In machinery fault detection and identification (FDI), decomposing vibration signals into corresponding intrinsic mode functions (IMFs) reduces the intricacy in extracting weak fault features at the early failure state. However, selecting a suitable IMF for fault information extraction is a challenging task. Analyzing the non-Gaussian IMFs allows extracting effective fault-related information rather than the entire signal or other IMFs because the vibration signals are random in nature. In this study, we present an IMF selection method based on the maximum kurtosis value of each IMF. A kurtosis computation method named autogram is used. It considers the autocovariance function to characterize the 2nd order cyclostationary. We deploy the AdaBoost algorithm with a decision tree classifier to gain a better performance compared with other tree-based classifiers. The proposed FDI framework can effectively detect and classify multiple fault features at the incipient failure stage.

在机械故障检测和识别 (FDI) 中，将振动信号分解为相应的本征模态函数 (IMF) 降低了在早期故障状态下提取弱故障特征的复杂性。然而，为故障信息提取选择合适的 IMF 是一项具有挑战性的任务。分析非高斯 IMF 允许提取有效的故障相关信息，而不是整个信号或其他 IMF，因为振动信号本质上是随机的。在这项研究中，我们提出了一种基于每个 IMF 的最大峰度值的 IMF 选择方法。使用名为 autogram 的峰度计算方法。它考虑自协方差函数来表征二阶循环平稳。与其他基于树的分类器相比，我们使用决策树分类器部署 AdaBoost 算法以获得更好的性能。所提出的 FDI 框架可以在初始故障阶段有效地检测和分类多个故障特征。