fig2

Figure 2. The architecture of the mechanics-informed deep learning constitutive model. The overall architecture illustrates input/output layers, the mechanics-informed layer integration, and the data flow from strain inputs through GRU, MHA, and FFN modules to stress and microstructural outputs. $$ \dot{\varepsilon} $$: strain rate; ε_t: strain; σ_t: stress; ρ_t: dislocation density; f_twin: twin volume fractions; GRU: gated recurrent unit; MHA: multi-head attention; FFN: feed-forward network.