CN111260113A - Online prediction method of junction temperature in the whole life cycle of SiC MOSFET module - Google Patents

Abstract

The invention discloses a full-life-cycle junction temperature online prediction method for a SiC MOSFET module, which is implemented by the following steps: step 1, a brand-new SiC MOSFET module is adopted for power circulation until the module is completely aged and invalid; sampling electrical parameters of the SiC MOSFET module as a data set during a power cycle; step 2, establishing a junction temperature prediction model of the SiC MOSFET module, wherein the junction temperature prediction model comprises a first BP neural network model and a second BP neural network model which are connected in sequence; step 3, performing model training by adopting a data set pair; and 4, transplanting the trained junction temperature prediction model of the SiC MOSFET module into the FPGA, inputting current Id and outputting the corresponding junction temperature Tc in real time in the actual operation of the SiC MOSFET module to be tested. The invention can accurately obtain the junction temperature of the SiC MOSFET module and ensure the safe and stable operation of the belt module.

Description

Online prediction method of junction temperature in the whole life cycle of SiC MOSFET module

technical field

The invention belongs to the technical field of electronics, and relates to an on-line prediction method for junction temperature in the whole life cycle of a SiC MOSFET module.

Background technique

SiC MOSFET modules combine the advantages of high temperature resistance, high voltage resistance, fast switching speed, and low switching loss. With the development of science and technology, aerospace, communications, nuclear energy and many other fields are in urgent need of a power electronic device that can still work normally in high temperature, high frequency and other environments. The high temperature characteristics of SiC are obvious. The maximum temperature of the chip can reach 600℃, and the thermal breakdown junction temperature of the module can reach 250℃. However, due to the particularity of the operating environment of the device, a large amount of heat will be generated in the operating state, resulting in temperature rise and thermal stress deformation. Due to the long-term operation and continuous power cycle or thermal cycle, the aging speed of the module is accelerated, and long-term failure accumulation is formed, thereby reducing the reliability of the device and even the entire system. Usually, the junction temperature monitoring of power devices is carried out according to the state of health of the device. However, it is ignored that the internal parameters of the module change as the module gradually ages during work, and the prediction result is inaccurate.

SUMMARY OF THE INVENTION

The purpose of the present invention is to provide an online prediction method for the junction temperature of the SiC MOSFET module in the whole life cycle, which solves the problem of inaccurate prediction results of the junction temperature of the SiC MOSFET module in the prior art.

The technical scheme adopted in the present invention is,

The online prediction method of junction temperature in the whole life cycle of SiC MOSFET module is implemented according to the following steps:

Step 1, use a brand new SiC MOSFET module for power cycling until the module is completely aged and fail; during the power cycle, the electrical parameters of the SiC MOSFET module are sampled as a data set, and the electrical parameters are Vds, Id and R, where Vds is the saturation voltage. drop, Id is the current value, R is the resistance value;

Step 2, establish a junction temperature prediction model of the SiC MOSFET module. The junction temperature prediction model of the SiC MOSFET module includes a first BP neural network model and a second BP neural network model connected in sequence, and the input of the first BP neural network model corresponds to Vds, Id and R, the output of the first BP neural network model corresponds to the real-time power cycle number Nc, and the output of the first BP neural network model corresponds to an input of the second BP neural network model;

Step 3, using the data set to train the junction temperature prediction model of the SiC MOSFET module;

Step 4: Transplant the trained SiC MOSFET module junction temperature prediction model to the RAM of the FPGA. The FPGA is connected to the tested SiC MOSFET module. During the actual operation of the tested SiC MOSFET module, the current Id is input, and the corresponding Id is output in real time. Junction temperature Tc.

The present invention is also characterized in that,

The specific steps of sampling in step 1 are: set the current value Id of the DC power supply, the current value Id randomly selects the value in the range of [1,150], the junction temperature starts from 30 °C, and the electrical parameters are obtained by the method of interval sampling, and the sampling frequency is 2000 power cycles in the initial stage of aging and 1000 power cycles in the later stage of aging.

The number of neurons in the input layer of the first BP neural network model is 3, corresponding to Vds, Id and R respectively, and the number of neurons in the output layer is 1, corresponding to Nc.

The number of neurons in the hidden layer of the first BP neural network model is 7.

