Surrogate modeling of high-dimensional vibroacoustic problems using parametric model order reduction

The field of structural dynamics has already seen tremendous potential for approaching large-scale computations using (parametric) model order reduction. The various advantages include large savings in computation time and resources without compromising on accuracy thereby replacing the expensive solvers with easy to compute reduced order models. However, there are domain-specific challenges when it comes to using the techniques for practical applications. For vibroacoustic models, high-dimensional models with a large number of inputs-outputs, frequency-dependent nature of the system and a high number of parameters considered for variation, require special attention. This dissertation focuses on handling these various challenges to produce reduced order models efficiently. Another focus of the work is to enhance existing reduction algorithms for robustness and adaptivity, especially while reducing problems exhibiting high dynamics such as systems with fluid-structure interaction. The thesis addresses model intrusive projection-based Krylov subspace reduction techniques to obtain an accurate reduced order model in the frequency domain. The proposed techniques have shown to be robust for the challenging problems from vibroacoustics delivering promising potential for practical applications. With the help of representative examples from vibroacoustics including a complex aircraft fuselage model used for cabin noise assessments, the efficiency of the proposed methods is showcased. Furthermore, to address the challenges that arise while handling high-dimensional parametric problems, the method of active subspaces with extension to clustering techniques and the method of spatially-adaptive sparse grids has shown good results. Accordingly, an interpolatory approach using interpolation of underlying subspaces thereby yields accurate reduced order models in the parameter domain. Finally, the thesis offers extensive tools and recommendations to efficiently reduce vibroacoustic models with the aforementioned challenges. To increase the quality and credibility of the presented computational solutions, the code implementation part of the entire work is developed and published according to sustainable research software development principles.