Surrogate-Assisted Peak-Slope Optimization of Dynamic Vibration Absorbers

mechanical systems by shifting the associated response peaks. However, optimizing their performance is computationally demanding, especially in systems with many degrees of freedom or numerous components, such as inerter-based DVAs. This study proposes a computationally efficient, surrogate-assisted optimization framework that leverages a novel Peak-Slope (PS) performance metric. As a reinterpretation of the classical H_∞ approach, the PS metric evaluates the effectiveness of vibration absorbers by measuring the secant slope between adjacent resonance peaks in the frequency response function. A well-tuned DVA yields a PS value approaching zero, indicating minimal variation between peaks and thus optimal mitigation of resonance. To reduce complexity, the structural parameters are held constant, and the influence of absorber parameters on the PS metric is isolated. The optimization space is simplified using surrogate models constructed via quartic polynomial regression. A novel decoupling algorithm introduced in this study enables efficient estimation of the PS metric as the Decoupled Peak-Slope (DPS) by expressing it as a sum of independent surrogate functions, each dependent on a single DVA parameter. Optimization is then performed by minimizing this total sum. A fully coupled 1DOF–1DOF system, incorporating masses, springs, dampers, and inerters, is used as the benchmark to validate the method. The DPS approach is compared against traditional genetic algorithm (GA)-based optimization, demonstrating substantial gains in both speed and accuracy. Further validation is conducted using reduced-order systems from the literature, confirming the true decoupling capability of the framework. For four distinct structural configurations, the decoupled surrogate equations are generated and summarized, forming a catalogue of precomputed polynomial functions that enables rapid evaluation of optimal DVA parameters across a range of systems. The results show strong agreement with analytical solutions and superior performance over GA-based methods. This positions the DPS framework as a fast, accurate tool for future semi-active DVA systems, enabling real-time tuning via precomputed surrogate functions.