基于多层超限学习机的滚动轴承故障诊断方法

针对目前轴承故障诊断领域存在的海量数据问题及快速学习、实时监测的诊断要求,采用一种多层超限学习机方法对滚动轴承故障数据进行诊断测试。该方法直接学习轴承故障振动时域信号,与传统诊断方法相比,省去了复杂的信号处理过程,更加简便。将多层超限学习机方法的诊断结果分别与单层超限学习机、深度神经网络方法的诊断结果进行比较,多层超限学习机具有明显优势:(1)与单层超限学习机相比,多层超限学习机具有更好地学习和特征提取能力,其诊断准确率可达到98.29%;(2)与深度神经网络相比,多层超限学习机能够在保证较高诊断准确率的前提下,获得较快的训练速度,其训练速度较深度神经网络提高了41倍。结果表明,所采用的方法在滚动轴承故障诊断方面具有很好的效果和应用价值。

Multi-layer extreme learning machine methods based fault diagnosis of Rolling bearing

Aiming at the problem of massive data and the requirement of rapid learning and real-time monitoring existed in the field of bearing fault diagnosis currently, this paper deals with the fault diagnostic data of rolling bearings utilizing a kind of multi-layer extreme learning machine. Compared with the traditional diagnostic methods, this method analyses the time domain signal of fault vibration instead of the frequency domain signal, thus eliminates the time for complex signal processing. We use the SLFNs and the DNN to test the diagnosis data of rolling bearings, and make the comparison with multi-layer extreme learning machine methods. It turns out to be that multi-layer extreme learning machine methods has obvious advantages as follows: 1) compared with the SLFNs, multi-layer learning machine methods has better learning and feature extraction capabilities, and the diagnostic accuracy rate could be 98.29%; 2) compared with the DNN method, it can obtained faster training speed in the premise of high diagnostic accuracy, which could be improved 41 times. The results showed that the method established in this paper has a good effect and application in fault diagnosis of rolling bearings.