Incremental Decoupling Neural Architecture for Wide‐Rate Nonlinear Viscoelastic Damage Modeling of HTPB Propellants via Data‐Model Synergy

Solid propellants have been identified as nonlinear viscoelastic materials. However, existing nonlinear constitutive models have failed to simultaneously characterize stress‐strain relationships under both low and high strain rates, and their accuracy has remained insufficient. To address these limitations, an Incremental Form Separated Residual Deep Neural Network (IFSR‐DNN) has been proposed as a nonlinear constitutive surrogate model for hydroxyl‐terminated polybutadiene (HTPB)‐based solid propellants. By decoupling viscous and elastic components based on the generalized Maxwell model, dedicated viscous subnetworks and elastic subnetworks have been designed. Further integrating Schapery's damage‐inclusive nonlinear viscoelastic theory, a softening function subnetwork has been developed, and these subnetworks have been systematically coupled. A subnetwork‐wise training strategy has been implemented, embedding physics‐based loss terms derived from discrete integral incremental formulations. The IFSR‐DNN has proven capable of effectively replacing conventional constitutive models across a wide strain‐rate range. Validations of both subnetworks and the integrated network have demonstrated their ability to capture relaxation effects, rate dependence, and high consistency with experimental data. The error of the linear viscoelastic component has been quantified as 10 −15 , while the total network error compared to experimental results has been reduced to 1.3%. This model has innovatively combined data‐driven methodologies and theoretical frameworks, offering a novel paradigm for damage‐integrated constitutive modeling of solid propellants.