滑动轴承磨损智能诊断与寿命预测方法研究

Research on intelligent diagnostic and prognostic method for sliding bearing wear

滑动轴承是现代机械设备的核心部件之一,在润滑不充分的情况下,轴承接触面易出现磨损退化,为设备运行带来极大安全隐患。为实现对滑动轴承磨损智能诊断与寿命预测,该文设计并建立了滑动轴承磨损试验台,开展了滑动轴承磨损退化试验和振动测量试验,得到不同磨损深度下的轴承振动信号,并构建了时域、频域、非线性结合的多域特征,可有效表征轴承磨损退化信息;提出了基于稀疏相关向量迭代指数退化的预测方法,对新输入的振动信号进行定量的磨损深度诊断,并基于历史数据的稀疏相关向量加权拟合退化模型,进行磨损趋势预测。试验证实了所提方法能够有效支持滑动轴承磨损状态的智能诊断与寿命预测。

[Objective] Sliding bearings are critical components of modern machinery,and their proper function is critical.However,inadequate lubrication can cause significant wear and degradation of the bearing contact surfaces,posing significant safety risks.Monitoring the wear state of sliding bearings during operation and predicting their remaining useful life(RUL) is crucial for ensuring equipment safety and reducing maintenance costs.Despite this need,the current online diagnostic and prognostic methods for sliding bearings are lacking.To address this issue,this study proposes an intelligent diagnostic and prognostic method for sliding bearing wear based on multidomain features and relevance-vector-based iterative exponential degradation(RV-IED).[Methods] This paper designed and built a test rig to simulate real-world operating conditions of sliding bearings and collected data under different wear conditions.Wear degradation and vibration measurement tests were conducted to measure the maximum wear depth(MWD) and vibration signals during the tests.For feature engineering,multidomain features combining the time domain,frequency domain,and nonlinear characteristics were constructed.From the energy of the vibration signals,time-domain features were derived.A Fourier transform was then applied to these signals to obtain frequency-domain waveforms,which are decomposed into multiple normal distributions.Calculating the intensity values of the top ten peaks' ■ frequency-domain vector,which was then reduced to one-dimensional frequency-domain features using principal component analysis.To handle the high-dimensionality of feature spaces,dynamic time warping was used to compute distances between different spectra as nonlinear features.These multidomain features served as input vectors for diagnosis,with the corresponding MWDs used as labels for training the relevance vector machine(RVM).New samples were diagnosed by outputting the current MWD.After each diagnosis,another RVM extracted sparse relevance vectors and corresponding weights from historical diagnosis data,fitting an exponential model using nonlinear least squares.This model predicts the bearing RUL by extending the trend to a preset threshold.[Results] The proposed method was evaluated against traditional time domain and frequency domain features combined with SVM/RVM methods as a control group.The experimental results showed that:(1) In diagnosing wear depth,the proposed method achieved a diagnostic error within 10%,outperforming the control group;(2) Although the diagnostic error increased as the training set size decreased,the changes were minimal beyond a reduction of 30%,making the method suitable for small sample sizes.We recommend a dataset size that does not exceed 1×10~4;(3) For RUL prediction,the proposed method's cumulative relative accuracy is 0.59,compared to 0.36 for the control group.[Conclusions] By leveraging the constructed sliding bearing wear test rig and monitoring data,multidomain features were created to accurately reflect bearing wear degradation.A diagnostic and prognostic method based on RV-IED for sliding bearing wear provides accurate diagnostics and RUL predictions,even for small sample sizes.This method surpasses traditional approaches and effectively supports intelligent diagnostics and predictive maintenance of sliding bearing wear states.This innovative approach holds promise for further advancement in fault prediction and health management in mechanical systems.