.png){kind=icon}
.png){kind=icon}
Return
Generative AI-driven inverse design optimization of composite blended-wing-body aircraft under multidisciplinary constraints
.png){kind=icon}
PRE
AI
M
Z
X
H
DOI：10.1007/s00158-026-04375-x.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
The multidisciplinary collaborative design of blended-wing-body (BWB) aircraft structures is fundamental to achieving lightweight and high-reliability objectives in next-generation aircraft. However, traditional design methods are constrained by their limited capability to explore high-dimensional design spaces and by the inherent complexity of multidisciplinary interactions, making it challenging to effectively balance structural performance requirements with design efficiency. To overcome these limitations, this study proposes a generative artificial intelligence (AI)-based inverse multidisciplinary design optimization method for composite BWB aircraft. The method addresses three challenges: generating high-fidelity design samples under multidisciplinary constraints, establishing efficient mappings between design spaces and performance metrics, and performing collaborative optimization under highly nonlinear constraints. To enhance both the quality and efficiency of design sample generation, this work integrates high-fidelity fluid–structure interaction (FSI) simulations with a conditional generative adversarial network (CGAN), and proposes an inequality constrained entropy-regularized CGAN (ICE-CGAN) design generation model. By incorporating inequality constraint conditions and objective guidance into the generative training process, combined with a Sinkhorn divergence-based distribution regularization mechanism, the model significantly improves the diversity and feasibility of generated design candidates while maintaining consistency with the underlying data distribution. Optimization results demonstrate that BWB configuration candidates satisfying all multidisciplinary constraints are successfully generated. Surrogate model predictions demonstrate an average constraint satisfaction rate exceeding 99% across diverse settings. High-fidelity validation on 11 generated designs confirms that all designs satisfy the physical constraints, achieving complete real-world feasibility. The small high-fidelity dataset leads to large surrogate errors in sparse design regions. To address this limitation, Sinkhorn regularization anchors generated designs to data-dense areas, thereby improving reliability. This study provides an efficient and reliable data-driven collaborative design approach for composite BWB aircraft, demonstrating strong potential for engineering applications.
Keywords:
Generative AI
Conditional generative adversarial network
Multidisciplinary constraints
Blended-wing-body
Inverse design
Journal
IF:
4
Papers:
4.8K
Citations:
1.7W
Organization
Cited Papers