庞皓乾,张攀,王文,等.基于降噪–残差神经网络的发动机部分失火故障诊断[J].内燃机工程,2023,44(3):91-100.
基于降噪–残差神经网络的发动机部分失火故障诊断
Partial Misfire Fault Diagnosis of An Engine Based on Noise Reduction–Residual Neural Network
DOI：10.13949/j.cnki.nrjgc.2023.03.011
关键词:部分失火  故障诊断  振动信号  小波阈值降噪  残差神经网络  短残差块
Key Words:partial misfire  fault diagnosis  vibration signal  wavelet threshold noise reduction  residual neural network  short residual blocks
基金项目:国家自然科学基金重点项目（51636005）
作者单位E-mail
庞皓乾* 天津大学 内燃机燃烧学国家重点实验室，天津 300354 panghq@tju.edu.cn 
张攀 天津大学 内燃机燃烧学国家重点实验室，天津 300354  
王文 中国重汽集团青岛重工有限公司，青岛 266111  
王彦军 天津大学 内燃机燃烧学国家重点实验室，天津 300354  
邹佳华 天津大学 内燃机燃烧学国家重点实验室，天津 300354  
高文志* 天津大学 内燃机燃烧学国家重点实验室，天津 300354 gaowenzhi@tju.edu.cn 
摘要点击次数: 836
全文下载次数: 547
摘要:针对发动机单缸部分失火故障，提出基于小波阈值降噪和残差神经网络的“降噪–残差神经网络”故障诊断方法。通过降噪与深度学习算法相结合，将小波阈值降噪后的振动信号输入到残差神经网络进行故障诊断；使用短残差块进一步防止网络的退化，并利用大卷积核增大长数据输入的卷积视野，提高信号故障特征的提取能力。测试结果证明该方法不仅实现了未参与训练的运转工况97%以上的故障诊断准确率，而且对于加入高斯噪声后的含噪声信号也能实现较高的诊断准确率。通过与其他故障诊断网络进行对比证明了该方法的优越性。
Abstract:A fault diagnosis method of “noise reduction–residual neural network” based on wavelet threshold denoising and residual neural network was proposed for partial misfire fault diagnosis of engine cylinders. Combining the noise reduction and deep learning algorithm， the signal was denoised by wavelet threshold and fed into the residual neural network for fault diagnosis. Unlike the previous residual network， the short residual block would be utilized to further prevent network degradation. Besides， the large convolution kernel was also used to expand the convolution field of long data input and improve the ability to extract fault characteristics. Experimental results show that this method can not only achieve more than 97% fault diagnosis accuracy for the operation conditions without training， but also achieve high diagnosis accuracy for the noisy signals with Gaussian noise. The performance of the proposed method is proved to be more superiority and excellent than those of other misfire fault diagnosis algorithms.
查看全文  HTML   查看/发表评论