limitations highlight the need for a more adaptive and comprehensive approach that can adjust to time-evolving conditions and detect subtle variations indicative of damage. This study aims to develop an innovative framework for damage identification by integrating cepstral and time-image-based features to leverage the strength of both sets of indicators. Time image-based features like the Gramian Angular Field capture underlying structural dynamics by encoding temporal correlations among data points as angular values in a polar coordinate system. This technique allows for the application of image-based machine learning methods to time-series analysis by converting the data into a two-dimensional matrix. Additionally, combining cepstral coefficients improves the capability of tracking stationary and nonstationary changes in the frequency content of the systems. By fusing these sets of parameters, this study aims to enhance the accuracy and efficiency of structural damage identification, particularly in systems with time-varying behaviors.