Multi-fidelity robust design optimisation for composite structures based on low-fidelity models using successive high-fidelity corrections

Multi-fidelity probabilistic optimisation of composite structures under thermomechanical loading using Gaussian processes

“…The equivalent high-fidelity FEM simulation number to this model is around 80, which is twice more computationally expensive than the proposed multi-fidelity method. This proposed multi-fidelity method is still more efficient than the multi-level multi-fidelity modelling approach developed before [37]. The multi-level multi-fidelity approach uses both 20 HFM training points and 110 LFM training points, which equates to around 55 high-fidelity FEM simulations; hence the proposed multi-fidelity method enables 30 % of computation time savings.…”

“…The developed multi-fidelity formulation adapts the concept of multi-level optimisation to improve computational efficiency compared to other multi-fidelity methods. This adaptation ensures that the multi-fidelity formulation can offer significant computational time savings while safeguarding the optimisation process against different levels leading to information loss [37]. Multi-level optimisation presents a wide range of benefits when it comes to large-scale optimisation problems.…”

“…In particular, the multi-fidelity approach is integrated with multi-level optimisation to manage large-scale problems. Yoo et al [37] have suggested a multi-level multi-fidelity optimisation approach. The multi-level multi-fidelity formulation separately creates two surrogate models of the HFM and the LFM with different numbers of design variables.…”

“…The LFM does not require extra FEM simulations to build up a new training dataset for the next level. This can provide additional computational gains compared to the multi-level multi-fidelity method that should establish the low-fidelity training dataset for each level [37]. It is not surprising that even low-fidelity FEM simulations can cause a computational burden when a problem is complex and large-scale.…”

Design and optimization of self-deployable damage tolerant composite structures: A review

“…With this approach, the authors were able to optimize the ply orientation and thickness of the laminate while considering: uncertainty in both random design variables and random parameters (robustness assessment) and uncertainty on the individual mechanical properties of each laminate (reliability assessment). More recently, Yoo et al (2021) [282] proposed a novel multifidelity modelling-based optimisation framework, aimed at the robust design of composite structures. Through the use of high-fidelity model for exploitation and low-fidelity model for the exploration of the design space, the authors were capable of achieving computational time savings of at least 50% compared to conventional multi-fidelity and high-fidelity modelling methods.…”