基于BP神经网络构建简化翼结构动态代理模型

Construction of dynamic surrogate model of simplified wing structure based on BP neural network

  • 摘要: 建立精确且可信度高的代理模型是数字孪生技术中的关键环节之一。为了研究航天结构动力学中动态代理模型的构建方法,选择简化翼结构作为研究对象,分别利用单点正弦定频激励及全场加速度扫频激励获得结构在瞬态动力学分析下的位移和加速度响应;将时间和部分节点响应结果作为输入,将希望关注的节点处的响应值作为输出,利用BP神经网络构建动态代理模型,当所建立的代理模型精度达标后即构建了翼结构瞬态动力学的动态代理模型。该法构建的模型不仅可用于快速预测多个节点的位移和加速度响应,亦可为后续构建航天复杂结构动力学数字孪生体奠定基础。

     

    Abstract: Establishing an accurate and reliable surrogate model is one of the key links in digital twin technology. In order to study the method for constructing the dynamic surrogate model in space structural dynamics, a simplified wing structure was selected as the research object. The displacement and acceleration responses of the structure under transient dynamic analysis were obtained by single point sinusoidal constant frequency excitation and full field acceleration sweep frequency excitation, respectively. Taking the time and the response results of some nodes as inputs, and the response values at the concerned nodes as outputs, back propagation (BP) neural network was used to build a dynamic surrogate model. The dynamic surrogate model of the wing structure under transient dynamics was deemed to have been constructed after the established model’s accuracy met the requirement. The model constructed by this method can not only be used to quickly predict the displacement and acceleration responses of multiple nodes, but also lay a foundation for the subsequent construction for dynamic digital twins of space complex structures.

     

/

返回文章
返回