Embedding artificial neural networks into twin cohesive zone models for composites fatigue delamination prediction under various stress ratios and mode mixities

This paper presents for the first time a novel numerical technique for modelling fatigue delamination growth in fibre reinforced composites, which is based on coupling two twin cohesive zone models with a single-hidden-layer artificial neural network. The simulation approach proposed here can describe composites fatigue delamination under negative & positive stress ratios and the full range of mode mixities. In the modelling strategy, each segment of a composites interface is described by two twin cohesive elements, which jointly provide local fracture mechanics parameters into a feedforward single-hidden-layer neural network, without the need to know the global load R ratio. In turn, the neural network algorithm feeds the fatigue crack propagation rate da/dN back into the twin cohesive elements, which follow a static and fatigue cohesive law in a synchronous fashion. The novel modelling methodology has been implemented in an explicit finite element scheme. The modelling strategy is first verified and validated by several benchmark cases, involving mode I Double Cantilever Beam tests, mode II End Loaded Split tests with and without reversal, as well as Mixed-Mode Bending tests. A relevant application of the modelling technique is demonstrated considering a tapered laminate, which experiences non-proportional loading due to the presence of combined static tension and cyclic bending.