Global damage detection techniques based on natural frequencies and mode shapes have proven ineffectual in detecting damage smaller than 10 percent of the surface area of the structure and near-surface damage. A damage index involving a convolution product of the difference in the layer-wise in-plane modal strains between a reference and damaged states of a structure was developed to qualify damage. This is an inverse multi-modal problem, associated with a large number of local optima. To further complicate this problem, both continuous and discrete design variables need to be adopted. A tailored genetic algorithm has been used in conjunction with an artificial back-propagation neural network, which is used as a function approximator with satisfactory accuracy to determine the damage index response of the delaminated laminate with various de-lamination patterns.