Intelligent Multi-Physics Collaborative Design and Layout Optimization for Irradiated Inertial Microsystems Based on Cascade Surrogate Models

Abstract

To address the "radiation-electromagnetic-thermo-mechanical" multi-physics coupling challenges faced by inertial microsystems in deep space, alongside the bottleneck of prohibitive computational costs associated with traditional Finite Element Methods (FEM), this paper proposes an intelligent collaborative design methodology driven by a physical-prior cascade surrogate model (C-MLP) and Particle Swarm Optimization (PSO). The C-MLP utilizes the maximum PCB temperature as a bridging variable: the first-stage network rapidly predicts spatial thermo-mechanical distributions, while the second-stage network evaluates high-frequency signal integrity under synergistic effects. Testing reveals that even under an extreme temperature perturbation of 15°C, the system's Mean Absolute Percentage Error (MAPE) remains stable at 2.31%. This model drastically reduces the time required for a single multi-physics evaluation from 37 minutes to 0.9165ms, enabling tens of thousands of global optimization iterations to be executed in just 27.13 seconds. Ultimately, the algorithm yields a collision-free "optimal layout," minimizing the total thermal stress on the main chips while ensuring excellent signal integrity.