Delamination detection in composite laminates using improved surrogate-assisted optimization

Abstract Vibration-based delamination detection can be posed as an optimization problem, wherein the discrepancy of the frequency shifts obtained from a numerical model with assumed delamination parameters is minimized from target values of the frequency shifts. Unlike cracks, delamination is characterized by a mix of continuous and discrete parameters. The continuous variables comprise the in-plane locations and sizes, whereas discrete parameters correspond to the interface where the damage occurs. While the recent studies have demonstrated the effectiveness of surrogate-assisted optimization in determining the locations and sizes of the delamination, the prediction of interface has received scarce attention. To improve on this aspect, in this paper, individual surrogate models are built corresponding to each of the interfaces. Furthermore, we also attempt to improve the underlying optimization method by using multiple populations instead of one in order to reduce the likelihood of being trapped in local minima. The performance of the proposed approach is numerically investigated by considering various factors, such as the different number of samples for training the surrogate model, the combinations of different modes of frequency shifts used as input for damage detection, and the addition of artificial noise to the numerical frequency shifts to simulate the unavoidable measurement errors. Experimental investigations are also conducted on six delaminated beam specimens and one intact specimen. The proposed approach showed significant improvements in identifying the damage parameters, and in particular improving the prediction accuracy of the damage interface.