Reducing time and memory requirements in topology optimization of transient problems (Q6592356)

This paper deals with strategies to reduce time and memory requirements in topology optimization of transient problems. A total of five algorithms are examined, which are: Global-in-Time (GT), Checkpointing (CP) and Local-in-Time (LT), in addition to the new ones Hybrid (CP/LT) and Parallel-Local-in-Time (PLT) introduced by the authors. In particular, the CP/LT algorithm introduces an error measure and corrections for the LT algorithm, whereas the PLT algorithm parallelizes the optimization process by decomposing the time domain. The algorithms are compared in the context of transient thermal and flow optimization problems in two spatial dimensions, which are solved with the help of the SIMP method. The performances of the algorithms are discussed in terms of memory requirements, computational cost, and stability, exposing their strengths and weaknesses. A guideline for selecting the appropriate algorithm is summarized through a comprehensive decision tree, which is useful for practical applications and very welcome. Finally, some research directions are suggested in order to improve the performance of the proposed algorithms.