Complex Dynamic Engineering System Codition Monitoring Research Using Cointegration Theory

The complex structure and dynamic feature of modern engineering system cause more trouble for the condition monitoring (CM) task than ever. This is because those well developed monitoring method based on signal analysis techniques can not provide the status information of whole system since they usually deal with single variable. And those strategies based on system model face the difficulty of building an accurate model to characterize the system behavior. To achieve the CM task of complex engineering systems, this paper creatively applies the cointegraton theory which is form econometrics to the CM task.According to the cointegration theory, there could be a long-run equilibrium between the nonstationary system variables. In that case, these stochastic variables follow a long term common trend in despite of individual nonstationarity. The deviation of variable level from this common trend is stationary. This equilibrium is called cointegration relation, and can be described as a simple linear combination form. The cointegraton relation is determined by the system's inherent mechanism. Consequently, if the cointegration relation can be found between the engineering system variables, the cointegration model residuals (innovations) then are able to detect the relation change caused by system faults.Based on this idea, this paper explores a new CM method based on basic cointegration concept and test method. For the building of CM model, two different methods are researched. One is building CM model based on single cointegrating vector which is according to the maximum eigenvalue. And the other one is building CM model through linear combination of multi cointegrating vectors under maximum entropy criterion. Subsequently, CM model residuals are analyzed both in time domain statistically and in frequency domain for signal signature.The analysis results show that, when the system condition is normal, the CM model residuals fluctuate around mean zero randomly. But when faults occur, the CM model residual's dynamic characters change obviously. Furthermore, each type of system fault shows a unique pattern of influence on the model residuals. These evidences prove that the cointegration based method is feasible for complex engineering system's CM task. During this research, a hydraulic flap servo control system is designed and simulated. This simulation model is employed to play the role of a complex engineering system which the CM method is applied to. The simulation model provides not only the system process variable data to build the CM model, but also a platform to test it. Therefore, the hydraulic simulation model is also an important part of this research project.