故障特征基于多源信息和约束条件的多尺度诊断方法

项目名称： 故障特征基于多源信息和约束条件的多尺度诊断方法

项目编号： No.60801048

项目类型： 青年科学基金项目

立项/批准年度： 2009

项目学科： 数理科学和化学

项目作者： 文传博

作者单位： 上海电机学院

项目金额： 20万元

中文摘要： 本项目以旋转机械为研究对象，开展故障诊断方法的研究。主要包括：1）分析传动链的相互关系，给出系统的约束条件；2）针对受约束模型，给出了一种多尺度约束滤波算法，通过对系统状态的分块和小波变换，建立一个包含约束条件的新的动态系统，然后对小波系数进行滤波，最后通过重构得到原系统的估计值 ；3）在对永磁同步电机磁链的检测中，提出了一种基于卡尔曼滤波算法和小波变换的多尺度在线监测算法。它结合了卡尔曼滤波的实时性和小波变换的多尺度性，通过测量电机的端电压和流过定子线圈的电流在线估计电机永磁体的磁场状况；4)当有多个传感器对系统进行测量时，先根据传感器采样特点进行归类，然后对每类分别进行约束滤波，最后再把不同类的融合结果融合在一起；5)在验证算法时，考虑故障样本数与非故障样本数不均衡这一情况，通过基于Relief的Easyensemble算法来解决故障诊断中的这类，新方法提高了分类器在不均衡数据集中的分类性能和预报能力。

中文关键词： 故障诊断；状态估计；小波变换；多传感器；多尺度分析

英文摘要： This project treats rotating machinery as a research object, and gives some methods of fault diagnosis, which are summarized as follows. 1)Analysis the relationship of neighboing two parts and a equality constraint condition is given. 2)A Multiscale constrained filtering method is proposed to deal with these constrained problems. The original state is rewritten as blocked form and wavelet transform are used to obtain new dynamic system containing constrained information. The wavelet coefficients are estimated with measurements and finally obtain the estimate results of original state by using inverse wavelet transform.3)A new method based on Kalman fitler and wavelet technique is proposed. By monitoring the voltages and currents of the stator, it is possible to estimate the permanent magnet flux linkage on-line. 4）When several sensors are used together to measument the object, we firt set the sensors in different gather according to the sampling rate, then, develop the constrained filter for every sort, and finally fuse these constrained esitmate results.5) In imbalanced problem, some features are redundant or even irrelant. We propose REE(Rlief based feature selection for Easyensemble) to solve the imbalaced problem in the fault diagnosis of gear. The new method improve the classification performance and prediction ability on the imbalanced data set.

英文关键词： Fault Diagnosis;State Estimate;Wavelet Transform;Multi sensors;Multiscale anslysis