Dynamic robust fusion neural network assisted multi-objective optimization framework for scramjet inlet design

•Developed a dynamic robust fusion neural network (DRFN) for high-speed inlet design.•Achieved high-efficiency optimization with improved aerodynamic performance.•Validated DRFN shows higher accuracy and robustness than conventional surrogate models.•Advanced AI-driven framework improves aerospace e...

Full description

Saved in:
Bibliographic Details
Published in:Aerospace science and technology Vol. 167; p. 110691
Main Authors: Ren, Hu, Ma, Yue, Tong, Shuhong, Tian, Ye
Format: Journal Article
Language:English
Published: Elsevier Masson SAS 01.12.2025
Subjects:
Inlet design
Multi-objective optimization
Neural network
Scramjet
Transfer learning
Inlet design
Multi-objective optimization
Neural network
Transfer learning
Scramjet
ISSN:1270-9638
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:•Developed a dynamic robust fusion neural network (DRFN) for high-speed inlet design.•Achieved high-efficiency optimization with improved aerodynamic performance.•Validated DRFN shows higher accuracy and robustness than conventional surrogate models.•Advanced AI-driven framework improves aerospace engineering application capabilities. To address the challenges of designing efficient inlets for aerospace vehicles under varying Mach number conditions, this study presents an adaptive multi-objective optimization framework incorporating a dynamic robust fusion neural network surrogate model. This method leverages advanced machine learning techniques to establish precise mappings between inlet geometry parameters and performance metrics, enhancing both prediction accuracy and decision-making in aerodynamic design. The framework employs the adaptive reference vector guided evolutionary algorithm (ARVEA), benchmarked against traditional optimization methods such as NSGA-II and PSO, demonstrating significant improvements in both computational efficiency and solution quality. Experimental validations across Mach numbers 6 to 8 reveal the framework's capability to achieve smooth transitions and reliable startup at Mach 5, while satisfying stringent static pressure ratio constraints essential for spacecraft and high-speed aircraft performance. Compared to the conventional Busemann inlet design, the optimized configuration yields a 19.8 % reduction in length, a 2.14 % increase in total pressure recovery coefficient, and a 3.83 % reduction in drag—enhancements that directly contribute to the aerodynamic efficiency and propulsion effectiveness of aerospace vehicles. This study underscores the potential of integrating AI-driven surrogate models with adaptive optimization algorithms to advance both theoretical understanding and practical applications in aerospace engineering, particularly in the realm of complex system design and high-speed propulsion technologies. The findings align with the thematic focus of knowledge-based systems and their interdisciplinary applications in modern aerospace science and technology.
ISSN:1270-9638
DOI:10.1016/j.ast.2025.110691