Research On Grid Inverter Fault Diagnosis Based On Bond Graph And The BP Neural Network

With the development of semiconductor technology and upgrade of power electronic devices, ac motor variable frequency speed regulation system get more and more widely used in industrial production, at the same time for reliability and security of the variable frequency speed regulation system is also more and more high expectations and demands are put forward. A large number of data show that the failure of switching devices in inverter is main reason leading to the invalid variable frequency speed regulation system. In fault occurs, can quickly and accurately judge the fault position and the cause of the problem, and selectively removed fault element is to guarantee the safe operation of the system one of the most effective way. In view of the nonlinear and strong coupling inverter, this paper puts forward the bond graph model is applied to fault diagnosis of inverter, the main research content is mainly from the three-phase grid inverter of the two aspects of modeling and fault diagnosis, in view of the switch power devicesFirst of all in the modeling method on the selection, considering high-power inventers contain both continuous dynamics and discrete dynamics, which take the system on complex mixed properties owning to the above two characteristics. In order to realize fault diagnosis for hybrid system, the model must cover the two characteristics, namely the continuous dynamics and discrete dynamics. Thus, for the modeling of this kind of system, it is very complex and liable to make mistakes if using general mathematical method. Besides, the inhomogeneity of mathematics due to the isomerism of physical system often slip up, even can’t solved by numerical method. While, the method—divide the overall system into multiple subsystems and each subsystem is modeled by ordinary principles of mechanics or electromagnetism, then assemble these subsystems into unified equations and then transform into state equations is also hard to realize in technology, vulnerable to associated variables and missing modality. To make this problem come out effectively and guarantee the model can response its performance accurately, in this paper, from the view of energy, put forward a new bond graph-based modeling and NN approach. As for switched power devices, a new concept named switched power junction(SPJ) is presented and a model is developed for the inverters analysis using SPJ.Secondly, taking three-phase grid-connected inverters for subjects, in this paper the bond graph model based on SPJ was established and also carried out simulation on 20-sim. The response curves of the system were also obtained without any faults of each power devices. In this case, the unique superiority of BG-based modeling, which regarded as some advanced modeling tool in three-phase grid-connected inverters, was testified.At last, the fault diagnostic methods for switching power devices, mainly focused on stuck-open faults and short-circuit faults, designed and described bond graph-based modeling and NN under. Based on bond graph model, the export system time causality diagram, a term used to describe the causal relationship between the system variables. According to causality analytical redundancy relationship of each node is deduced, and then deduces each switch inverter device fault characteristic matrix, through the fault of analytical redundancy relationship caused by the change of fault characteristic matrix and the switching device failure characteristic matrix is thus concluded that failure and their status. Due to the fault characteristics of the same bridge arm up and down, so rely on analytical redundancy relations to analyze the failure state of the system and the cause of the problem was more difficult, so as to put forward the application of neural network to analyze the extracted data, fault type and fault location are obtained. This diagnosis method is verified by simulation and the feasibility and correctness of the compared with qualitative methods, this method has more.In this paper, not only for bond graph theory and other methods in power electronic system of the combination of fault analysis to broaden the vision, also for bond graph theory in the modeling of hybrid system provides the basis...