<p>Early identification of structural damage is crucial for reducing maintenance costs and extending the lifespan of engineering structures. This relevant problem is addressed in this work with a novel Population-based MCMC method with a Delayed-Acceptance strategy (PopDA-MCMC) for sampling the joint posterior probability density function of uncertain parameters in a Bayesian damage identification framework. To enhance computational efficiency, a response surface model of the reduced flexibility matrix of the structure is used to construct a computationally inexpensive posterior probability density function for the first stage of the acceptance decision step. Only proposals that pass this initial screening proceed to the second stage, where the true posterior is evaluated using time-domain response data, thereby reducing unnecessary computations. The proposed approach not only accelerates a Population-based MCMC through Delayed-Acceptance but also integrates modal parameters and time response data for damage identification. The RSM training data selection is optimized by combining a relatively small number of cohesion parameters, significantly reducing the number of required simulations compared to traditional design of experiments, such as D-Optimal and Central Composite Design. Numerical results on a beam demonstrate that the PopDA-MCMC method achieves up to a 50% reduction in computational cost compared to a conventional Population-based MCMC method, highlighting its effectiveness in accelerating the sampling process. Beyond damage identification, the proposed method provides a robust framework for addressing general inverse problems of parameter estimation.</p>