Physics-Informed and Data-Driven Fusion Approach for Designing MDOF Dynamic System

Abstract To address the problems of high computational cost, poor generalization ability, and inadequate physical consistency associated with frequency response prediction and inverse parameter design for multi-degree-of-freedom (MDOF) vibration systems, we propose an intelligent design method integrating physics-informed and data-driven approaches. Taking a 10-degree-of-freedom lumped mass-spring system as the research object, we first construct a forward prediction model based on the one-dimensional Residual Network-Physics-Informed Neural Network (1D-ResNet-PINN), then design an inverse generation model of the Physics-Constrained Conditional Wasserstein Generative Adversarial Network (PC-cWGAN), and finally establish a closed-loop intelligent design framework integrating the aforementioned forward and inverse models to realize intelligent structural design. Experimental results demonstrate that the forward model exhibits remarkable advantages in prediction accuracy and generalization ability, the inverse model enables end-to-end generation from response requirements to structural parameters with a substantial improvement in computational efficiency, and the construction of the closed-loop framework provides effective technical support for the intelligent design of complex vibration systems.