摘要
针对在RV减速器往复运动过程中所采集的振动信号干扰大,传统滤波方法过分依赖专家经验以及参数选择困难等问题,提出一种基于卷积自编码的生成对抗网络(Convolutional Auto-encoder GAN,CAE-GAN),应用于RV减速器振动信号降噪。首先,针对生成对抗网络(Generative Adversarial Networks,GAN)训练时收敛困难的问题,通过引入距离函数改进生成器的损失函数,提高模型的稳定性。其次,引入跳跃连接改进生成器的网络结构,在增强模型收敛能力的同时,进一步提升模型的降噪性能。最后,使用RV减速器振动数据对所提方法进行验证。实验结果表明:所提方法具有更好的降噪性能且能够提高故障诊断准确率。
Aiming at the problems of large interference in the collected vibration signals,excessive reliance on expert experience in traditional filtering methods,and difficulty in parameter selection,A method based on Convolutional Autoencoder GAN(CAE-GAN)was proposed to denoise the vibration signals of RV reducers during reciprocating motion.Firstly,in response to the difficulty of convergence in training Generative Adversarial Networks(GANs),the stability of the model was improved by introducing a distance function to improve the loss function of the generator.Secondly,the skip connections was introduced to improve the network structure of the generator and enhance the model's convergence ability,so as to further enhance the noise reduction performance of the model.Finally,the proposed method was validated using RV reducer vibration data.The experimental results show that the proposed method has better noise reduction performance and can improve the accuracy of fault diagnosis.
作者
范啸宇
刘韬
王振亚
陈朝阳
王亚南
王贵勇
FAN Xiaoyu;LIU Tao;WANG Zhenya;CHEN Zhaoyang;WANG Yanan;WANG Guiyong(Collage of Mechanical and Electrical Engineering,Kunming University of Science and Technology,Kunming 650500,China;Inner Mongolia First Machinery Group Co.,Baotou 014000,Neimenggu,China)
出处
《噪声与振动控制》
北大核心
2025年第5期84-91,共8页
Noise and Vibration Control
基金
云南省重大科技专项计划资助项目(202202AC080003)
云南省教育厅科学研究基金研究生类资助项目(2024Y128)。
