Steam generators (SGs) in nuclear pressurized water reactors play an important role as heat exchangers between the primary and secondary circuits. Over time, SGs experience various complex degradation mechanisms, including clogging, a physical phenomenon that occurs gradually during operation. Clogging can impact SGs in multiple ways, such as altering heat transfer efficiency between the circuits or causing vibrations due to flow redistribution. Clogging data for specific SGs is typically gathered through televised camera inspections, conducted during reactor outages, leading to expensive and limited data. Costly chemical cleanings are performed in case the clogging rate is found to be above a certain threshold. For a better predictive maintenance and health management policy of its assets, EDF has set the long-term goal to develop digital twins of high-stake components such as the SG. As an initial step toward the digital twin, EDF R&D has developed a physical model and the computational code THYC-Puffer-DEPOTHYC [1,2] to simulate clogging kinetics over long operational periods (1 run of the code amounts to 5h of high-performance computing time). We present an uncertainty quantification methodology applied to this computationally intensive code [3], using polynomial chaos expansion and Gaussian process surrogate models to perform global, target and conditional sensitivity analyses based on Sobol’ indices and more advanced kernel-based dependence measures (typically, the Hilbert-Schmidt Independence Criterion). This study allows us to assess the impact of various uncertain input variables on the dispersion of the clogging rates over the simulation time and to uncover the influential variables on the degradation phenomenon as well as to estimate the remaining useful life (RUL) distribution in relation with conservative thresholds for the clogging rate. To further improve the predictive accuracy of the model, we propose a hybrid approach that applies Bayesian calibration [5] to a physical parameter of the model that drives the clogging kinetics. This approach leverages various types of information (clogging field data, fictitious but realistic data generated from statistical regression models and simulations from the original computational code and derived surrogate models), providing more accurate estimates of the SG's remaining useful life, thus supporting prognostics and maintenance planning—key features of the digital twin.