The need for efficient and low-cost techniques adequate for damage detection has become of great interest in engineering applications where structural health monitoring (SHM) is of paramount importance. Promising algorithms for SHM have to deliver results with very low computational and response time requirements and be trustworthy within a certain accuracy. Different algorithms (artificial neural networks (ANN), response surface methodology (RSM), and optimization techniques -gradient-based local search (GBLS) and nondominated sorting genetic algorithms (NSGA-II)) are proposed to fill this research gap. The concept of a surrogate model as a fast-executing model is also introduced. Because the objective of this paper is to concentrate on viable techniques suitable for damage detection using vibration methods with very low computational requirements, surrogates are therefore employed to curtail the computational expense. Particularly of interest among the proposed algorithms is RSM, the principle of which has proved successful in the pharmaceuticals industry over the years. However, RSM has not been so widely used in the field of structural engineering for delamination detection. In this paper, we have demonstrated that a fourth-order polynomial has the capability to detect delaminations in composite structures. In order to reduce the size of training data required to solve the inverse problem by the proposed algorithms, the idea of a suitable design space is brought to the limelight as the combination of all possible simulations that one is concerned about. Since the overall sum of design space is usually prohibitively large, we have used K-means clustering to effectively achieve this. This research concerns the application of ANN, RSM, and optimization techniques for delamination detection using changes in natural frequencies before and after damage. Efficiencies of algorithms (ANN, GBLS, and NSGA-II) are compared with the developed RSM models in terms of the accuracy of delamination detection and response time requirements. The methods have been shown to compete effectively for delamination detection and are accurate in detecting the size and locations of delaminations at midplanes. RSM has a unique feature in that it produces models with a small training dataset requirement and also generates mathematical models that are easy to interpret and implement. The optimization techniques, when integrated with surrogate models, require small training sets clustered through the entire design space. ANN, however, requires large training datasets to achieve its results. As such, the potential of these algorithms as tools for on-board damage detection when integrated into a SHM system is successfully demonstrated.