Exploration of Particle Swarm Optimization for Parameter Identification in Phase-Field Modeling

To address the issue that phase-field (PF) simulation parameter calibration has long relied on manual experience and comparisons with static final-state results, a frame-by-frame parameter identification framework was developed. Three strategies were designed: standard Particle Swarm Optimization (PSO), a hybrid PSO incorporating elitism and random reset mechanisms, and Elitist Random Search (ERS). A composite cost function was established, which combined the Dice coefficient and Chamfer distance through a weighting ratio. Hybrid PSO converged earlier and achieved a superior final global optimum cost, demonstrating that this strategy effectively mitigated the risk of entrapment in local optima. Finally, hybrid PSO went through extended validation with longer duration (15s) and larger simulation scale (400 µm × 1700 µm), the average error of the key parameters identified is approximately 2.8%. This work combined frame-by-frame dynamic comparison with hybrid PSO for PF parameter identification for the first time, fundamentally resolving the long-standing problem of degradation in this field.