It has been proved that the incipient fault diagnosis plays a significant role in condition monitoring/health management.This decade has witnessed a huge development in transducer technology,electronic engineering and database technology,with tremendous amount of information collected and used to analysis the system health condition,which makes data-driven methods attract a lot attention and really make sense.This paper researches on the four-tanks system and satellite control system,concerning from diagnosability analysis to the improved diagnosis method.The main work includes the following three parts:(1)The diagnosability is the inherent property of a system,mainly including the fault detectability and isolability.Analysis of diagnosability evaluation based on K-L divergence is conducted in this paper,and its deficiency is analyzed.The model uncertainties and process noise are taken into consideration,and in this work we consider a method using Energy-statistics to evaluate the system diagnosability,propose the definition and the calculation of fault detectability and isolability,which overcomes the inavailability of K-L based method in low-dimensional cases.(2)Considering the regular faults in systems,we analyzed the principles of fault detection with PCA based methods,and the corresponding procedures are exhibited.The T~2 statistics and Q statistics are verified in fault diagnosis in nonlinear non-stationary system(four-tanks system).For incipient faults,the PCA method are proved to be inefficient in a simulation,for its amplitude is small.(3)Concerning the incipient fault exist in the system,the procedures of general process for incipient fault diagnosis is shown,as well as the noise-reduction based incipient fault diagnosis algorithm is proposed.Since the traditional diagnosis statistics is insensitive to incipient fault,a method combined with K-L divergence and principle component analysis is proposed.The case study shows that this method improves the detection ratio of incipient fault. |