LVQ神经网络在滚动轴承故障诊断中的应用研究

A Learning Vector Quantization Neural Network for Faults Diagnosis of Rolling Bearing and its Application

滚动轴承多部位多类别故障诊断属于大规模或者较复杂分类问题,利用智能方法诊断时需要设计结构合理的神经网络才能实现高精度诊断。因此,笔者对基于学习向量量化(LVQ)神经网络的滚动轴承多故障诊断进行了研究。首先通过模拟故障实验采集到滚动轴承不同部位和类型的10种故障振动信号。然后选择db16母小波对实验所采集的数据信号进行三层小波变换,并提取第三层小波8个节点信号的能量特征。最后将能量特征组成LVQ神经网络的输入特征向量进行网络训练与检测,以实现滚动轴承的故障定位和模式识别。实验结果证明,所提出的诊断方法避免了故障定位和故障类型的分别诊断,能够在网络训练后同时较精确地实现滚动轴承10种故障的定位与模式识别。

Multi-site and muhi-class fault diagnosis of roiling bearing is a large-scale and complex classification problem. Since the properly designed structure of the neural network is a key factor to high-precision diagnosis in intelligent diagnostic system, the Learning Vector Quantization （LVQ） neural network based rolling bearing faults diagnosis method is studied in this paper. The experiments were firstly implemented to record the vibration signals of seeded faults under ten different working conditions. Then the measured signals were decomposed with wavelet transform （WT） to 3 level using dbl6 wavelet function, and hence the statistics energy of each sub-band at the 3rd level could be calculated. Finally, the statistics energies were used as input feature vectors to train the LVQ and recognize the rolling bearing faults. The experimental results show that the proposed method can identify the location and pattern of multi-site and multi-class rolling bearing faults accurately at the same time.