振荡行为下变转速工况轴承故障特征提取及诊断研究

The vibration signal under variable speed operation is always more information-rich than that under constant speed operation; however, the signal is more non-linear and non-stationary. It, therefore, is crucial as well as challenging to extract bearing fault feature from the vibration signal under variable speed condition. As vibration signals arising from faulty bearing under variable speed are not only non-stationary and non-linear but also exhibit a mixture of oscillatory and non-oscillatory transient behaviors, this project proposes a non-linear signal decomposition methodology based on signal’s oscillatory behavior to extract bearing fault feature under variable speed condition. Based on the extracted signal, the bearing fault diagnosis strategy free from shaft rotating speed is then explored to overcome the difficulty that the shaft rotating speed cannot be estimated as it is coupled by fault types. Specifically, this project consists of: 1) the adaptive determination of bases for signal components with different oscillation behaviors and the improvement of the signal decomposition algorithm for fast convergence, leading to an effective bearing fault feature extraction; 2) the local maximal energy based instantaneous frequency (IF) estimation from the extracted signal; 3) the iterative generalized demodulation for signal using the extracted IF and the establishment of the bearing fault diagnosis strategy without involving shaft rotating speed and resampling. The diagnosis results can then be made based on the developed strategy. To conclude, this project intends to develop a methodology which is not based on the frequency/scale information for bearing fault feature extraction and does not require shaft rotating speed and resampling for nonstationary signal processing, leading to a novel and reliable method for bearing fault diagnosis under variable speed condition.

轴承在变转速工况下的振动信号相比平稳状态而言包含更多反映轴承状态的信息，但信号本身也表现出非平稳性和非线性。因此，如何对该工况下的信号进行有效特征提取和分析是轴承故障诊断的关键，也是难点。基于轴承故障引起的冲击在振荡行为上与其他信号成分表现出瞬态和持续的差别，本项目拟以振荡行为理论为基础，研究非线性的轴承故障特征信号提取方法，进而针对变转速工况轴承转速估计困难的问题，探究构建无需轴转速的变转速轴承故障诊断策略。主要内容包括：首先，研究基于振荡行为的信号分解中字典基的自适应选择准则和优化信号分解迭代算法，建立快速提取轴承故障特征信号的方法；其次，在所提取信号上开展基于局部能量最大的瞬时频率估计算法的研究；然后，利用瞬时频率，研究信号的循环广义解调，构建无需转速和重采样的变转速轴承故障诊断策略。本项目研究的振荡行为下故障特征提取方法和无需转速的故障诊断策略将为变转速轴承故障诊断提供一种新途径。

轴承作为旋转机械的重要零部件之一，常运行于变速工况且经受强冲击作用，是极易危及设备安全的易损件。因此，对轴承实施故障诊断具有重要的实际意义。但变转速下轴承的故障诊断不同于恒定转速情形，变转速的工作模式将增加信号的非平稳性和非线性；同时，转速的变化也将导致故障诊断方法中各参数甚至诊断方法也随之改变，由此增加了变转速下轴承故障特征提取和诊断的难度。为此，本项目研究了适用于变转速下轴承故障特征信号提取的方法；基于此，构建了基于振动信号本身准确提取瞬时频率的算法；而后，建立了不依赖于轴转速信息的变转速轴承故障检测和故障类型智能识别策略，实现了对变转速轴承的故障诊断。主要研究内容及重要结果如下：.(1) 分析了轴承在故障激励下的振动特性，得出转速变化对故障引起的瞬态特征的振动特性影响可忽略的结论，据此，在构造稀疏过完备字典时只需构造涵盖单个形态的小波基函数； . (2) 针对稀疏框架目标函数求解收敛速度缓慢的问题，引入了优化最小算法求解稀疏表示系数。该算法使得原本相对较难的目标函数的求解变得更加简便可行，将其运用于轴承、齿轮的复合故障诊断，验证了其有效性，且与传统的求解算法进行了比较，体现了最小优化算法在运算效率上的优越性； . (3) 针对瞬时频率提取面临的挑战，提出了基于概率密度分布的多源时频脊线融合算法，该算法通过对异常时频脊线区间进行定位、多源时频脊线预处理和融合实现对瞬时频率的精确提取，为在无需转速计的情况下实现轴承的故障诊断提供了基础； . (4) 提出了基于平均故障特征系数的变转速轴承故障诊断策略，避免了重采样过程中插值算法对阶比分析结果的影响，在一定程度上也简化了轴承的故障诊断流程；. (5) 提出了基于改进深度置信网络的分层自适应方法用于轴承故障诊断，构造分层深度学习结构从轴承中提取深层代表特征以识别轴承故障类型和程度，实验表明在诊断精度和运行效率方面，该算法均具有优越的性能。