KT-MDO: a knowledge-template-driven multidisciplinary design optimization framework

Multidisciplinary design optimization (MDO) typically employs surrogate models to alleviate the high computational cost of multidisciplinary simulations. However, under data-scarce engineering conditions, purely data-driven surrogates often suffer from accuracy degradation. Although knowledge–data fusion can mitigate this problem, existing MDO frameworks lack a unified mechanism for knowledge management and on-demand invocation. Domain knowledge is usually hard-coded in a static form within individual discipline modules, when design requirements change frequently, this rigid integration cannot accommodate dynamic reconfiguration of the optimization workflow, thereby constraining both design efficiency and system scalability. To address this issue, this paper proposes a Knowledge-Template-Driven Multidisciplinary Design Optimization framework (KT-MDO). First, domain knowledge is categorized into four types, attribute, monotonicity, shape, and formula, and a systematized representation is established for each type. Then, two types of knowledge templates are constructed: one for automatically formulating MDO problem models, and the other for automatically generating the corresponding code, enabling dynamic adaptation to diverse design requirements. In two representative lightweight design scenarios of an automotive body-in-white, KT-MDO achieves optimization performance comparable to baseline methods while reducing manual model configuration workload by approximately 54%. It also enables rapid adaptation across different scenarios with minimal additional cost, thereby significantly improving the efficiency and practicality of MDO.