Wu Huifeng, Dong Rui, Liu Zheng, Wang Hui, Liang Lei
National Engineering Research Center of Fiber Optic Sensing Technology and Networks, Wuhan University of Technology, Wuhan 430070, China.
School of Electronic Information and Automation, Guilin University of Aerospace Technology, Guilin 541004, China.
Micromachines (Basel). 2022 Jul 31;13(8):1237. doi: 10.3390/mi13081237.
To reduce the dependence of real-time deformation monitoring and shape reconstruction of flexible planar structures on experience, mathematical models, specific structural curvature (shape) sensors, etc., we propose a reconstruction approach based on FBG and a data-driven model; with the aid of ANSYS finite element software, a simulation model was built, and training samples were collected. After the machine learning training, the mapping relationship was established, which is between the strain and the deformation variables (in three directions of the -, -, -axis) of each point of the surface of the flexible planar structure. Four data-driven models were constructed (linear regression, regression tree, integrated tree, and BP neural network) and comprehensively evaluated; the predictive value of the BP neural network was closer to the true value (R = 0.9091/0.9979/0.9964). Finally, the replication experiment on the flexible planar structure specimen showed that the maximum predictive error in the -, -, and -axis coordinates were 2.93%, 35.59%, and 16.21%, respectively. The predictive results are highly consistent with the expected results of flexible planar structure deformation monitoring and shape reconstruction in the existing test environment. The method provides a new high-precision method for the real-time monitoring and shape reconstruction of flexible planar structures.
为降低柔性平面结构实时变形监测与形状重建对经验、数学模型、特定结构曲率（形状）传感器等的依赖，我们提出一种基于光纤布拉格光栅（FBG）和数据驱动模型的重建方法；借助ANSYS有限元软件，构建了仿真模型并采集了训练样本。经过机器学习训练，建立了柔性平面结构表面各点应变与变形变量（在x、y、z轴三个方向）之间的映射关系。构建了四个数据驱动模型（线性回归、回归树、集成树和BP神经网络）并进行综合评估；BP神经网络的预测值更接近真实值（R = 0.9091/0.9979/0.9964）。最后，在柔性平面结构试件上进行的重复实验表明，x、y和z轴坐标的最大预测误差分别为2.93%、35.59%和16.21%。预测结果与现有测试环境中柔性平面结构变形监测和形状重建的预期结果高度一致。该方法为柔性平面结构的实时监测和形状重建提供了一种新的高精度方法。
