A robust topology optimization approach is presented which uses the probabilistic first-order second-moment method for the estimation of mean value and variance of the compliance. The considered sources of uncertainty are the applied load, the spatially varying Young's modulus and the geometry with focus on the latter two. In difference to similar existing approaches for robust topology optimization, the presented approach requires only one solution of an adjoint system to determine the derivatives of the variance, which keeps the computation time close to the deterministic optimization. For validation, also the second-order fourth-moment method and Monte Carlo simulations are embedded into the optimization. For all approaches, the applicability and impact on the resulting design are demonstrated by application to benchmark examples. For random load, the firstorder second-moment approach provides unsatisfying results. For random Young's modulus and geometry however, the robust topology optimization using first-order second-moment approach provides robust designs at very little computational cost.

In the current work, a fail-safe optimization of beam structures is carried out. This approach may provide an insight into the robustness of lattice structures. The use of beam elements allows a commonly used engineering approach for obtaining a fail-safe design to be applied. This consists of removing one beam element at a time and optimizing the remaining structure. At the end of the process, the maximum beam radii are used for the final design. This approach is computationally extremely expensive for lattice structures, as it requires one optimization per removed beam. In our contribution, we show that the design obtained from this approach does not actually achieve the desired fail-safe behaviour. We therefore apply a multi-model approach in which the fail-safe requirement is an optimization constraint. This is still computationally demanding and therefore, methods for reducing the number of failure cases to be considered within the optimization are discussed. Furthermore, the p-norm is applied to the stress constraints to reduce the computational effort for the gradient calculation. Reduction of failure cases and use of the p-norm have opposite effects on the conservatism of the result and therefore compensate each other to some extent. Highlights Stress constrained optimization of beam structures. Evaluation of different approaches for fail-safe design. Strategies for efficiency improvement (reducing failure cases/p-norm).

The paper presents a rigorous formulation of adjoint systems to be solved for a robust design optimization using the first-order second-moment method. This formulation allows to apply the method for any objective function, which is demonstrated by considering deformation at certain point and maximum stress as objectives subjected to random material stiffness and geometry, respectively. The presented approach requires the solution of at most three additional adjoint systems per uncertain system response, when compared to the deterministic case. Hence, the number of adjoint systems to be solved is independent of the number of random variables. This comes at the expense of accuracy, since the objective functions are assumed to be linear with respect to random parameters. However, the application to two standard cases and the validation with Monte Carlo simulations show that the approach is still able to find robust designs.

No abstract

Explicitly considering fail-safety within design optimization is computationally very expensive, since every possible failure has to be considered. This requires solving one finite element model per failure and iteration. In topology optimization, one cannot identify potentially failing structural members at the beginning of the optimization. Hence, a generic failure shape is applied to every possible location inside the design domain. In the current paper, the maximum stress is considered as optimization objective to be minimized, since failure is typically driven by the occurring stresses and thus of more practical relevance than the compliance. Due to the local nature of stresses, it is presumed that the optimization is more sensitive to the choice of the failure shape than compliance-based optimization. Therefore, various failure shapes, sizes and different numbers of failure cases are investigated and compared on the basis of a general load-path-based evaluation scheme. Instead of explicitly considering fail-safety, redundant structures are obtained at much less computational cost by controlling the maximum length scale. A common and easy to implement maximum length scale approach is employed and fail-safe properties are determined and compared against the explicit fail-safe approach.