针对滚动轴承故障振动信号非平稳性的特点,对二进制粒子群优化算法(binary particles swarm optimization,BPSO)和深度信念网络(deep belief network,DBN)进行研究,提出一种基于局部均值分解(local mean decomposition,LMD)和IBPSO-DBN的轴承故障诊断方法.提出用加权惯性权重改进BPSO迭代过程中的固定权重,再用改进BPSO优化DBN的隐含层神经元个数和学习率.该方法先对信号进行LMD,提取出各PF分量的散布熵和时域指标,并构建特征矩阵,然后把特征矩阵输入改进BPSO-DBN模型中训练,实现滚动轴承故障诊断和分类.采用试验轴承数据做验证并与其他诊断方法对比,结果表明,基于LMD和BPSO-DBN的滚动轴承故障诊断方法具有较好的故障识别率.
DBN is Optimized Based on Improved Binary Particle Swarm Algorithm Bearing Fault Diagnosis
Aiming at the characteristics of non-stationarity of rolling bearing fault vibration signals,binary particles swarm optimization(BPSO)and deep belief network(DBN)are studied,and a bearing fault di-agnosis method based on local mean decomposition(LMD)and IBPSO-DBN is proposed.The weighted inertia weights are proposed to improve the fixed weights in the BPSO iteration process,and then the num-ber of hidden layer neurons and learning rate of DBN are optimized by improving BPSO.In this method,the signal is LMD,the dispersion entropy and time domain indicators of each PF component are extracted,and the feature matrix is constructed,and then the feature matrix is input into the improved training in the BPSO-DBN model to realize rolling bearing fault diagnosis and classification.Using the test bearing data as verification and comparing with other diagnostic methods,the results show that the rolling bearing fault di-agnosis method based on LMD and IBPSO-DBN has a good fault identification rate.
local mean decompositionbinary particles swarm optimizationdeep neural networkrolling bearing fault diagnosis
NETL
NSTL
万方数据
维普
