【作者】 王凯；

【导师】 万小金；

【作者基本信息】 武汉理工大学 ， 动力机械及工程， 2017， 硕士

【摘要】 轴承作为旋转机械最常使用的部件,其运行状态直接影响设备的精度和可靠性。因此,对轴承性能退化趋势进行预测具有重要意义。此过程中,常面临两个关键问题:预测模型的建立以及表征机械系统性能退化的指标选取。本文正是从这两方面展开研究,提出基于集成软竞争自适应共振理论(Adaptive Resonance Theory,ART)的滚动轴承性能退化趋势预测方法。在预测模型构建方面,训练速度快、能逼近任意非线性函数的径向基函数(Radial Basis Function,RBF)神经网络可用来建立预测模型。但其仍存在以下问题:样本学习和分类是通过硬竞争机制进行,对混叠处样本分类易造成误分现象,实际上混叠处的样本对预测精度起着重要作用;其次,RBF网络训练时需要大量的样本,而面对大数据样本时,隐层节点增加,网络复杂程度增加,运算效率降低。本文将软竞争ART与RBF神经网络相结合,建立基于软竞争的ART-RBF神经网络模型。该模型通过设置警戒参数,能自适应控制隐含层节点的生成,处理大量的样本时可以简化网络,提高运算效率。软竞争机制的采用,可有效减少混叠处样本误分的问题,提高模式识别以及趋势预测的精度。利用时间序列对预测模型及关键参数进行验证和分析,表明其在一定程度上可以解决RBF模型存在的问题。另外,为了提高预测的精度及稳定性,在软竞争ART-RBF模型基础上,构建基于加权平均的集成预测模型,并使用墨西哥草帽函数对集成预测模型进行了验证。在指标选取方面,本文使用置信度CV值(Confidence Value,CV)作为表征滚动轴承性能退化的综合指标,通过对比预测目标区间与性能正常区间CV值的变化判断其性能退化情况。使用滚动轴承加速疲劳试验获取的振动信号,提取常用的特征参数,对这些参数进行敏感性和趋势性分析后,选取4个特征参数的相对值,再通过自组织映射(Self-Organizing Map,SOM)网络获得综合指标CV值。对比单一特征,该指标在趋势性以及早期退化的敏感性方面优势明显。最后,利用滚动轴承加速疲劳试验获取的加速度信号对上述方法进行验证,得到的CV值的预测结果表明了文中所提预测方法的有效性。此外,该方法对滚动轴承的预知维修和寿命预测的相关研究和应用具有一定的参考价值。更多还原

【Abstract】 As the most commonly used parts of the rotating machinery,the performance of the bearing directly affects the accuracy and reliability of the equipment.Therefore,the prediction for bearing performance degradation trend is of great significance.Two key issues in the prediction of bearing performance degradation are the establishment of prediction model and the selection of bearing performance degradation.This paper aimed to put forward the prediction method of bearing performance degradation based on integrated soft ART-RBF model.The establishment of prediction model is a key problem to be solved.The radial basis function(RBF)neural network can be used to establish a prediction model due to its characteristics such as fast training speed and approach to any nonlinear function.However,there are still some problems: 1)sample learning and classification are performed by the hard competition mechanism,which usually causes misclassification and reduces the prediction accuracy;2)training RBF network requires many samples.Processing so many samples will generate a large number of nodes,leading to a challenge to the network complexity.In this paper,the soft adaptive resonance theory was introduced into the RBF neural network,and the soft ART-RBF neural network prediction model was established.The model can adaptively control the generation of the hidden layer nodes by setting the vigilance parameters,and it can simplify the network and improve the computing efficiency when processing different number of training samples.The adoption of soft competition mechanism can effectively reduce the misclassification and improve the accuracy of pattern recognition and trend prediction.The results of verification of time series and analysis of model parameters showed that this prediction model can solve those problems above-mentioned to a certain extent.The other key problem to be solved is how to select the index which can describe the bearing performance degradation.In this paper,the confidence value(CV)was used as a comprehensive index,and the performance degradation was judged by comparing the change of the CV value between a target interval and the normal stateinterval.This index was based on the minimum quantization error obtained from the self-organizing map(SOM)network.Using the data obtained from the accelerated fatigue test of bearings,some common characteristic parameters were extracted,and the relative value of 4 characteristic parameters were selected to construct the comprehensive index according to the sensitivity and consistency analysis.Compared with the single feature,the comprehensive index based on the multiple feature parameters has obvious advantages in terms of bearing degradation sensitivity.In order to improve the prediction accuracy and stability,the prediction values of the 4 relative characteristic parameters to construct CV prediction samples were obtained based on integration method.In summary,this paper presents a prediction method of bearing performance degradation trend based on integrated soft ART model.The experiments showed that this method can be used to evaluate and predict the bearing performance degradation trend,and has some reference value for the evaluation and prediction of performance degradation research and the practical engineering application.更多还原