Risk-averse approach for topology optimization of fail-safe structures using the level-set method (Q2667305)

This paper is concerned with the derivation and numerical evaluation of a level-set method in the context of topology optimization for structures under uncertain failures. The authors inspect the linearized elasticity equation \[ -\nabla\cdot \sigma(u(x,\omega)) = b(x) \quad \text{in } D \times \Omega\tag{1} \] with uncertain Dirichlet and Neumann boundary conditions in a spatial domain \(D\subset \mathbb{R}^2\) and sample space \(\Omega\) of a probability space. The random variable \(\omega\) is assumed to follow a discrete probability function. The topology optimization is then performed by minimization of the cost functional \[ \min_{\chi_\mathcal{O}\subset \mathcal{U}_{ad}} \mathcal{J}(\chi_\mathcal{O},\omega) = \int_D \chi_\mathcal{O} b u~dx + \int_{\Gamma_N} \bar{t}u~ds, \] where \(\bar{t}\) represents the body and surface forces applied in normal direction of the boundary and \(\mathcal{U}_{ad}\) is the set of admissible shapes. The uncertainty of the optimization problem is investigated by comparing multiple variants of the cost functional, namely the worst case formulation (where the worst failure is optimized), its smoothed approximation with a penalized exponential-logarithmic transformation, the excess probability (the probability of the performance exceeding a given threshold \(\eta\)) and the expected excess (expected value of the highest overshoot of the cost function over the threshold \(\eta\)). To tackle the problem numerically, following [\textit{S. Osher} and \textit{J. A. Sethian}, J. Comput. Phys. 79, No. 1, 12--49 (1988; Zbl 0659.65132)], it is approximated with the level-set method, which is a novel approach in this area. The system is therefore transformed into a suitable reaction-diffusion equation which is then solved using a semi-implicit time integration and finite elements. In total, three numerical experiments are conducted, investigating the performance of excess probability and expected excess compared to the worst case formulation. ``The numerical results included show that the proposed risk-averse formulations yield redundant structures which are less sensitive to inherent losses of stiffness resulting from possible failures. Subtle differences have been reported from the results obtained by means of the two risk-averse functions. On the one hand, the excess probability formulation permits to minimize the number of scenarios with compliance above a given threshold, but does not take into account the possible excess or shortfall with respect to the threshold, yielding (few cases) with considerably larger values than the threshold. On the other hand, the expected excess formulation does take into account the amount of degradation of the structural performance, or equivalently, the excess with respect to the specified threshold value. Furthermore, the numerical examples reveal that the worst-case approach yields a topology very similar to that of the expected excess formulation for high values of the threshold. However, in contrast to the worst-case formulation, which penalizes all the damage cases regardless of their probability of occurrence, the expected excess formulation allows designers to assume an a priori level of risk by means of the use of a threshold in the structural performance.'' (as written in the original manuscripts conclusions)