The most common types of induction rotating machine failures are the mechanical faults induced by misalignment, mechanical imbalance and bearing fault. It is well known that the vibration is the best and the earliest indicator of arising mechanical defect. Thus, this paper presents a novel practical bearing fault diagnosis method based on wavelet package decomposition (WPD) associated with neural network. Firstly, the raw signal is segmented by the use of WPD to a set of sub-signals (coefficients futures). Then, the energy related to the most sensible coefficients that contained the greatest dominant fault information is selected as a distinctive feature fault. The analysis results show that this fault indicator varies under different loads and states (healthy or defective). In order to automate the detection and the location of bearing defect, this feature can be used as an input to the artificial neural network. The proposed approach is capable of discriminating faults from four conditions of rolling bearing, the healthy bearing and the three different types of defected bearings: outer race, inner race, and ball. The experimental results prove the effectiveness of this approach.

感应旋转电机故障最常见的类型是由于未对准，机械不平衡和轴承故障引起的机械故障。众所周知，振动是产生机械缺陷的最好且最早的指标。因此，本文提出了一种基于神经网络的小波包分解（WPD）的新型实用轴承故障诊断方法。首先，通过使用WPD将原始信号分段为一组子信号（系数期货）。然后，将与包含最大的主断层信息的最敏感的系数相关的能量选择为特征性断层。分析结果表明，该故障指示器在不同的负载和状态（健康或故障）下会发生变化。为了自动进行轴承缺陷的检测和定位，此功能可以用作人工神经网络的输入。所提出的方法能够从滚动轴承，健康轴承和三种不同类型的缺陷轴承（外圈，内圈和滚珠）的四个条件中识别出故障。实验结果证明了该方法的有效性。