22 March 2023
Structural integrity and durability of aircraft structure
Subject: Solid Mechanics
Lecturer: Zlatan Kapidzic, IEI
Location: ACAS, A Building, Campus Valla
Structural strength, durability and damage tolerance are important issues for aircraft flight safety and economic life. To achieve and maintain adequate levels of these properties during the entire service life of the aircraft, the manufacturer performs a structural integrity assessment program. The program addresses all aspects that affect aircraft structural integrity, such as material properties, manufacturing, function, loading, operational conditions and maintenance. In this context, understanding the structural and material behavior under loading and the ability to predict them is essential. Traditionally, this knowledge has largely been based on experimental and development testing and today numerical simulations are increasingly used also. A large part of contemporary research is directed towards development of material modelling and simulation technique. As the requirements for light weight in aircraft structures continue to increase, so does the need to consider new materials and structural solutions and to understand their properties, function and the effects it has on service life and structural integrity. This lecture will give an outline of the aircraft structural integrity program and show some fatigue and damage tolerance related issues in fighter aircraft structure. Typical failure behavior of composite and metal will be presented and how they can be modelled in a hybrid, bolted and integrated structures. Some highlights of the resent research and an outlook towards the future trends will be given.
1 Mars 2023
Image synthesis and augmentation for data-centric machine learning
Subject: Visualization and Media Technology
Lecturer: Gabriel Eilertsen, ITN
Location: K2, Kåkenhus, Campus Norrköping
Machine learning (ML), especially by means of deep learning, has made substantial progress over the last decade, e.g., for solving complex problems such as image classification, medical diagnosis, personalized recommendations, and natural language processing. However, the data-hungry nature of deep learning means that the full potential of a model is often inhibited by lack of data. Deep learning so far has to a large extent been model-centric, and the general view on data has been "more is better". In most situations there could be larger gains in improving the training data, in a data-centric formulation. In imaging applications, one promising technique for expanding and improving training data is through image synthesis. With the advancements of generative deep learning, there are many possibilities for data-centric ML in the intersection between computer graphics and deep generative modeling.
This lecture will introduce data-centric ML and image synthesis, and highlight some projects within computer vision and computer graphics where data has been in focus. This includes, e.g., data augmentation techniques for deep high dynamic range image reconstruction and for self-supervised learning, as well as deep generative modeling for data augmentation, anonymization, and anomaly detection. One of the main application areas has been medical imaging, and in particular digital pathology. Here, data-centricity and image synthesis is especially promising since data is expensive to capture, relies on medical expertise for annotation, and is of sensitive and protected nature.