The shock-structure solution of the hyperbolic balance laws is analyzed using Physics-Informed Neural Networks (PINNs). The usefulness of the present approach is demonstrated in the \(2 \times 2\) hyperbolic dissipative model, which satisfies all the requirements of rational extended thermodynamics. First, we classify the possible regions of the sub-shock formation in terms of the velocity of the shock wave and the field value evaluated in the unperturbed equilibrium state. Each region represents the number of possible sub-shocks: no sub-shocks, one sub-shock, or multiple sub-shocks. Second, we propose optimizing the PINNs architecture based on the classified region and show that PINNs can capture the shock structure with sub-shocks using multiple neural networks. The developed methodology can be applied to all systems in rational extended thermodynamics and is expected to play an important role in the study of sub-shock formation.
