摘要
针对滚动轴承早期故障时间点难以检测问题,提出一种基于卷积深度置信网络(convolutional deep belief network, CDBN)与多核极限学习机自编码器(multi-kernel extreme learning machine based autoencoder, MKELM-AE)的无监督健康指标故障检测方法。首先,引入CDBN自适应提取滚动轴承健康状态振动信号频谱的深层高维特征信息,去除高维数据冗余信息后得到表征滚动轴承健康状态的低维特征;然后,采用粒子群优化(particle swarm optimization, PSO)的MKELM-AE对提取的低维特征进行重构训练;最后,将待测信号输入训练好的CDBN-MKELM-AE模型中计算重构误差作为反映滚动轴承退化的健康指标,并采用Bootstrap Pettitt异常检测方法检测待测健康指标发生突变的时间。实验结果表明,所提方法建立的健康指标能反映轴承退化的不同阶段,可有效检测出早期故障中健康指标发生突变的时间,定位早期故障点。
To address the problem of difficult detection of early failure time points of rolling bearings,an unsupervised health indicator fault detection method based on convolutional deep belief network(CDBN)and multi-kernel extreme learning machine based autoencoder(MKELM-AE)was proposed.Firstly,CDBN was introduced to adaptively extract the deep high-dimensional feature information from the vibration signal spectrum of rolling bearing health condition,and the low-dimensional features representing rolling bearing health condition were obtained after removing redundant information of high-dimensional data.Then,MKELM-AE with particle swarm optimization(PSO)was used to train the extracted low-dimensional features.Finally,the signal to be measured was input into the trained CDBN-MKELM-AE model to calculate the reconstruction error as the health indicator reflecting the rolling bearing degradation,and the Bootstrap Pettitt anomaly detection method was used to detect the time of abrupt change of the health indicator to be measured.The experimental results show that the health indicator established by the proposed method can reflect the different stages of bearing degradation,and can effectively detect the time of abrupt change of health indicator in early failure and locate the early failure point.
作者
肖飞
马萍
张宏立
王聪
XIAO Fei;MA Ping;ZHANG Hongli;WANG Cong(School of Electrical Engineering,Xinjiang University,Urumqi 830017,China)
出处
《组合机床与自动化加工技术》
北大核心
2024年第6期151-155,160,共6页
Modular Machine Tool & Automatic Manufacturing Technique
基金
国家自然科学基金资助项目(52065064,52267010)
新疆维吾尔自治区自然科学基金资助项目(2022D01E33,2022D01C367)。
关键词
滚动轴承
早期故障预测
卷积深度置信网络
多核极限学习机
健康指标
rolling bearing
early fault detection
convolutional deep confidence network
multicore extreme learning machine
health indicator
