A generative design framework for passive vibration absorbers using Lévy flight-based mixed-integer optimisation

Passive vibration absorbers (PVAs), comprising stiffness, damping, and inerter elements, play a critical role in enhancing the dynamic performance of engineering systems. Conventional design methods, such as network synthesis and enumeration, struggle to efficiently explore the vast configuration space (layout and parameters) of complex absorber design problems, often resulting in redundant or suboptimal designs. This paper presents a novel computational intelligence-guided PVA design framework, integrating a generative network that captures all valid configurations for a given element catalogue with a Lévy flight-inspired, computationally efficient evolutionary algorithm. The design task is formulated as a mixed-integer optimisation problem, where binary and continuous variables represent the layout and element parameters, respectively. The generative network operates as a rule-based system, systematically encoding domain knowledge to enable scalable layout–parameter co-optimisation. The Lévy flight algorithm emulates natural foraging behaviour to balance global exploration and local exploitation during search. The framework is validated against three well-established metaheuristics through two benchmark problems: (i) a quarter-car shock absorber under white-noise excitation, and (ii) a multistorey building absorber under wind excitation. A third case study, involving concurrent optimisation of front–rear absorbers in a full-car model subjected to realistic rough-road profiles, is presented as a demonstration problem. Results show that the proposed framework consistently identifies high-performing PVA configurations with substantially reduced computational cost, offering an efficient and scalable alternative to traditional design approaches.