摘要
针对传统深度学习故障诊断方法在滚动轴承中诊断效果不理想的问题,提出一种细菌觅食优化算法(BFO)优化卷积神经网络(CNN)学习率使诊断效果提升的模型。在模型逐次迭代过程中,将CNN中的学习率参数带入BFO中,生成一个自适应的学习率,用于更新CNN的权重和偏置,使模型故障诊断效果达到最佳。通过实验证明基于细菌觅食算法优化的卷积神经网络训练的模型在分类精度上优于CNN训练的模型,并与CNN多种学习率对比,可将故障诊断准确率提升至97.25%,并提高了全局的收敛能力。
Aiming at the problem that the diagnosis effect of traditional deep learning fault diagnosis method is not satisfactory in rolling bearings,a model of bacterial foraging optimization algorithm(BFO)optimization convolutional neural network(CNN)learning rate improves the diagnostic effect.During the model iteration,the learning rate parameter in the CNN is brought into the BFO to generate an adaptive learning rate that is used to update the weights and biases of the CNN to achieve the best model fault diagnosis.Through experiments,it is proved that the model trained by the convolutional neural network optimized by the bacterial foraging algorithm is better than the CNN-trained model in terms of classification accuracy,and compared with the multiple learning rates of CNN,the fault diagnosis accuracy can be improved to 97.25%,which improves the overall convergence ability.
作者
崔坤
刘美
高翔宇
孟亚男
张斐
CUI Kun;LIU Mei;GAO Xiangyu;MENG Yangnan;ZHANG Fei(Jilin Institute of Chemical Technology,Jilin 132022,China;Guangdong University of Petrochemical Technology,Maoming,Guangdong 525000,China;Dongguan University of Technology,Dongguan,Guangdong 523419,China)
出处
《自动化与仪器仪表》
2023年第7期235-239,共5页
Automation & Instrumentation
基金
国家自然科学基金面上项目(62073091)
广东省高校重点领域(新一代信息技术)专项(2020ZDZX3042)
东莞理工学院机器人与智能装备创新中心项目(KCYCXPT2017006)
广东省普通高校机器人与智能装备重点实验室项目(2017KSY009)
广东省普通高校特色创新项目(2017KTSCX176)
湖南省重点实验室开放基金项目(21903)。
关键词
细菌觅食算法
卷积神经网络
自适应学习率
故障诊断
bacterial foraging algorithms
convolutional neural network
adaptive learning rate
fault diagnosis