Application of a hybrid of particle swarm and genetic algorithm for structural damage detection

Abstract

This study presents a novel optimization algorithm which is a hybrid of particle swarm optimization (PSO) method and genetic algorithm (GA). Using the Ackley and Schwefel multimodal benchmark functions incorporating up to 25 variables, the performance of the hybrid is compared with pure PSO and GA and found to be far superior in convergence and accuracy. The hybrid algorithm is then used to identify multiple crack damages in a thin plate using an inverse time-domain formulation. The damage is detected using an orthotropic finite element (FE) model based on the strain energy equivalence principle. The identification is carried out using time-domain acceleration responses. The principle is to minimize the difference between the measured and theoretically predicted accelerations. Since the computational effort of identifying the use of global FE model proved prohibitive, a quarter substructure was identified which contains 72 damage variables. Using numerically simulated experiments, three cracks in a plate were reliably detected using this method in the presence of noise. While the pure particle swarm algorithm proved to be fast, the hybrid algorithm proved to be more accurate in damage prediction. GA performed worst in speed and accuracy.