组合核RVM与EEMD信息熵的机械设备可靠度评估与预测

机械设备的运行可靠度反映的是机械设备在预期服役环境中正常工作的能力，传统的机械设备运行可靠度评估方法通常是建立在大量历史样本信息的基础上，然而针对没有历史样本信息的机械设备，传统方法难以对机械设备的实际运行状态进行准确评估，也很难对其未来的性能退化趋势进行有效预测。为此，提出通过构建归一化EEMD信息熵与组合核函数相关向量机对机械设备的运行状态进行评估和预测。首先采集机械设备运行过程中的振动信号，采用经验模式分解方法（ensemble empirical mode decomposition，EEMD）对振动信号进行分解，获得多个本征模态函数（intrinsic mode function，IMF）分量，并分别计算其相对能量和归一化EEMD信息熵，构造表征机械设备运行状态的特征指标。随后，构建组合核相关向量机对机械设备的运行状态指标量样本进行学习，并采用粒子群算法对组合核相关向量机中的权值参数和核函数参数进行优化，建立反映机械设备运行状态的可靠度预测模型。最后，将所构造的运行状态特征指标输入给相关向量机进行可靠度的性能退化预测。滚动轴承运行状态评估及预测的结果表明，所提出的机械设备运行状态评价方法能够充分提取反映滚动轴承运行状态的特征信息，运行可靠度预测方法也充分考虑了滚动轴承性能退化状态的历史规律性，相对于单一核函数相关向量机智能预测模型，组合核相关向量机提高了滚动轴承运行状态的预测精度和鲁棒性，为机械设备的运行状态评估和性能退化趋势预测的工程应用提供了一种新的方法。

Assessment and Prediction of Mechanical Equipment Reliability Based on Combined KernelRelevance Vector Machine and EEMD Information Entropy

The ability of mechanical equipment to work normally in the expected service environment is usually reflected by operational reliability. Traditional methods for evaluating operational reliability of mechanical equipment are often based on a large number of historical samples. However, traditional methods are difficult to accurately evaluate the actual operation status of mechanical equipment without historical samples and its future performance degradation trend effectively. The normalized EEMD information entropy and relevance vector machine with combined kernel function were constructed to evaluate and predict the operation reliability of mechanical equipment. Firstly, the vibration signals of mechanical equipment during the operation process were collected, and the vibration signals were decomposed by the ensemble empirical mode decomposition method (EEMD). Several intrinsic mode functions (IMF) components were obtained, and their relative energy and normalized EEMD information entropy were calculated to construct characteristic indicators to characterize the operation status of mechanical equipment. Secondly, the operational samples of mechanical equipment were used to train the relevance vector machine, and particle swarm optimization algorithm was used to optimize the weights and kernel function parameters in the RVM. Finally, the constructed operation state characteristic indicators were input into the relevant vector machine to predict the performance degradation trend for mechanical equipment. The results of evaluation and prediction of rolling bearing operation state showed that the proposed method can fully extract the characteristic information reflecting rolling bearing operation state, and the prediction method of operation reliability also fully takes into account the historical regularity of the rolling bearing performance degradation. Compared with single kernel function relevance vector machine and other intelligent prediction models, the combined kernel relevant vector machine improves the accuracy and robustness of rolling bearing operation state prediction, and provides a novel mode for mechanical equipment operation state evaluation and performance degradation prediction in engineering application.