基于优化VMD和深度卷积神经网络的柴油发电机组故障诊断

Fault Diagnosis of Diesel Generator Set Based on Optimized VMD and Deep Convolution Neural Network

针对柴油发电机组故障信号非平稳非线性特征参数难以提取的问题,结合深度学习的优势,提出一种基于变分模态分解(VMD)和深度卷积神经网络(CNN)相结合的故障诊断模型。为克服VMD算法中分解模态数较难确定的问题,采用峭度准则来选取最优分解模态数,将优化的VMD算法用于不同工况下的柴油发电机组声信号进行分解,转化为灰度图像作为网络输入,通过CNN自动进行特征提取,并利用训练集样本进行网络训练。为避免背景噪声和提高故障诊断精度,使用双传感器采集发电机组声信号。通过测试集的验证,表明该模型在对柴油发电机组的故障诊断中实现了不同工况下的可靠判别,进一步提升了故障判别的准确性。通过对比其他4种故障诊断方法,结果表明所提方法诊断精度更高且鲁棒性好。

In order to solve the problem that it is difficult to extract the non stationary nonlinear characteristic parameters of diesel generator set fault signals,a fault diagnosis model based on the combination of variational mode decomposition(VMD)and deep convolution neural network(CNN)is proposed combined with the advantages of deep learning.In order to overcome the problem that it is difficult to determine the number of decomposition modes in VMD algorithm,the kurtosis criterion is used to select the optimal number of decomposition modes.The optimized VMD algorithm is used to decompose the acoustic signals of diesel generator sets under different working conditions,and is converted into gray images as network input.The feature is automatically extracted through CNN,and network training is conducted using training set samples.In order to avoid background noise and improve the accuracy of fault diagnosis,dual sensors are used to collect the acoustic signals of the generator set.The verification of the test set shows that the model can realize reliable discrimination under different working conditions in the fault diagnosis of diesel generator sets,and further improve the accuracy of fault discrimination.Compared with other four fault diagnosis methods,the results show that the proposed method has higher diagnosis accuracy and better robustness.