Research On The Design And Operation Reliability Of Power Electronic Conversion Devices

With the gradually deepening of green,low-carbon,efficient and energy-saving concepts,new energy power generation,electric vehicle charging,bidirectional DC/DC conversion and other technologies which take power electronic conversion as the core have been widely used.The reliability of power electronic conversion device would affect the safe and stable operation of the whole equipment.This paper takes the typical buck converter as the research object and improve the reliability of power electronic converter from two aspects of design reliability and operation reliability.This paper first adopts the life prediction method based on the data model to evaluate the reliability of the circuit in the design stage,which is based on the reliability prediction manual of the electronic equipment,and combined with the components selection,parameters selection and operation conditions of the circuit to forecast the theoretical operation life of the circuit reasonably for analyzing the reliable working ability of the designed circuit and providing guidance for the reliability analysis and design optimization of power electronic conversion device in the design stage.Power semiconductor device is the most vulnerable component in the power electronic conversion device,and the temperature fluctuation in the operation is the key factor to affect its reliable operation.Therefore,this paper achieved to estimate the case temperature of power devices on-line and propose a method of case temperature estimation method based on Artificial Neural Network whose model is developed and trained by using VS Code software and Tensor Flow framework in Python environment,and then the designed network model is implemented in DSP controller.The temperature value estimated by this method can provide an important reference for the adjustment of converter operation strategy.Finally,the hardware test platform of Buck converter is built and arranged for testing whose results prove the correctness of the key steps in the process of reliability analysis;The network model is trained and tested with the measured off-line data in the testing process;The training method of extreme point combination is put forward in which the accurate fitting of neural network under various working conditions can be realized by using only a small amount of data;The trained neural network model is transplanted into the DSP controller whose experimental results under different working conditions prove the feasibility of online case temperature estimation.