High performance model order reduc2on techniques in computa2onal mul2scale fracture

J. OLIVER,; CAICEDO, M.; A E. HUESPE; LLOBERAS-VALLS, O.

nantes

Conferencia; CFRAC 2017 International Conference on Computational Fracture and Failure of Materials and Structures; 2017

ECCOMAS

Mul9-scalecomputa9onalhomogeniza9onmethods(FE2)offerahugepoten9altoincorporate,tomacroscopicpropaga9ngfracturemodeling,thecomplexityof the material meso/micro-structure morphology, through simple and physicallyrealis9c models. However, they presently suffer the burden of their enormouscomputa9onal cost, since the computa9onal complexity (number of the involvedopera9ons)hasamul9plica9vecharacterintermsoftheinvolvedsamplingpointsateveryscale.In this context, a powerful, and conceptually simple but efficient, mul9scalecomputa9onalmodel for fracture has beenrecently presented bythe authors in [1]andsuccessfullyappliedtoanumberof2Dbenchmarks.Inthiswork,twocombinedtechniquesareexploredandappliedtothatmodel:1)low-dimension-spaceprojec9onof the solu9on (reduced order modeling, ROM) and 2) reduced op9mal quadrature(ROQ).Theyleadtoahighperformancereducedordermodel(HPROM)oftheoriginalhigh fidelity (HF) mul9scale fracture problem [2], which drama9cally diminish thecomputa9onal demand, at the cost of a very small addi9onal error. A number ofrepresenta9ve simula9ons in 2D problems show high resul9ng speedups and linearscalability in terms of the problem complexity, which make think of the HPROMtechniques as a possible way to make computa9onal mul9-scale computa9onalfractureavailablefordaily-lifeengineeringproblems.