Multifidelity Model–Assisted Knowledge Transfer Optimization Method for Computationally Intensive Solid Rocket Motor Design

Solid rocket motor design based on high-fidelity simulation models is extremely time-consuming in practical engineering problems. To improve the efficiency of solid rocket motor design problems, this study proposes a multifidelity model–assisted knowledge transfer optimization (MFM-KTO) method. The similarity measurement method is proposed to select several similar tasks and a multifidelity model is then constructed using the knowledge extracted from these tasks. The local density enhanced radial basis function method is employed and the hyperparameters are trained based on the source knowledge to improve the generalization performance on the target model. A two-dimensional benchmark function and two engineering applications are utilized to test the proposed MFM-KTO method and popularly used sequential approximation optimization (SAO) method. The results indicate that the MFM-KTO method converges after several iterations, while the SAO method for comparison needs multiple computation costs before locating the optimum, which proves that the proposed MFM-KTO provides an effective knowledge transfer optimization method to utilize previous experience and successfully improve the design efficiency compared to normal SAO method.