Аңдатпа
Hydrogel-based lung sealants are optimized largely based on formulation and post-cure mechanics, even though deformation history during delivery and placement plays a critical, yet under-characterized, role in determining in situ performance. This disconnect is particularly consequential on the application of lung sealants, where excessive stiffness promotes delamination while insufficient stiffness compromises airtight sealing. Here, we establish a quantitative process–structure–property framework for a gelatin–tannic acid–transglutaminase lung-mimetic sealant by integrating time-resolved rheology with physics-informed machine learning. Using small-amplitude oscillatory shear and steady torsional flow at 37 °C, we show that gelation follows a reproducible kinetic clock (storage modulus, G′– loss modulus, G″, crossover at ~100 s; modulus plateau by ~600 s), while the attainable viscoelastic state is strongly governed by deformation history. Increasing oscillatory strain amplitude from 0.1 % to 50 % suppresses network maturation, reducing the post-gelation storage modulus, G′, from ~10⁶ to ~10⁴ Pa, whereas sustained steady shear (0.01–1 s⁻¹) decreases magnitude of complex viscosity by 2–3 orders of magnitude and permanently downshifts elastic moduli, G′–, from ≈3.5×10³ to ≈4×10² Pa. These modulus ranges span the effective compliance of lung parenchyma under physiological tidal strain, delineating mechanical regimes associated with airtight sealing, strain accommodation, or premature interfacial failure. Controlled aeration during the processing of the sealant further decreases stiffness by ~30 % without altering gelation kinetics, providing an additional, physically interpretable compliance-tuning mechanism. To unify these effects, we introduce a dimensionless Degree of Gelation (DoG), serving as a rheological state variable that collapses oscillatory and steady-shear histories into a single, time-resolved descriptor of network evolution. Machine-learning models trained on experimentally accessible inputs (time, strain amplitude, shear rate, frequency, aeration) accurately predict DoG (R² ≈ 0.9) and, in inverse mode, identify handling conditions required to achieve targeted in situ mechanical states. This rheology–machine-learning framework reframes lung sealant development from a static materials optimization problem to a controllable, process-driven design strategy. By quantitatively linking applicator-level parameters to failure-relevant mechanical outcomes—airtightness, compliance, and resistance to delamination—it provides a mechanistic and generalizable foundation for the design of injectable hydrogels, bioadhesives, and tissue-interfacing