摘要
针对电子设备的健康性能退化问题,提出一种改进流形学算法与隐半马尔可夫模型(hidden semi-Markov model,HSMM)相结合的电子设备健康评估与故障预测方法。首先,在有监督邻域保持投影(supervisedneighborhood preserving projection,SNPP)算法中引入非相关约束并加入核函数形成核有监督非相关邻域保持投影(kernel supervised uncorrelated neighborhood preserving projection,KSUNPP)算法,将其用于原始特征的提取,获得有效的特征集作为HSMM的输入进行训练;其次,建立了电子设备健康评估与故障预测模型,该模型用Kullback-Leibler(KL)距离来衡量故障程度,实现设备退化程度的评估,又可根据各状态驻留时间,预测出设备故障发生的时间。最后,将该方法应用于某型导弹电子设备的健康评估与故障预测,验证其有效性。
To deal with the health performance degradation of electronic equipment,a new health evaluation and fault prognostics method based on improved manifold learning algorithm and hidden semi-Markov model（HSMM） is proposed.Firstly,according to the supervised neighborhood preserving projection（SNPP） algorithm,a kernel supervised uncorrelated neighborhood preserving projection（KSUNPP） algorithm is proposed by introducing an uncorrelated constraint and kernel method,and the improved algorithm is used for feature extraction.Secondly,the health evaluation and fault prognostics model of electronic equipment is constructed.Then,by calculating Kullback-Leibler（KL） distance which can measure the fault degradation,the model can evaluate the health performance degradation.And according to the dwell time of every state,it can also predict the time that faults occur.Finally,the proposed method is applied to the health evaluation and fault prognostics of electronic equipment of a certain type of missile.Experiment results demonstrate that the method is effective.
出处
《系统工程与电子技术》
EI
CSCD
北大核心
2012年第5期1068-1072,共5页
Systems Engineering and Electronics
基金
国家自然科学基金(60971118)资助课题
关键词
健康评估
故障预测
流行学习
特征提取
隐半马尔可夫模型
health evaluation
fault prognostics
manifold learning
feature extraction
hidden semi-Markov model
