VMD-模平方阈值与PNN相结合的齿轮故障诊断
Fault Diagnosis of Gear By Using VMD-Modulo Square Threshold and PNN

针对故障齿轮振动信号的非平稳和调制特性，提出了在变分模态分解（VMD）-模平方阈值降噪的基础上利用概率神经网络（PNN）进行齿轮故障诊断的方法。首先，利用VMD将原始振动信号分解为若干个本征模态函数分量，采用模平方阈值方法对各分量处理后并重构；然后，提取重构信号的峭度和均方根作为特征值组成特征向量；最后，将特征向量输入PNN实现故障类型识别。通过齿轮故障试验分析，将其与基于EMD-模平方阈值、LMD-模平方阈值和EEMD-模平方阈值的BP神经网络故障诊断方法相比较。结果表明，该方法能有效的提取特征信息，故障诊断准确率高达96.875%，证明了所提方法的可行性和有效性。
Aiming at non-stationary and modulation characteristics of fault vibration signals of gear, a fault diagnosis method is proposed by using the probabilistic neural network(PNN) on the basis of the variational mode decomposition(VMD)-modulo square threshold diagnosing. Firstly, the signals was decomposed into the several intrinsic mode functions(IMF) and each one was processed with the modulo squared threshold method. Then, the kurtosis and root mean square of the signals which were reconstructed with IMFs after the modulo squared threshold method were extracted as fault feature vectors for pattern recognition. Finally, the feature vector was input PNN model to realize the fault type identification. Through gear fault tests, based on EMD-modulo square threshold, LMD-modulo square threshold and EEMD-modulo square threshold, the accuracy was up to 96.875% comparing with BPNN, in which the feasibility and effectiveness of the present method was verified