Towards Improved Acoustic Emission-based Methodologies for Structural Damage Identification and Visualization
Abstract
Non-destructive testing (NDT) techniques have emerged as a valuable tool for detecting damage and evaluating the overall structural condition, leading to enhanced safety and optimized maintenance of large-scale structures. Acoustic Emission (AE) approach is one of the powerful NDT techniques that are suitable for damage detection due to its high sensitivity to localized damage. Depending on the nature of structures and loading conditions, a suite of AE parameters such as duration, signal strength, amplitude, rise time, counts, and energy indices reflect their as-is state and detect any anomalies. However, these parameters are sensitive to environmental conditions and background noise in the measured data, which often leads to misleading information (i.e., false alarms) about the existing condition of the monitored system. This thesis aims to develop a suite of improved AE waveform-based damage detection, localization and visualization approaches by employing advanced signal processing techniques and classification models. The primary aim is to revolutionize AE methods that are free of any user-defined AE parameters and do not involve manual and subjective preprocessing tasks (e.g., setting a threshold and using denoising filters). Over the course of this thesis, a suite of advanced signal processing techniques is employed to decompose the nonstationary AE waveform collected from structures using limited sensors and extract the key AE components containing damage signature. Furthermore, several classification and visualization tools are explored to detect and identify the presence of damage using denoised AE components. Overall, this research makes a paradigm shift from AE parameters-based analysis to AE waveform-based analysis by extracting the damage information directly from raw data.
When it comes to long-term monitoring of real-life structures using an AE monitoring system, it leads to massive AE data. Dealing with such big data becomes a time-consuming exercise using traditional feature extraction methods. In this thesis, a Deep Learning (DL) technique augmented with time-frequency decomposition is further explored to automate the process of damage prediction and classification by leveraging the capabilities of Artificial Intelligence. The performance of improved AE waveform-based methodologies is validated using a suite of experimental and full-scale studies. Overall, the proposed methodologies can undertake robust damage identification using time and time-frequency information of AE waveforms, making them suitable candidates for robust damage detection, classification, and visualization.