轴承退化状态表征的流形融合指标构建方法

Study on Fusion Indicator Construction Method for Bearing Degradation Condition Evaluation Based on Manifold Learning

针对传统数据融合构建的轴承退化状态表征指标易受噪声影响,存在波动和难以兼顾轴承退化状态的全局结构和局部结构等问题,提出了基于一致流形近似与投影的轴承退化状态表征的融合健康指标构建方法。首先,分别在时、频域内计算轴承退化状态表征指标,构建原始高维退化趋势特征集;然后,建立退化状态敏感特征筛选准则,从高维特征集中筛选出较好表征轴承退化状态的特征指标;最后,引入一致流形近似与投影算法和指数加权滑动平均对筛选出的敏感特征进行融合与平滑,构建轴承退化状态表征指标。通过两组全寿命轴承振动数据对所提方法进行验证,结果表明,与单一指标以及传统的指标融合构建方法相比,该方法可以克服单一指标表征能力有限和传统指标融合方法能力不够全面等缺点,能有效提高轴承健康指标的单调性与趋势性,以准确表征轴承的退化状态。

The health indicators extracted by the traditional data fusion methods might be failure for characterizing the bearing degradation condition under the strong background noise,such as violent fluctuation,difficulty in balancing the global and lo-cal structures.As a result,a novel fusion health indicator construction method for bearing degradation condition evaluation is presented based on uniform manifold approximation and projected.First,the health indicators of bearing degradation condition were calculated in time and frequency domain,respectively,which were employed to produce an original high-dimensional fea-ture set.Then,the finer health indicators were selected from the high-dimensional feature set via defining the sensitive criterion for bearing degraded process.Finally,these selected sensitive features were fused by the uniform manifold approximation and pro-jection,meanwhile,the exponentially weighted moving average was utilized to make the fusion feature more smooth.Two run-to-failure bearing data sets were used to verify the effectiveness of the proposed method.The results validate that the presented method owns a better ability for enhancing the monotonicity and tendency of bearing health indicators than the traditional data fusion methods and overcomes the shortcomings of limited representation ability of single indicator.