摘要
传统故障诊断模型训练时易陷入局部最优、模型泛化能力差,且故障识别精度易受人工特征提取质量的影响,针对这一问题对滚动轴承故障诊断方法进行了研究。首先,提出了基于深度置信网络(DBN)的滚动轴承故障诊断模型,研究了DBN模型的逐层自适应特征提取能力;然后,提出了一种改进的混合蛙跳算法(ISFLA),对DBN各隐含层神经元个数和反向微调算法学习率进行了优化;最后,在不进行任何特征提取的情况下,利用美国凯斯西储大学的轴承数据集进行了实验研究,提取了原始时域振动信号,进行了故障特征分析,并与BP、DBN和PSO-DBN算法进行了对比。研究结果表明:与其他方法相比,ISFLA-DBN的故障识别精度最高,算法收敛速度最快,模型泛化能力最好。
Aiming at the problems that the traditional fault diagnosis model was easy to fall into local optimality,the model generalization ability was poor,and the fault recognition accuracy was easily affected by the quality of artificial feature extraction,the fault diagnosis method of rolling bearing was studied.Firstly,a fault diagnosis model of rolling bearing based on deep belief network(DBN)was proposed and the hierarchical adaptive feature extraction capability of DBN model was studied.Then,an improved shuffled frog algorithm(ISFLA)was designed to optimize the number of neurons in each hidden layer and the learning rate of reverse fine-tuning algorithm.Finally,the bearing data set of Case Western Reserve University was used for experimental research without any feature extraction of the data.In the experiment,the original time-domain vibration signal was extracted,and the fault features were analyed and compared with the BP,DBN and PSO-DBN algorithms.The results show that ISFLA-DBN has the highest fault recognition accuracy,the fastest algorithm convergence rate,and the best model generalization ability.
作者
齐洪方
黄定洪
QI Hong-fang;HUANG Ding-hong(School of Intelligent Manufacturing,Wuhan Huaxia University of Technology,Wuhan 430223,China;School of Mechanical and Electronic Engineering,Wuhan University of Technology,Wuhan 430070,China)
出处
《机电工程》
CAS
北大核心
2021年第1期62-68,共7页
Journal of Mechanical & Electrical Engineering
基金
湖北省自然科学基金资助项目(2019CFC911)
湖北省高等学校优秀中青年科技创新团队计划资助项目(T201837)
武汉华夏理工学院校级科研基金资助项目(18016)。
关键词
滚动轴承
故障诊断
深度置信网络
改进混合蛙跳算法
rolling bearing
fault diagnosis
deep belief network(DBN)
improved shuffled frog leaping algorithm(ISFLA)
