欠定条件下数控机床滚动轴承微弱信号的自适应提取方法研究

Rolling element bearing is one of important machine components for numerical-control machine. Condition monitoring and fault diagnosis for bearings is an imperative part for preventive maintenance procedures and reliability improvement of these machines. When a localized fault occurs at the early stage of bearing failure, the impulses generated by the fault are relatively weak and usually overwhelmed by large noise and other higher-level macro-structural vibration signals generated by adjacent machine components and other machines. For such a multi-component vibration signal, it is difficult to extract the critical bearing signal effectively and then give accurate diagnosis decisions. With the aim of adaptively extracting the weak bearing signal from the vibration signal collected by limited sensors as early as possible, this proposal focuses on the following research topics: Firstly, develop an adaptive noise reduction method for the collected bearing vibration signal with large noise by using the spectrum difference between noise and vibration signals along with empirical mode decomposition, so that the filtered signals only involve multiple vibration sources and are used for further signal separation; Secondly, develop an adaptive weak bearing signal extraction method using the sparsity of vibration signals generated by faulty bearings and underdetermined blind signal separation model so that the weak bearing signal would be effectively extract from multiple vibration sources and used for accurate fault diagnosis; Finally, investigate optimization for stopping criterion of EMD to reduce the computation complexity and load, so that the advanced signal processing methods to be developed can be used for online condition monitoring and fault diagnosis. The research results to be obtained will provide available and effective methods for solving the problem of weak bearing signal extraction for condition monitoring and fault diagnosis of mechanical machines; meanwhile, extend advanced signal processing methods along with nonlinear methodology from theoretic analyses to real problems.

滚动轴承是数控机床的关键部件，即使微小故障也会影响机床的加工精度和安全运行。状态监测与故障诊断是实现轴承预防性维修和提高可靠性的关键技术，但实际观测信号中包含强噪声和多振源信息，严重制约了轴承故障早期的微弱信号提取，致使难以及早发现设备异常。本项目将按照噪声去除、振源分离和算法优化三个步骤来解决上述难题。首先，根据噪声和振动信号的频谱差异，研究基于自适应信号分解的强噪声去除方法，解决强噪声降低信号分析性能的难题。然后，在观测信号数小于振源数(欠定)的情况下，研究基于稀疏表示的振源分离方法，由此构成轴承微弱信号的自适应提取方法。最后，通过优化信号分解方法来提升分解的性能和效率，使所提微弱信号提取方法不仅能够解决微弱信号不易提取的难题，还能满足现场监测与故障诊断的及时性要求。本项目的研究成果将为数控机床滚动轴承现场监测与故障诊断中的微弱信号检测提供行之有效的方法，具有重要的理论意义和应用价值。

轴承微弱信号自适应处理与分析是轴承动态性能检测与大数据评估方法研究的关键环节。本项目通过分析大型旋转机械信号处理中存在的关键问题，制定了微弱信号提取、噪声去除和算法改进三项研究内容。. 首先，针对轴承故障信号微弱、难以提取的问题，提出了基于能量窗口合并的自适应共振解调方法，可自适应优化滤波器参数，改善滤波器性能，实现了强噪声工况下的轴承微弱信号的提取。. 其次，针对强噪声工况下轴承多故障振动信号检测问题，本项目提出了多段串联随机共振系统，结合并联与串联随机共振系统的特点，将高频段的信号能量不断向低频段集中；同时，定义了联合信噪比自动调整系统参数，实现了强噪声工况下的轴承多故障信号检测。. 最后，针对自适应信号分解方法中普遍存在的端点效应问题，本项目提出了基于频域相似度的自适应波形延拓方法，可以较好地保存信号的时频特征，解决了端点效应导致的误差累积问题。在此基础上，针对分解产生的多个信号分量，本项目提出了基于综合特性的敏感分量提取方法，解决了轴承特征信号的选择问题，并且分析了常用统计指标的故障敏感性，避免了特征选择的盲目性，提高了特征分析与选择的效率。. 在此基础上，结合实验信号与工程实测信号对上述方法进行了验证与分析。同时，将改进后的自适应信号分解方法与其他先进信号处理方法相结合，进一步提高了复杂工况下轴承信号处理的性能，而且为旋转机械的滚动轴承动态检测与性能评估提供了有效的方法与指标。.项目执行期间研究成果包括：累计发表、录用期刊论文6篇、会议论文8篇，其中SCI检索5篇、EI检索4篇；申请发明专利2项，授权1项；参加学术会议与交流6次，发表英文专著中的1章。. 本项目研究工作不仅提供了有效的轴承微弱信号自适应处理方法，也可应用于具有相似运行机理的旋转机械其他关键部件的信号处理中，具有可推广性。同时，相关方法与分析结果可为旋转机械动态检测与性能评估提供了实用的处理方法与监测指标，具有实用性。