Betrouwbaarheidsgebaseerd ontwerp van gewikkelde composietdrukvaten: integratie van meerschalige ruimtelijke materiaalvariabiliteit

Hydrogen gas is a promising clean energy carrier in a rapidly decarbonizing global economy. Storage remains difficult and typically occurs in high-pressure vessels made of carbon fiber-reinforced composites. The use of this high-performance material keeps vessel weight manageable, but comes at a significant cost that hinders the broad adoption of composite pressure vessels in industry. Reliability-based design methods offer the potential to reduce the composite thickness, and thus manufacturing cost, while guaranteeing vessel safety. Such a reliability-based design procedure faces several challenges: Spatial variability significantly increases the input parameter space of the numerical simulation, rendering efficient uncertainty propagation difficult. This issue is exacerbated by the multiscale nature of composite materials; vessels are measured in meters but individual reinforcement fibers are only several micrometers wide. Homogenization of the microstructure is required to provide effective material properties for the full vessel simulation. A substantiated analysis should also explicitly simulate the material variabilities that cause vessel performance variability. This requires that both their physics are well-understood and that sufficient measurement data are available. Literature currently lacks a composite pressure vessel design technique that simultaneously tackles these challenges. This dissertation develops a nondeterministic simulation methodology for the reliability-based design of filament-wound composite pressure vessels. To this end, it develops an efficient multiscale uncertainty quantification approach that incorporates spatial variability of fiber misalignment, fiber volume fraction and fiber strength using the concept of random fields. A vine copula-based metamodeling methodology is leveraged to reduce the computational cost of homogenization. It bypasses the "curse of dimensionality" that typifies the large input parameter space of spatial variability by not relying on spatially smooth input and output responses, at the expense of losing spatial information on the output response. Verification results on a unidirectional tensile coupon show that, for the problem at hand, the metamodel reduces time spent on the homogenization step a hundredfold. The approach also accurately predicts the failure stress at a significantly lower computational cost. The multiscale uncertainty quantification framework is validated by predicting the burst pressure of a small composite pressure vessel. The workflow predicts a mean burst pressure of 1504 bar, whereas the mean experimental burst pressure equals 1597 bar, a difference of 5.8%. An investigation into the causes of this discrepancy reveals that fiber strength parameter uncertainty and fiber break model uncertainty are both significant contributors. The validated methodology is used to minimize the composite layup thickness of a small composite pressure vessel for a certain burst reliability target. The optimization strategy reduces the layup thickness from a nominal baseline 10.692 mm to 7.776 mm, a ratio of 27.3%, without compromising on the burst reliability of the vessel.