Simulated Annealing (SA) is a widely used optimization technique in which the temperature parameter T plays a crucial role in controlling the acceptance probability of non-improving trial solutions (negative changes) during the search process. Traditionally, the initial value of T is determined by the user, often through trial and error, which can be inefficient and lead to suboptimal performance. In this study, we present an automated method for calibrating the T parameter based on the statistical behavior of the objective function. By leveraging the same sequence of changes (both favorable and unfavorable) generated by the SA algorithm, our method dynamically computes an appropriate initial T value, eliminating the need for manual tuning. Extensive experiments on benchmark and real-world optimization problems demonstrate that our approach yields good results. In particular our approach is successfully applied to the topological optimization of discrete structures made of prefabricated or 3D-printed members.