Li Ai

Date of Award

Summer 2021

Document Type

Open Access Dissertation


Civil and Environmental Engineering

First Advisor

Paul Ziehl


Some complex infrastructure systems, such as nuclear facilities and bridges, are subject to structural damage due to environmental erosion, material deterioration, and other factors after long periods of use. Stress corrosion cracking (SCC) and alkali-silica reaction (ASR) have been identified as the primary degradation mechanisms for steel and concrete structures in nuclear facilities and bridges. Ensuring the integrity and operational safety of structures during their lifetime is an important task. Nondestructive methods and structural health monitoring can be used to detect damage caused by SCC and ASR instead of conventional visual inspection. Among the nondestructive methods, acoustic emission (AE) is a suitable method because it is extremely sensitive to the initiation and propagation of the damage in materials. Detection and localization of damage in structures can be achieved by deploying a network of AE sensors. However, the complexity of real-world structures is a challenge for the application of AE. In some cases, the area available for sensor attachment is limited. It would be challenging to deploy sensor arrays. An approach using a single AE sensor may be beneficial for damage detection in complex infrastructure system.

The purpose of this dissertation is to investigate the intelligent damage detection and localization approach for infrastructure system such as spent nuclear fuel storage containers and concrete bridge components leveraging deep learning techniques and a single AE sensor. In addition, a novel transfer learning approach for damage localization without labelled historical signals for training is proposed. The finite element model is developed to generate numerical AE signals for training the supervised learning model. Unsupervised domain adaptation technology is used to reduce the difference in distribution between the generated numerical AE signals and the realistic AE signals.

The results suggest that the intelligent approach using a single AE sensor and deep learning techniques has a good performance. The transfer learning approach is able to localize AE signals with high accuracy without using labelled training data, demonstrating that it could be a potential approach to localize ASR and SCC events on infrastructure systems. However, further research is needed to standardize the method for field application.