中国科学院南京天文光学技术研究所机构知识库

In this paper, we developed a micro-vibration measurement platform using five piezoelectric sensors. Compared  to the traditional four-sensor measurement configuration, this design offers enhanced stiffness and expands the  measurable frequency bandwidth. Through finite element analysis and optimized design, we determined the  structural parameters of the platform and maximized its fundamental frequency. Additionally, a novel transfer  learning algorithm was implemented for dynamic calibration. Within the calibration frequency band, this  transfer learning approach utilizes the optimized parameters of the deep neural network (DNN) from the pre vious frequency as the initialization parameters for the next frequency. This process enables the reuse of  structural information across the entire frequency band, thereby improving the network’s convergence rate and  reducing the workload required for tuning training parameters. The calibrated platform can accurately measure  generalized forces within the frequency range of 10 ~ 600 Hz. The relative errors for the three-dimensional  forces Fx, Fy and Fz are 2.38 %, 1.39 %, and 2.80 %, respectively, while those for the three-dimensional mo ments are 2.18 %, 2.65 %, and 1.90 %. These results indicate that the calibrated measurement platform meets the  requirements for precise measurements.