Open-Circuit Fault Diagnosis Algorithm For Multi-phase Boost Converter

This thesis mainly studies on the problem of open circuit fault diagnosis for multiphase boost converters.Due to its structure and dynamical characteristics,multi-phase boost converters are widely applied in high-power applications such as wind turbines,photovoltaics,electric vehicles,etc.By parallel connection of identical single converter units,this topology has advantages of small current ripple,high transmission efficiency,and small power factor.However,when an open-circuit fault occurs in one phase of a multi-phase converter,the input current ripple will largely increase,reducing circuit performance significantly.Therefore,fault diagnosis is crucial for the stability of such structure.For multi-phase structures,most of the current diagnosis methods are only for single fault,this paper is dedicated to solve the problem of diagnosing multiple concurrent faults.Firstly,the four-phase boost converter is taken as an example to help explaining the structural characteristics,working principle as well as the change of circuit outputs during fault transition process,which provides priori knowledge for the design of fault diagnosis algorithm.A simulation model based on MATLAB as well as a hardware experimental platform were built for the purpose of effectiveness verification and performance comparison of the proposed methods.In this thesis a simple yet effective model-based open-circuit fault diagnosis method is proposed.A linear model is obtained according to the relationship between the input current and the phase current.Each time,by identifying the model parameters,all the branch statuses can be obtained in one switching cycle serving the goal of multiple concurrent fault diagnosis.In the process of modeling,a key step is to predict the branch current under normal conditions.According to prior knowledge of the circuit,two current prediction methods are proposed,one based on pre-stored current and the other output voltage.Experiments verified both methods on phase current estimation.Besides model-based methods,data driven have become popular in the field of fault diagnosis during recent years due to its advantages of not requiring accurate mathematical models and easy implementation.Based on the temporal correlation characteristic of the input data,a fault diagnosis method is proposed using recursive neural network(RNN).Both simulation and experiment show that the three proposed methods can diagnosis multiple concurrent faults quickly and accurately.At last performance comparisons of all three proposed methods are conducted based on indexes such as accuracy,recall rate and AUC.In addition,robustness of the proposed methods under different load,noise intensity and PWM control signal duty cycles are taken into account.According to the indexes taken into account,the data driven method performs better.However,the computational cost and required network training time of this method are larger than those of the model-based methods.Also,the performance increase provided by data driven method is not significant enough for that trade off.In practice,model-based methods are more applicable as they require less computing power,shorter diagnosis time and easier implementation.Last but not least,by applying an ubiquitous relation in multiple interleaved structures,model-based approaches are scalable and can be generalized to any other interleaved parallel DCDC converters without remodeling.The model-based approach proposed in this thesis can also be easily transplanted to other multi-phase interleaved structures.