Fracture degrades the performance of reinforced concrete structures by reducing the load-carrying capacity. Understanding this fracture behavior is critical for assessing the structural integrity and longevity. Highly time-intensive physics-based simulations for fracture motivate the development of machine learning-based surrogate model. However, most existing models applied to fracture prediction operate as black boxes and require large training datasets for reliable generalization.In this study, we develop a thermodynamics-informed latent space dynamics identification (tLASDI) framework that integrates a Convolutional Autoencoder to extract low-dimensional fracture patterns from high-dimensional damage fields and couples this latent representation with parametric GENERIC formalism-informed neural networks [1] (pGFINN). The tLASDI-pGFINN method enforces key thermodynamic principles such as total energy conservation and entropy generation to ensure accurate and efficient fracture prediction. The proposed model is validated by predicting fracture propagation in concrete under various loading magnitudes and reinforcement bar configurations, including variations in rebar layout, diameter, and spacing. A Neural Network-enhanced Reproducing Kernel Particle Method (NN-RKPM) is employed to generate 2D and 3D datasets, in which the neural network approximation and background RK approximation are coupled via a partition-of unity framework [2]. To evaluate the generalization ability of the tLaSDI-pGFINN, a purely data-driven Convolutional Neural Network LSTM (ConvLSTM) model is used as a baseline for comparison. The results show that the tLaSDI pGFINNaccurately captures the crack propagation under different loading conditions and reinforcement bar configurations and achieves comparable accuracy with a reduced amount of training data compared to the ConvLSTM. This finding indicates the efficiency of the proposed model in predicting concrete fracture behaviour with limited training data, paving new opportunities for broader implementations in reinforcement bar concrete design and analysis.