Machine learning driven structural health monitoring
View/
Keywords
Structural health monitoring ; Composite materials ; Physics enhanced machine learning ; Damage prediction ; Machine learning ; Deep learning ; Physics-informed feature engineeringAbstract
The goal of this diploma thesis is to develop a general methodology that uses engineering
knowledge to predict damage in composite materials for aeronautical structures. In more
detail, machine learning and deep learning algorithms are implemented to classify the defect
modes and predict their severity, under realistic limitations, with scarce and limited data.
This study begins by analyzing theoretical aspects of structural health monitoring and how it
can benefit from machine learning and also analyzes simulations conducted on composite
plates and its engineering background. These simulations help in data acquisition and they
provide signals that passed through the damaged plates, which are the training features and
their corresponding damage severity values and defect mode classes, which are the target
values. Then, the data handling and feature engineering methodologies, that are based on
engineering knowledge, are discussed and highlight how beneficial they are especially with
small datasets and the chosen algorithms are explained. At this point, common algorithms
are implemented to tackle the damage severity prediction task as a regression problem and
the defect mode prediction as a classification problem. Multiple external factors have been
introduced, to add difficulties, which occur in practice and make the experiments realistic.
Last the results of the machine learning experiments are displayed and they are generally
positive but specifically the results of the simpler algorithms surpass the results of the
complex ones, proving that knowledge application, enhances performance. This study
confirms that combining machine learning with structural health monitoring can provide
very insightful results about the structural condition of a plate, without the need of large
datasets and complex methodologies.
Postgraduate Studies Programme
Τεχνητή Νοημοσύνη - Artificial IntelligenceDepartment
Σχολή Τεχνολογιών Πληροφορικής και Επικοινωνιών. Τμήμα Ψηφιακών ΣυστημάτωνCorporate Department
National Center of Scientific Research "Demokritos"Number of pages
47Language
EnglishCollections
Except where otherwise noted, this item's license is described as
Αναφορά Δημιουργού-Μη Εμπορική Χρήση-Όχι Παράγωγα Έργα 3.0 Ελλάδα
Αναφορά Δημιουργού-Μη Εμπορική Χρήση-Όχι Παράγωγα Έργα 3.0 Ελλάδα