<p>Water quenching is a widely employed heat treatment technique to produce high-quality metallic components with desired properties. It is crucial to model the temperature distribution and assess its impact on residual stress and distortion to ensure the quality of quenched parts. However, quenching is a complex, multi-scale, and multi-physics problem involving many interplay phenomena, such as rapid evaporation, condensation, and thermal-mechanical interactions. The physical complexity makes developing an accurate and efficient quenching model to achieve this objective a significant challenge. This paper presents a coupled data-physics thermo-mechanical simulator (DPTMS) for quenching processes. DPTMS is built on a data-physics coupling framework, which leverages physics-informed machine learning and finite element method to achieve accurate temperature prediction and thermo-mechanical analysis. Firstly, a physics-informed machine learning model is developed to quickly reconstruct the full-field temperature profile from limited temperature monitoring data. In the machine learning temperature model, we apply a re-combination method to reorganize monitored temperature data to remedy the challenge of data scarcity. Following this, we develop a machine learning framework using appropriate neural network architectures and inherent physics laws to restructure the 3D temperature profiles. Subsequently, the machine learning-based temperature model is seamlessly integrated into a parallelized finite element-based thermo-mechanical model using FEniCS to predict residual stress and distortion. The accuracy of this coupled machine learning and FEniCS approach are validated with high-fidelity simulation and experimental data. The comparison with existing models in terms of accuracy and efficiency is presented to show the superior performance of the proposed DPTMS.</p>