The number of neurons in the input layer of the second BP neural network model is 4, respectively corresponding to Vds, Id, R and the output of the first BP neural network model.

The number of neurons in the hidden layer of the second BP neural network model is 9.

The beneficial effects of the present invention are

The invention adds aging parameters to the SiC MOSFET module through power cycle sampling to compensate and correct the junction temperature prediction. By analyzing the experimental data with the BP network algorithm, the BP neural network has a strong nonlinear mapping ability, can approximate any nonlinear function with arbitrary precision, and has a strong self-learning and adaptive ability. junction temperature is predicted. Then the algorithm is transplanted into the RAM of the FPGA, and the online extraction of the junction temperature of the SiC MOSFET module is realized. Over-temperature protection of traditional SiC MOSFET modules is overcome, and the junction temperature of SiC MOSFET modules is accurately obtained. Ensure safe and stable operation of devices with SiCMOSFET modules.

Description of drawings

Fig. 1 is a flow chart of steps in the on-line prediction method of junction temperature in the whole life cycle of SiC MOSFET module of the present invention;

Fig. 2 is the network structure diagram of the first BP neural network model in the whole life cycle junction temperature online prediction method of SiC MOSFET module of the present invention;

FIG. 3 is a network structure diagram of the second BP neural network model in the online prediction method of the junction temperature of the whole life cycle of the SiC MOSFET module of the present invention.

Detailed ways

The present invention will be described in detail below with reference to the accompanying drawings and specific embodiments.

The on-line prediction method for junction temperature of SiC MOSFET module in the whole life cycle of the present invention is characterized in that it is specifically implemented according to the following steps:

Step 1, a brand new SiC MOSFET module is used for power cycling until the module is completely aged and fails; the electrical parameters of the SiC MOSFET module are sampled as a data set during the power cycle, and the electrical parameters are Vds, Id and R respectively; The specific steps are: set the current value Id for the DC power supply, and the current value Id randomly selects the value in the range of [1,150]. The junction temperature starts from 30°C, and the electrical parameters are obtained by the method of interval sampling. The sampling frequency is 2000 times in the early stage of aging. Power cycle, 1000 power cycles in the later stage of aging;

Step 2, establish a junction temperature prediction model of the SiC MOSFET module. The junction temperature prediction model of the SiC MOSFET module includes a first BP neural network model (as shown in Figure 2) and a second BP neural network model (as shown in Figure 3), which are connected in sequence. The input of the neural network model corresponds to Vds, Id and R, the output of the first BP neural network model corresponds to the real-time power cycle number Nc, and the output of the first BP neural network model corresponds to an input of the second BP neural network model;

Step 3, using the data set to train the junction temperature prediction model of the SiC MOSFET module;

Step 4: Transplant the trained SiC MOSFET module junction temperature prediction model to the RAM of the FPGA. The FPGA is connected to the tested SiC MOSFET module. During the actual operation of the tested SiC MOSFET module, the current Id is input, and the corresponding Id is output in real time. Junction temperature Tc.

The number of neurons in the input layer of the first BP neural network model in step 2 is 3, corresponding to Vds, Id and R respectively, and the number of neurons in the output layer is 1 corresponding to Nc; The number of neurons in the layer is 7.

The number of neurons in the input layer of the second BP neural network model is 4, corresponding to Vds, Id, R and the output of the first BP neural network model; the number of neurons in the hidden layer of the second BP neural network model is 9 indivual.

When training the junction temperature prediction model of the SiC MOSFET module, the first BP neural network model adjusts the weights and thresholds according to the error between the predicted output value and the expected value until the output value Nc approaches the expected value Nf, and the second BP neural network model according to the predicted output value The error between the value and the expected value adjusts the weight and threshold until the output value Tj approaches the expected value Tc;

Example 1

Step 1. For the 1200V/300A model BSM300D12P2E001 produced by ROHM, a brand new SiCMOSFET module, perform power cycling until the module is completely aged and fail; during the power cycle, the electrical parameters of the SiC MOSFET module are sampled as a data set, and the electrical parameters are respectively are Vds, Id and R, where Vds is the saturation voltage drop, Id is the current value, and R is the resistance value;

The specific sampling method is:

