基于IVMD及GNN的外啮合齿轮泵剩余寿命预测研究

Research on Residual Life Prediction of External Gear Pump Based on IVMD and GNN

齿轮泵剩余使用寿命预测对液压元件预防性维护具有重要意义。从流量退化的角度构建一种基于改进的变分模态分解(IVMD)方法及图神经网络(GNN)的外啮合齿轮泵寿命评价模型。首先运用经粒子群优化算法(PSO)优化后的变分模态分解(VMD)方法对齿轮泵原始振动数据进行降噪重构,再从时域、频域、时频域选取重构信号的特征指标并组成特征矩阵,将特征矩阵进行归一化处理后得到齿轮泵寿命评估指标。最后将评估指标与对应流量信号输入到GNN模型中进行训练,进而得到外啮合齿轮泵寿命评价模型。为验证IVMD-GNN模型的优越性,将其与模糊神经网络(ANFIS)及贝叶斯网络(Trainbr-RBFNN)进行对比。结果表明:IVMD-GNN模型预测结果的均方误差为1.68e-04,明显小于其他两种模型,且预测下的寿命分布与真实分布基本吻合,表明该模型拥有更高的准确性,能够对齿轮泵的剩余使用寿命进行评估。

The prediction of the remaining service life of the gear pump is of great significance to the preventive maintenance of hydraulic components.From the perspective of flow degradation,a life evaluation model of external gear pump based on improved variational modal decomposition(IVMD)method and graph neural network(GNN)is constructed.First,use the variational modal decomposition(VMD)method optimized by the particle swarm optimization algorithm(PSO)to reconstruct the original vibration data of the gear pump with noise reduction,and select the characteristics of the reconstructed signal from the time domain,frequency domain,and time-frequency domain.The indicators are combined to form a feature matrix,and the feature matrix is normalized to obtain the gear pump life evaluation indicators.Finally,the evaluation index and the corresponding flow signal are input into the GNN model for training,and then the life evaluation model of the external gear pump is obtained.In order to verify the superiority of IVMD-GNN model,compare it with fuzzy neural network(ANFIS)and Bayesian network(Trainbr-RBFNN)and evaluate the reliability index.The results show that the mean square error of the prediction results of the IVMD-GNN model is 1.68e-04,which is significantly smaller than the other two models,and the predicted life distribution is basically consistent with the true distribution,indicating that the model has higher accuracy and can be used for gears.The remaining service life of the pump is evaluated.