摘要详情

ID / 提交时间

93 / 2025-05-14 21:37:34

标题

A Twin Model Construction Method for Unbalanced Vibration in Rotor Systems Based on Random Forest

关键字

rotor unbalance,dynamic twin model,K-means clustering,finite element analysis,random forest

主题及专题

Special Sessions > SSL 14 Artificial self-recovery of high-end mechanical equipment

状态

全文录用

作者

Xiang Li / Beijing University of Chemical Technology

Zhinong Jiang / Beijing University of Chemical Technology

Wenliang Zhang / Prime-rel Electronic Technology Co., Ltd.

Wenhao Bi / Beijing University of Chemical Technology

Mei Li / Beijing University of Chemical Technology

Yanfei Zuo / Beijing University of Chemical Technology

摘要

Aiming at the problems of insufficient real-time monitoring of the dynamic characteristics of rotating machinery under multiple working conditions and high data acquisition costs, A twin model construction method for unbalanced vibration in rotor systems based on random forests is proposed in this paper to achieve rapid prediction of the unbalance or vibration response. Based on the finite element dynamic model, the characteristic data are obtained by applying unbalanced excitation through harmonic response analysis. On the basis of considering the dynamic characteristics of the rotor system, the key node features are extracted based on K-means clustering sampling. According to the requirements of different application scenarios, a three-level prediction twin model is constructed based on the dynamic characteristics of key nodes after order reduction: Model Ⅰ for predicting unbalance, Model Ⅱ for predicting both unbalance and rotational frequency simultaneously, and Model Ⅲ for predicting the frequency response function covering the entire frequency band. Taking the finite element model of a certain rotor system as an example for verification, the results show that the root mean square error (RMSE) of Model Ⅰ in the unbalance prediction is 0.1726 g·m, and the mean absolute error (MAE) is 0.15 g·m. In the rotational frequency prediction of Model Ⅱ, the error does not exceed 0.9 Hz. The RMSE of the unbalance prediction is 0.1504 g·m, and the MAE is 0.116 g·m. In the frequency response function prediction of Model Ⅲ, the Pearson correlation coefficients of each measurement point are all higher than 0.98. The validation results demonstrate that the proposed method offers excellent prediction accuracy and practical engineering value. Real-time inversion of unbalance, rapid and accurate prediction of rotational speed and unbalance in the keyless phase, and overall dynamic balance within the working rotational speed range can be achieved through this method.