基于LSTM的航空发动机整机支承刚度识别方法

Identification method for support stiffness of whole aero-engine based on LSTM

针对旋转状态下航空发动机整机的支承刚度识别问题,提出了一种基于长短期记忆(long short-termmemory,LSTM)神经网络的航空发动机整机支承刚度识别方法.首先,建立航空发动机整机模型,基于该模型获得目标转速下不同支承刚度对应的位移响应;然后,建立以LSTM为核心层的深度学习网络,以位移响应为输入、支承刚度为输出训练该网络,构建位移响应与支承刚度之间的非线性关系;最后,利用深度学习网络的泛化特性对多个支承刚度进行直接识别.使用该方法对一航空发动机整机进行支承刚度识别,结果表明,支承刚度的识别误差小于2%,LSTM的识别精度高于径向基神经网络与支持向量机.该方法避免了动力学反问题中复杂的寻优过程,实现了复杂非线性结构的动态参数识别.

Based on the long short-term memory(LSTM)neural network,an identification method is proposed to identify the support stiffness of an aero-engine at rotating state.First,the dynamic model of a whole aero-engine with nonlinear support was established.The displacement responses corresponding to different support stiffnesses at the target rotating speed were obtained.Then,the deep learning neural network with LSTM as the core layer was established.The network was trained with the displacement responses as the inputs and the support stiffnesses as the outputs.The nonlinear relationship between the support stiffnesses and the displacement responses was constructed.Finally,the support stiffness was directly identified with the generalization of the deep learning network.The support stiffness of an aero-engine was identified by the proposed method.Results show that the recognition error is less than 2%,and the recognition accuracy of LSTM is better than that of the radial basis function neural network and support vector machine.The proposed method can avoid the complex optimization process in inverse dynamic problems and realize the identification of dynamic parameters for complex nonlinear structures.