To meet the increasingly demanding emissions and fuel economy standards, the thermal management of the aftertreatment devices has recently become a priority in powertrain control. In particular, the longitudinal distribution of temperature in the three-way catalyst is a critically important variable to monitor for catalyst light-off control, thermal protection, and for diagnosing aging and degradation. However, such information is typically unavailable in production applications, due to the cost and reliability issues of instrumenting the three-way catalyst with multiple temperature sensors. This work focuses on the development and the experimental validation of a control-oriented, physics-based thermal model of a three-way catalyst for the purpose of real-time temperature monitoring. Starting from the governing equations in partial differential equation form, a model order reduction technique that combines proper orthogonal decomposition and collocation is developed. The sensitivity of the selection of the empirical basis functions is studied. To include the exothermic effect from chemical reactions, a fully connected artificial neural network is trained. The reduced-order model executes more than 100 times faster when compared to the use of standard numerical methods and commercial simulation software such as GT-Power, while providing comparable accuracy. Finally, to verify the proposed methodology, the model is calibrated and validated against experimental data collected by instrumenting a three-way catalyst with multi-point temperature measurements, and installed on an engine test bench.