Theoretical background of the thesis

Nonlinear Structural Mechanics: Focus on Plasticity

The accurate simulation of structural components under real-world loading conditions often requires accounting for behaviors that extend beyond simple elasticity. Many engineering structures, especially those made from metals and alloys, exhibit distinct nonlinear characteristics such as yielding and permanent deformation. This phenomenon, broadly referred to as plasticity, is particularly significant when assessing a structure’s safety and performance under extreme loads, such as during accidental overloads or impact events. Capturing these effects within computational mechanics frameworks calls for mathematical models capable of describing not only the onset of plastic flow but also the subsequent material response, including hardening and softening behaviors.

One of the most widely adopted criteria for predicting the onset of yielding in ductile materials is the von Mises yield criterion. This theory postulates that yielding begins when the deviatoric stress in a material reaches a critical value, insensitive to hydrostatic pressure and uniquely suited for metals under complex stress states. To more accurately reflect real-world conditions, the criterion is often coupled with isotropic hardening rules, which model how materials strengthen as plastic deformation progresses. While the von Mises criterion is suitable for a broad range of engineering metals, alternative criteria such as the Tresca and Drucker-Prager models are applied for materials like soil, concrete, or polymers which are some contexts where mechanical behavior is governed by fundamentally different mechanisms.

Integrating plasticity models within the finite element method adds layers of complexity to both the mathematical formulation and the solution algorithm. The underlying system of equations becomes nonlinear, reflecting the intricate feedback between deformation, stress redistribution, and material evolution. Addressing these nonlinearities typically requires iterative linearization techniques. The Newton-Raphson method stands out as a mainstream approach: at each iteration, the nonlinear system is approximated by a linearized version, updated solution increments are computed, and the process is repeated until the solution stabilizes within a specified tolerance. For large-scale or ill-conditioned problems, customized preconditioners further enhance convergence by transforming the equations into more favorable forms.

Moreover, the accurate treatment of nonlinear material behaviors in numerical simulations is indispensable for safety-critical industries such as civil infrastructure, automotive, and aerospace engineering. It enables detailed prediction of failure modes, identification of weak zones, and informed decision-making during design and optimization. The extension of domain decomposition and overlapping Schwarz methods into this nonlinear realm equips engineers with powerful, scalable tools for tackling the increasingly ambitious demands of modern structural analysis, ensuring reliability and efficiency even in scenarios involving substantial, irreversible material changes.

Schwarz Domain Decomposition Methods in Practice

The deployment of Schwarz domain decomposition methods in computational workflows is a process that blends mathematical rigor with practical algorithm design. The first critical step is partitioning the computational domain into overlapping subdomains. These overlaps are intentionally introduced to improve inter-subdomain communication, which is pivotal for rapid convergence, especially in nonlinear settings. For each subdomain, operators called restrictions and extensions are defined: the restriction extracts relevant data from the global domain into the subdomain, while the extension propagates local corrections from the subdomain back into the overall solution space. This framework ensures that local improvements contribute effectively to the global problem.

Within each subdomain, the governing finite element equations are often nonlinear due to material behaviors like plasticity, and are solved using direct or iterative solvers. Direct solvers, such as those based on LU decomposition, offer robustness for small to moderate-sized subdomain systems, while iterative solvers are advantageous for very large or sparse systems. The independently computed corrections from each subdomain are then assembled into a unified update for the global system. This process constitutes a single Schwarz iteration. The global solution is updated, and subsequent iterations follow, exchanging the latest information between neighboring subdomains across overlaps.

A key advantage of additive Schwarz methods is their natural suitability for parallelization. Since each subdomain’s problem can be tackled simultaneously (subject to hardware constraints), these methods are well-aligned with multi-core CPUs and distributed computing environments. Additionally, when coupled with advanced iterative techniques like Krylov subspace methods, such as Generalized Minimal Residual (GMRES), or the Conjugate Gradient Method (CG), these algorithms provide even greater efficiency and robustness. Nonlinear preconditioning, in which the additive Schwarz approach is used to precondition the nonlinear global system itself before linearization, can further speed up convergence in challenging problems.

The practical value of these approaches lies in their flexibility and performance across a wide range of geometries, material behaviors, and problem scales. Whether addressing simple benchmark problems or complex, realistic engineering structures, overlapping Schwarz domain decomposition acts as a force multiplier, allowing practitioners to tackle larger, more intricate models with manageable computational resources. Their implementation in programming languages such as Python, leveraging scientific libraries and mesh generation software, places these powerful techniques within reach for researchers and practitioners across various engineering disciplines.

Bibliography