Research On Current Balance Model And Artificial Neural Network Control For Parallel-Connected IGBTs

The parallel-connected Insulated Gate Bipolar Transistor(IGBT)typology has become a common and effective way in extending power rating of IGBTs based power converters.However,current difference among IGBT branches is a main and critical problem because it induces device reliability issues,such as device fail earlier than its baseline lifecycle.The root cause of current imbalance is device fabrication or assembly processes induced physical difference and power circuit parasitic parameters induced difference in application.For instance,most of the spec limit and control limit of fabrication or assembly processes have a range,so even the devices in a same manufacturing lot have minor differences,the impact are strong enough to cause current differences.The same situation can also be find in IGBT based application circuits,such as the variance of parasitic parameters caused by layout and so forth.Eliminating the imbalance current of parallel-connected IGBTs has been a research hotspot in the world.However,there is no any reports that measure the IGBT current,including on-state current and transient current,by Virtual Sensing(VS)method,such as VS method based on Artificial Neural Network(ANN)and so on.For current balance control,there are no reports that solve this kind of problem by artificial intellegience as well,such as artificial neural network based PID(ANN-PID)and so on.This dissertation goes deep research in current balance model and ANN control for parallel connected IGBTs.The innovated solutions are proposed for both of the imbalance current measuring and controlling,including current balance virtual sensing(BCVS)model of parallel connected IGBTs by VS method,the ANN-PID control method for eliminating imbalance current and the gate quantity of electric charge regulator(GQR)technology and pototype design based on proposed model,method and technology to eliminate imbalance current.The research content in this dissertation is list as below:(1)The BCVS model of balance current for parallel connected IGBT is firstly propsed and the VS method is also established for IGBT global current measuring,including turn-on transient,on-state and turn-off transient.With establishing the relationship between IGBT on-state current and gate charge Q_G,derivation of the IGBT switching state-space equation and consideration of IGBT analytical model,the analytical virtual sensing(AVS)model can be obtained.The IGBT current related paramters that are easy to measure,can be screened out from AVS.The empirical virtual sensing(EVS)model of IGBT current can be expressed by the related parameters and the BCVS model can be found based on the EVS model.The BC model or EVS model can be solved by artificial intelligence(AI)means,such as back propagation ANN(BPANN)and K-means clustering in machine learning algorithm,to find the pure mathematical relationship between related parameters and IGBT current in a particular IGBT based power system.The BCVS model can be used to predict the IGBT current or the imbalance(balalance)current of parallel-connected IGBTs.The experimental results show that the prediction error is about 3%by average.The hardware based VS algorithm has also been developed and qualified on Field Prgrammable Gate Array(FPGA)platform.It results 0.6%prediction error by average with 100MHz machine frequency and it cosumes 310ns when performs the VS algorithm each time.The VS method can eliminate the physical current sensors in actual application engineering and saves cost.The VS real-time performance is determined by the speed of related parameters sensing circuits,computing resources and so on.(2)A new ANN control method is propsed for current balance control of parallel-connected IGBTs.At present,the common used methods for current balance control are switching time delay and gate volage compension.In this dissertation,the artificial neural network based PID(ANN-PID)algorithm is implemented for imbalance current compensation control because of its self-adaptive and real-time characteristics.The hardware based algorithm has also been developed and qualified on FPGA platform.It results 0.023%control error by average with 200MHz machine frequency and it comsumes 240ns when performs the ANN-PID algorithm each time.In actual application,the algorithm calculation period is also determinded by computing resource without change the architecture of algorithm.(3)A gate quantity of electric charge regulator(GQR)technology is proposed to regulate the IGBT current.GQR is one of the active gate control(AGC)technology.The realization of GQR is easy enough for both of the discrete components and integrated circuits.The simulation and experimentation prove the effectiveness of GQR in IGBT current regulating,such as di/dt,dv/dt regulation,over-shoot volage and current suppression,oscillation suppression and on-state current regulation.(4)The prototype of current balance control for parallel-connected IGBTs is designed and crafted.It realized ANN-PID control and GQR technology with reservation of data acquiring circuits for VS.The experimentation shows that the imbalance current is surppressed almost 7 times.The performance is improved about 3 to 23 times within the load current range of IGBT in the experimentation.