ADCS-ELM算法滚动轴承故障诊断

Rolling bearing fault diagnosis based on ADCS-ELM algorithm

针对滚动轴承的故障信息难以从复杂噪声背景下的非平稳振动信号中提取且传统方法分类精度低等问题,提出一种基于集合经验模态分解(EEMD)能量特征提取和优化极限学习机神经网络(ADCS-ELM)分类诊断相结合的轴承故障诊断方法。利用EEMD对非线性和非平稳信号的自适应分解能力,将待检测轴承故障信号分解为包含故障特征的固有模态函数集(IMFs),并提取能量特征向量;利用自适应动态搜索步长改进布谷鸟搜索算法(ADCS)优化ELM网络连接权值和隐层阈值;将提取的故障特征向量用于训练极限学习机神经网络,得到最优权值和阈值;利用ADCS-ELM进行轴承故障诊断实验。实验结果表明:与BP,LVQ和ELM网络轴承故障诊断方法相比较,所提方法能够有效提高故障识别准确率,并且具有更快的计算速度。

Aiming at the problem that the fault information of rolling bearings is difficult to extract from the non-stationary vibration signals from complex noise background,and the poor classification precision of traditional methods,a fault diagnosis method of rolling bearings based on energy feature extraction of ensemble empirical mode decomposition(EEMD)and classification diagnosis of an optimized extreme learning machine(ADCS-ELM)is proposed.Firstly,based on the adaptive decomposition ability of EEMD to nonlinear and non-stationary signals,the bearing fault signals are decomposed into an intrinsic mode function set(IMFs)which containing full of fault features,and the fault feature vectors are extracted.Then,the cuckoo search algorithm(ADCS),which is optimized by an adaptive dynamic search step strategy,is used to determine connection weights and the hidden layer threshold of the ELM.The optimal weights and thresholds can be obtained by training the ELM with the extracted fault feature vectors.Finally,the bearing fault diagnosis simulation experiments are carried out based on the ADCS-ELM to verify the effectiveness of the proposed method.Results show that,compared with the BP,LVQ and ELM methods,ADCS-ELM can improve the classification accuracy of the four typical bearing faults,and also have faster diagnosis speed.