基于特征选择和ELM神经网络的轴承可靠性预测

Reliability Prediction of Bearing Based on Feature Selection and ELM Neural Network Network

针对滚动轴承可靠性预测问题,提出了基于特征选择和ELM网络的可靠性预测方法。首先,对振动信号提取特征,构成特征参数初选集;其次,引入单调性、相关性、鲁棒性三个特征评价指标对特征参数初选集进行特征评价,并定义了一种新的限制性指标,得到可以反映轴承退化过程的参数,构成退化特征参数集;再次,对退化特征参数集进行维数约简,构成低维特征向量集;最后,以退化特征参数集和特征向量集分别为输入数据和标签带入ELM网络中做可靠性预测。通过西安交通大学轴承振动信号数据集证明了该方法的有效性。

Aiming at the problem of rolling bearing reliability forecast,a reliability forecast approach based on feature selection and ELM network is devised.Firstly,the features of vibration signals are extracted to form a preliminary selection of characteristic parameters;secondly,three characteristic evaluation indexes of monotonicity,correlation and robustness were introduced to evaluate the initial selection of characteristic parameters,and a new restrictive index was defined to obtain the parameters that could reflect the bearing degradation process,which constituted the degradation characteristic parameter set;thirdly,the dimension of the degraded feature parameter set is reduced to form the low dimensional feature vector set;finally,the degenerate feature parameter set and feature vector set are used as input data and labels respectively,and are brought into the ELM network for reliability forecast.Its validity is verified by the vibration signal data set of bearing in Xi'an Jiaotong University.