Anomaly and damage detection for structural health monitoring based on computational intelligence

  • Anomalie- und Schadenserkennung für die kontinuierliche Zustandsüberwachung basierend auf künstlicher Intelligenz

Hesser, Daniel Frank; Markert, Bernd (Thesis advisor); Fritzen, Claus-Peter (Thesis advisor)

Aachen : RWTH Aachen University (2022, 2023)
Book, Dissertation / PhD Thesis

In: Report. IAM, Institute of General Mechanics IAM-16
Page(s)/Article-Nr.: 1 Online-Ressource : Illustrationen, Diagramme

Dissertation, Rheinisch-Westfälische Technische Hochschule Aachen, 2022


Modern structures, systems and processes are required to fulfill high standards regarding safety, reliability and availability. One key element represents structural health monitoring (SHM), where a continuous datastream will be acquired in-service. As a result, a condition-based and predictive maintenance concept can be introduced by analyzing the recorded data, which provides ongoing information on the health state. What may sound simple and straightforward, is in reality associated with some pitfalls and challenges in the implementation process. Besides certification and compliance issues, the application of SHM is determined by the received sensor data and the interaction between the structure and the environment. Due to this interaction, the raw signal will always include scattering and noise, which might mask the presence of damage and therefore, has to be processed and filtered efficiently. In this doctoral thesis, the application of computational intelligence for different SHM applications is presented and discussed, namely acoustic emission (AE), tool wear monitoring and infrastructure monitoring. Methods from the field of computational intelligence present a promising tool to analyze the raw data and extract unique signal characteristics that are correlated to the specific damage cases. The goal of the present work is to provide guidance for new engineering applications by introducing the fundamentals of sensor technology, signal processing and computational intelligence. Three use cases provide a descriptive and detailed discussion on the application of such methods. AE sources in thin-walled structures have been analyzed by means of computational intelligence in order to predict the location and type of source. Tool wear has been monitored in a CNC milling machine based on the collected data during the manufacturing process. Furthermore, the infrastructure monitoring has been implemented for a suspension railway, which evaluates the health state of the infrastructure by analyzing the data that is collected in a bypassing vehicle. Depending on the application, physical experiments and/or numerical simulations have been performed to collect a diverse database. The raw data has been preprocessed in time or frequency domain, which involves methods, such as the wavelet transformation. Various machine learning architectures have been selected to address the corresponding SHM problem and to evaluate the health state. Furthermore, supervised and unsupervised algorithms have been implemented, which for example include artificial neural networks and one-class support vector machines. Each use case is discussed in terms of optimization and performance. The results prove that computational intelligence is a promising method to perform SHM in real-world applications in order to ensure a high safety, reliability and availability of the system. Furthermore, the proposed work provides helpful insights that can be easily applied for various damage cases and structures and is supported by the strongly growing field of computational intelligence and big data analysis.