A reliability analysis method for DC chargers affected by temperature

Abstract

The present invention discloses a reliability analysis method for a DC charger under the influence of temperature. First, the heat dissipation system of the DC charger is analyzed to determine the ventilation volume and the air volume of the cooling fan when the DC charger is fully loaded. Then, a simulation analysis is performed on the DC charger to obtain a temperature distribution cloud map when the DC charger is fully loaded. Based on the temperature distribution cloud map, the weak links of the DC charger are analyzed. Based on the thermal simulation results, the maximum internal temperature of the DC charger under full load is obtained. This temperature is used as the test stress value of the newly developed DC charger, and an accelerated life test is designed to obtain test data. The reliability analysis method for a DC charger under the influence of temperature provided by the above technical solution can effectively solve the problem that newly developed DC chargers currently lack analysis methods and are difficult to predict the life of newly developed DC chargers.

Description

Technical Field

[0001] The present invention belongs to the technical field of electronic equipment reliability research, and in particular relates to a reliability analysis method for a DC charger under the influence of temperature. Background Art

[0002] The DC charger is small in size, light in weight, easy to move and carry. It is easy to operate, has a fast charging speed, high charging recovery efficiency, and no risk of overcharging even if the charge exceeds the limit. It can be used in various computer rooms.

[0003] Currently, research on chargers is still in its early stages. For newly developed DC chargers, in particular, historical data is scarce, and the lifespan of newly developed DC chargers needs to be predicted based on the limited available data. This paper proposes a reliability analysis method for newly developed DC chargers that can identify device weaknesses under temperature stress and predict device lifespan. Summary of the Invention

[0004] The purpose of the present invention is to provide a reliability analysis method for a DC charger under the influence of temperature, so as to solve the problem that newly developed DC chargers currently lack analysis methods and are difficult to predict the life of newly developed DC chargers.

[0005] To achieve the above object, the present invention provides the following technical solutions:

[0006] A reliability analysis method for a DC charger under the influence of temperature, the specific steps are as follows:

[0007] Step 1: Analyze the heat dissipation system of the DC charger and select an appropriate heat dissipation method;

[0008] Step 2: Determine the ventilation volume of the DC charger when it is fully loaded, and determine the layout of the heat dissipation equipment corresponding to the selected heat dissipation method based on the ventilation volume;

[0009] Step 3: Perform simulation analysis on the DC charger to obtain a temperature distribution cloud diagram when the DC charger is fully loaded.

[0010] Step 4: Analyze the weak links of the DC charger based on the temperature distribution cloud map. Then, based on the thermal simulation results, determine the maximum internal temperature of the DC charger when operating at full load.

[0011] Step 5: Use the highest temperature as the test stress value of the newly developed DC charger and design an accelerated life test to obtain test data.

[0012] Step 6: Use BP neural network and Bootstrap method to expand the small sample data, combine with least squares method to estimate the distribution function parameters, calculate the mean time between failures reliability index, and finally, use Bootstrap method for sampling again to calculate the confidence interval and verify the effectiveness of the expansion method and calculation results.

[0013] Among them, step one specifically includes the following steps:

[0014] according to The formula is used to calculate the heat flux density of the heat dissipation surface of the DC charger in steady state. Where A is the total heat dissipation area of ​​the charger, in cm 2 ;

[0015] Compare the calculated heat flux density with the heat flux density value under natural cooling and select the appropriate cooling method.

[0016] Step 2 specifically includes the following steps:

[0017] (1) According to Formula to calculate air mass flow rate;

[0018] Where q m Indicates air mass flow rate in kg / s; C p It represents the specific heat capacity of air at a fixed pressure, in J / (kg·K). p =1005J / (kg·K); Q 总 Indicates the total heat dissipation, in W; Δt indicates the air temperature difference between the charger outlet and inlet, in °C;

[0019] (2) According to q v =60q m γ -1 Formula to calculate volume flow rate of air;

[0020] Where q v Indicates the air volume flow rate, m 3 / min; γ represents the air density;

[0021] (3) Determine the number and parameters of the heat dissipation equipment corresponding to the selected heat dissipation method based on the air volume flow value.

[0022] Step three specifically includes the following steps:

[0023] (1) Establish a DC charger simulation model;

[0024] The DC charger's charging module housing is modeled using a plate to simulate the chassis. The dust filter at the air inlet is created using the Grill module, with the velocity loss coefficient equivalent to the dust filter's resistance to airflow. The heat dissipation device is created using the Fan module, with the fan type, air volume, and air pressure parameters selected based on actual conditions. The heat sink is modeled using a Heatsink plate. The internal heat of the charging module is modeled using a Source heat source. Other low-power devices that significantly impact the airflow are simplified into corresponding cubes based on their physical characteristics.

