动态鉴别结构保持投影的故障诊断方法

Fault diagnosis method based on dynamic discriminant structure preserving projection

针对动态稀疏保局投影考虑数据的局部相似度信息,忽视局部差异度信息的问题,在综合考虑数据相似度信息和差异度信息的基础上,融合数据类标信息,提出了一种新的基于动态鉴别结构保持投影的故障诊断方法。该方法首先构造原始数据的扩展矩阵,利用稀疏表示获取数据的全局稀疏重构关系,并融入到相似度信息中;同时,考虑数据差异度信息和鉴别信息,构建的结构保持投影目标函数进行数据降维,最后利用鉴别函数值进行故障诊断。田纳西-伊斯曼过程的仿真结果表明,与保局鉴别分析方法相比,所提方法具有更好的诊断精度和更强的稳定性。

Dynamic sparse locality preserving projection （SLPP） focusing on the local similarity information of data, neglecting the diversity information, the proposed dynamic discriminant structure preserving projection （DSPP） approach for fault diagnosis takes into the similarity and diversity information account, and integrates the labels of data. In this study, the augmented matrix is firstly constructed. Then, the global sparse reconstructive relationship of data derived from sparse representation is integrated with the similarity information. Meantime, the diversity information and discriminant information are also combined to form the objection function of DSPP for dimension reduction. Finally, the value of discriminant function is obtained for fault diagnosis. The simulation results of Tennessee Eastman process demonstrate that the proposed method has better diagnosis accuracy and stronger stability than the locality preserving discriminant analysis method.