A comparative study of optimization techniques for tuning a finite element model of the lung to biomechanical data

A comparison between two stochastic optimization techniques, adaptive simulated annealing (ASA) and the multi-island genetic algorithm (MIGA) is conducted to investigate which is the more attractive option for tuning a finite element model to match experimental data. The study probes the repeatability, robustness and sensitivity of each algorithm. While both algorithms produced FE results within one standard deviation of the experimental mean, the ASA approach proved more effective than the MIGA. The ASA algorithm demonstrated better repeatability after 1000 trials, with an average parameter value change of 9%. The ASA algorithm also probed a wider range of the solution space and produced results with significantly lower sum-of-squares error than the MIGA (t-test, one tailed, p < 0.001).