基于异常检测算法的滚动轴承变工况性能退化评估技术

Performance degradation assessment underlies the condition-based maintenance of rolling bearings. Most available research assumes a stationary operating condition, which is contradictory to filed applications where nonstationary running is rather popular. Nonstationary operation will manifest itself in bearing vibrations and will mix with fault induced information, thus posting challenges to fault detection and performance degradation assessment. In the present proposal, a bearing dynamic model accounting for variable running regime will be developed to examine the relationship between vibrations and running conditions. Furthermore, experimental data is to be processed to investigate signal statistical characteristics and develop noise cancel approach. Above mentioned works will lead to the development of an algorithm able to recover the running condition profile from vibrations under study and eliminate the effect of running condition on vibrations. The resulted vibration signals immune to running conditions will then undergo feature extraction, feature evaluation and selection, as well as dimensionality reduction via manifold learning, resulting in an optimal feature subset. In addition, a hybrid novelty detection method will be constructed which combines the advantages of distanced-based and density-based novelty detection. Finally, a performance degradation indicator is to be presented by the optimal feature subset and the hybrid novelty detection. Such an indicator is anticipated to be sensitive to incipient fault, monotonous to performance degradation progress and locate in a limited range of value. The outcomes of the proposal will be conductive to bearing performance degradation assessment under nonstationary running conditions, and thus contribute to equipment management and maintenance decision as well as enhance the theoretic framework of bearing fault diagnostics and performance degradation assessment.

滚动轴承性能退化评估是实现状态维护的基础，目前研究大多集中在平稳工况下进行。作为外部激励，工业现场普遍存在的变工况将导致轴承动态响应发生变化，从而与故障特征信息相混杂，增加故障检测与性能退化评估难度。针对此问题，项目拟建立考虑工况变化的滚动轴承动力学模型，得到正常与不同性能退化程度下振动响应与工况变化之间的映射关系；结合实验数据分析变工况轴承振动信号统计特性与消噪方法，推导工况反演算法，从振动信号中剔除工况变化造成的影响。在此基础上，研究适合滚动轴承性能退化评估的特征提取、特征定量评价和非线性流形降维，获得最优特征子集；提出一种融合距离型与概率型异常检测算法优势的混合检测算法，获得一种对性能退化程度敏感、与其关系单调且取值范围界定的向量相似度评价方法；最终实现滚动轴承变工况性能退化评估。课题研究将为设备管理和维护提供科学决策依据，丰富和发展滚动轴承故障定性诊断和定量评估理论。

滚动轴承的性能退化评估是实现状态维护的基础，考虑到作为外部激励，工业现场普遍存在的变工况将导致滚动轴承的动态响应发生变化，从而与故障特征信息相混杂，增加故障检测与性能退化评估难度。针对此问题，项目组设计了一种具有变载变速功能的滚动轴承性能退化加速试验平台，建立了考虑载荷和转速影响的轴承非线性动力学模型，推导了工况与动态信号之间的响应机制，提炼了滚动轴承性能退化的关键影响因素。通过建立滚动轴承动力学建模来分析滚动轴承在滑移接触下的接触特性，分析了无故障滚动轴承滑移接触下振动特性，并以铁路机车车辆轴箱轴承为例，建立了机车车辆轴箱轴承多物理域虚拟系统动力学模型，实现了多场景轴箱轴承服役状态的高保真高效率建模和仿真。项目组揭示了变工况下滚动轴承振动信号的统计特性，提出了时频联合分析的信号消噪、工况无关特征提取、面向性能退化评估的特征选择和基于非线性流形学习的最优故障特征子集获取方法；研究了基于消噪后信号的多信息域特征融合的铁路机车轴箱轴承服役状态识别方法。提出了一种旋转机械振动特性评价指标工况敏感度分析方法以及一种融合距离型与概率型异常检测算法优势的混合检测算法，获得一种对性能退化程度敏感、与其关系单调且取值范围界定的向量相似度评价方法，完善了滚动轴承变工况性能退化程度定量评估方法和应用程序。研制了状态检测及智能故障诊断设备，开发了多域背景下铁路机车轴箱轴承状态检测及故障诊断可视化系统，实现了机车轴箱轴承服役状态的准确判断，并在轨道交通相关企业进行了成果应用，取得了较好效果。项目研究成果丰富和发展了滚动轴承故障定性诊断和定量评估理论，具有较为重要的理论意义和广泛的应用前景。