Transfer learning has promised to generalize structural health monitoring (SHM) of bridges, as it permits one to reuse long-term monitoring data across similar structures. Studies have been published using numerical and monitoring data from bridges sharing global similarities. This paper presents the first application of unsupervised transfer learning between twin concrete bridges. The two bridges are located side-by-side, but their construction time is separated by almost three decades. This paper proposes a framework to reuse monitoring data in the undamaged condition from the old bridge to address data scarcity and uncertainty in the training of machine learning algorithms for SHM of the new bridge. To deal with the scarcity of data, a numerical model is developed to simulate the undamaged condition of the new bridge. The model is calibrated using Bayesian inference through Markov-Chain Monte Carlo simulations with the Metropolis-Hastings algorithm. To deal with sources of uncertainty, a transfer learning is used to perform domain adaptation of data sets from both bridges. The results show the numerical model is capable of simulating the dynamics of the new bridge and transfer learning is capable of adapting the distribution domain of the data from the old bridge in such a way that it can be reused to train machine learning algorithms to classify observations from the new bridge.