摘要
针对传统单域特征指标无法充分表征轴承性能退化的状态信息,而基于多域高维特征向量的重构评估模型存在信息冗余且易受到不一致优化目标的影响而导致模型次优性能的问题,提出一种基于多元状态估计(multivariate state estimation technique, MSET)重构模型整体优化的轴承性能退化评估方法。首先,提取轴承振动信号的多个时域和频域特征、自回归模型系数和三层小波包Renyi熵组成高维多域特征向量,同时将健康状态的高维特征向量构建MSET重构模型的历史记忆矩阵;然后,利用遗传算法对轴承高维特征向量和MSET模型中的历史记忆矩阵进行同步联合优化,从而实现特征优选和重构评估模型的整体自适应优化,进一步提高降维后特征向量与重构模型的匹配性;最后,利用余弦相似度作为故障程度指标构建轴承性能退化评估曲线。西安交大-昇阳科技联合实验室滚动轴承疲劳试验全寿命数据分析结果表明,所提方法具有一定的有效性和可靠性。
Aiming at the problem that the traditional single domain feature index can not fully represent the state information of bearing performance degradation,and the reconstruction evaluation model based on multi-domain high-dimensional feature vector has information redundancy and is vulnerable to the influence of inconsistent optimization objectives,which leads to the suboptimal performance of the model,a bearing performance degradation evaluation method based on the global optimization of multivariate state estimation technique(MSET)reconfiguration model was proposed.Firstly,several time and frequency domain features of bearing vibration signals,autoregressive model coefficients and three-layer wavelet packet Renyi entropy were extracted to form high-dimensional multi-domain feature vectors.At the same time,the high-dimensional feature vectors of health state were used to construct the historical observation matrix of the MSET model.And then the genetic algorithm was used to synchronously optimize the high-dimensional feature vector of the bearing and the history memory matrix of the MSET model,so as to realize the overall adaptive optimization of feature selection and the reconstruction evaluation model,and further improve the matching between feature vector and the reconstruction model after dimensionality reduction.Finally,cosine similarity was used as the fault degree index to construct the bearing performance degradation evaluation curve.The analysis of the whole life data of bearing fatigue test provided by Xi’an Jiaotong University-Shenyang Science and Technology Joint Laboratory show that the method proposed in this paper is effective and reliable.
作者
张龙
刘杨远
吴荣真
王良
承志恒
颜秋宏
ZHANG Long;LIU Yangyuan;WU Rongzhen;WANG Liang;CHENG Zhiheng;YAN Qiuhong(Key Laboratory of Transportation Tools and Equipment,Ministry of Education,East China Jiaotong University,Nanchang 330013,China)
出处
《振动与冲击》
EI
CSCD
北大核心
2023年第16期251-261,共11页
Journal of Vibration and Shock
基金
江西省自然科学基金(20212BAB204007)
江西省教育厅科学技术研究项目(GJJ200616)
江西省研究生创新专项资金(YC2021-S422)。
关键词
多域特征向量
多元状态估计重构模型
历史记忆矩阵
遗传算法
同步联合优化
multi-domain feature vector
reconstruction model of multivariate state estimation technique
historical memory matrix
genetic algorithm
synchronous joint optimization