The current value Id is set for the DC power supply. The DC value is randomly selected in the range of [1,150]. The junction temperature starts from 30 °C, and the electrical parameters are obtained by the method of interval sampling. The sampling frequency is 2000 power cycles in the early stage of aging. The aging period is 1000 power cycles, and the sampled data is divided into training group data and test group data according to 3:1; the training group is used as the data set, and the test group data is used to verify the correctness of the prediction results of this method;

Step 2, establish a junction temperature prediction model of the SiC MOSFET module. The junction temperature prediction model of the SiC MOSFET module includes a first BP neural network model (as shown in Figure 2) and a second BP neural network model (as shown in Figure 3), which are connected in sequence. The input of the neural network model corresponds to Vds, Id and R, the output of the first BP neural network model corresponds to the real-time power cycle number Nc, and the output of the first BP neural network model corresponds to an input of the second BP neural network model;

The number of neurons in the input layer of the first BP neural network model is 3, corresponding to Vds, Id, and the number of neurons in the output layer is 1 corresponding to Nc; the number of neurons in the hidden layer is 7.

The number of neurons in the input layer of the second BP neural network model is 4, respectively corresponding to the outputs of Vds, Id, R and the first BP neural network model; the number of neurons in the hidden layer is 9.

Step 3, using the data set to train the junction temperature prediction model of the SiC MOSFET module;

Step 4: Put the trained SiC MOSFET module junction temperature prediction model into the RAM of the FPGA, and the FPGA is connected to the tested SiC MOSFET module. In the actual operation of the tested SiC MOSFET module, the input current Id=65A, through the SiCMOSFET module The output Vds=1.7008, and the corresponding junction temperature Tc=67.5℃ for real-time output.

Claims (6)

1. The method for online prediction of junction temperature in the whole life cycle of SiC MOSFET module is characterized in that, it is specifically implemented according to the following steps: Step 1, a brand new SiC MOSFET module is used for power cycling until the module is completely aged and fails; the electrical parameters of the SiC MOSFET module are sampled as a data set during the power cycle, and the electrical parameters are Vds, Id and R, where Vds is saturated Voltage drop, Id is the current value, R is the resistance value; Step 2, establish a SiC MOSFET module junction temperature prediction model, the SiC MOSFET module junction temperature prediction model includes a first BP neural network model and a second BP neural network model connected in sequence, and the input of the first BP neural network model corresponds to Vds, Id and R, the output of the first BP neural network model corresponds to the real-time power cycle number Nc, and the output of the first BP neural network model corresponds to an input of the second BP neural network model; Step 3, using the data set to train the SiC MOSFET module junction temperature prediction model; Step 4: Transplant the trained SiC MOSFET module junction temperature prediction model to the RAM of the FPGA. The FPGA is connected to the tested SiC MOSFET module. In the actual operation of the tested SiC MOSFET module, the input current Id is output in real time. The corresponding junction temperature Tc. 2. The method for online prediction of junction temperature in the whole life cycle of SiC MOSFET module as claimed in claim 1, wherein the concrete step of sampling in the step 1 is: set the current value Id in the DC power supply, and the current value Id randomly takes the value in the range of [1,150], the junction temperature starts from 30℃, and the electrical parameters are obtained by the method of interval sampling. The sampling frequency is 2000 power cycles in the early stage of aging and 1000 power cycles in the later stage of aging. 3. SiC MOSFET module full life cycle junction temperature online prediction method as claimed in claim 1, is characterized in that, the input layer neuron number of described first BP neural network model is 3, corresponds to Vds, Id and R, the number of neurons in the output layer is 1, corresponding to Nc. 4 . The method for online prediction of junction temperature in a full life cycle of a SiC MOSFET module according to claim 3 , wherein the number of neurons in the hidden layer of the first BP neural network model is 7. 5 . 5. SiC MOSFET module full life cycle junction temperature online prediction method as claimed in claim 1, is characterized in that, the input layer neuron number of described second BP neural network model is 4, corresponding to Vds, Id, R and the output of the first BP neural network model. 6 . The method for online prediction of junction temperature in a full life cycle of a SiC MOSFET module according to claim 1 , wherein the number of neurons in the hidden layer of the second BP neural network model is 9. 7 .

Images