Intelligent identification of nonlinear parameters in negative stiffness mechanism using SCINet

The negative stiffness mechanism is extensively utilized in aerospace vibration isolation systems owing to its high load-bearing capacity, minimal deformation, and excellent control performance. However, the strong nonlinear behavior inherent poses significant challenges for accurate structural parameter identification. To address the above issue, this paper proposes a parameter identification method for negative stiffness systems based on the novel SCINet deep learning framework. The model incorporates a binary tree-style block interaction architecture that enables hierarchical decomposition of signal features. By integrating both self-learning and mutual-learning mechanisms, the framework enhances sensitivity to nonlinear characteristics. Each sub-block retains a global receptive field, which effectively mitigates the influence of noise on identification accuracy. Furthermore, we introduced pretraining and transfer learning techniques to address the common issue of data scarcity in practical engineering applications. By transferring features across datasets, the method reduces dependence on data from a single task, thereby enhancing the model's robustness. Comparative experiments with baseline models confirm the proposed method's superiority in terms of accuracy, robustness, and effectiveness. Simulation results demonstrate that the method can accurately identify nonlinear parameters and maintain high robustness under complex conditions, including Coulomb friction-viscous damping coupling. Finally, the method is tested in negative stiffness mechanism experiments to validate its engineering applicability. These findings provide reliable parameter calibration support for the optimal design of nonlinear buffering systems.