[0025] (2) Thermal simulation of DC charger based on ANSYS Icepak;

[0026] Set the computational domain fluid to air, the ambient air temperature to 30°C, the flow state to turbulent, the turbulence equation to zero equation, select the corresponding cooling method, check the radiation calculation option, and set the simulation state to steady state;

[0027] By performing thermal simulation analysis on the DC charger, the temperature field distribution diagram of the charger section is obtained, and the temperature distribution diagram of the charger and the airflow trajectory distribution diagram inside the charger are obtained by calculation and simulation.

[0028] Step 5 specifically includes the following steps:

[0029] The maximum temperature Tmax is determined as the stress level for the constant stress accelerated life test of the charger, and a constant truncation accelerated life test is selected. The charging modules of several newly developed DC chargers are placed in a Tmax constant temperature test chamber. According to the control variable method, other external factors are kept consistent. A monitoring module is set at the output end of the charging module to monitor the output current and voltage parameters of the charging module. If the output current of the charging module fails to meet the expected requirements, the charging module is considered to have failed. The failure time of each charging module prototype is recorded, and the test is terminated after all charging modules have failed.

[0030] Step six specifically includes the following steps:

[0031] (1) Use the least squares method to convert the nonlinear expression of the reliability function or distribution function into a linear form to determine the failure distribution model that the test data obeys;

[0032] For the exponential distribution, the reliability function is: R(t) = e -λt ;

[0033] Where λ is the failure rate and t is the time;

[0034] Taking the logarithm on both sides of the equal sign can be transformed into the following form:

[0035] The data are represented on the coordinate axis, and then a slant line that best fits the data trend is fitted according to the data trend. If the data trend and the straight line are basically consistent, it means that the sample follows an exponential distribution, and the slope of the slant line is the estimated value of λ. The vertical and horizontal coordinate data are:

[0036] For the Weibull distribution, the reliability function is:

[0037] Where m is the shape parameter and η is the scale parameter;

[0038] Similarly, taking the logarithm on both sides of the equal sign can be transformed into the following form:

[0039] The data are expressed on the coordinate axis. The more the fitted slope line matches the trend of the data, the more it indicates that the data sample has the characteristics of Weibull distribution. The slope of the straight line is the estimated value of the shape parameter m, and the horizontal coordinate of the intersection of the straight line and the x-axis is the natural logarithm of the estimated value of the scale parameter η. The vertical and horizontal coordinate data are:

[0040] When the Weibull distribution is selected as the data distribution model, the reliability error calculated based on the approximate median rank formula is the smallest; when the exponential distribution is selected as the data distribution model, the reliability error calculated using the mathematical expectation is the smallest;

[0041] Approximate median rank formula:

[0042] Mathematical expectation formula:

[0043] Where, t i is the failure data, N is the number of samples, R(t i ) is the reliability of the i-th sample;

[0044] (2) Expansion of charging module failure data samples based on BP neural network;

[0045] Calculate the reliability value R(t i ), and then the calculated reliability values ​​are combined into a reliability vector [R(t1), R(t2),…, R(t r )], as the input data of BP neural network, and then the test data is combined into failure time vector [t1, t2,…, t r] and input it into the BP neural network, which is the data output by the BP neural network; then the two sets of data are trained in the BP neural network. After multiple trainings, the weights and thresholds are optimized, and then the BP neural network is used to simulate and generate new reliability data that meets the expected requirements;

[0046] If you want to expand the sample size N to M (M>N), first set the range of random reliability, then use MATLAB to randomly generate MN random reliabilities within this range, and then arrange these MN random reliabilities in ascending order as input data, call the trained BP neural network, and use network simulation to predict MN new data;

[0047] (3) Bootstrap-based failure data expansion;

[0048] Assume that θ = θ(F) is a parameter in the population distribution F, F m+n Defined as the empirical distribution function of the sample, is the estimated value of θ, then the error between the estimated value and the original actual value is:

[0049]

[0050] Calculating the R(T,F) distribution characteristics includes the following steps:

[0051] ①[t1,t2,…,t m ] represents the original real data set, [N1,N2,…,N n ] represents the new data set predicted by the BP neural network, and the expanded sample is T = [t1, t2,…, t m ,N1,N2,…,N n ], construct the empirical distribution function F m+n , that is: after arranging the data in sample T in ascending order, a new data sample set is obtained, and then the empirical cumulative distribution function of the sample is constructed based on the new data sample set as follows:

[0052]

[0053] ②From F m+n Perform independent random resampling with replacement, and let the sample set of the sample be:

[0054] ③According to Calculating Bootstrap Statistics in R * (T * ,F m+n );

[0055] ④ Repeat ② and ③ to obtain the Bootstrap statistic R * (T * ,F m+n ) a series of possible values;

[0056] ⑤Use the Bootstrap statistic R of the sample * (T * ,F m+n ) to infinitely approach the distribution of the original data R(T,F), that is, using R m+n The distribution is approximately equal to T m+n The distribution of θ(F) can be obtained, and then a series of possible values ​​of the parameter θ(F) can be obtained. Finally, the distribution of the parameter θ and its characteristic value can be calculated based on statistical knowledge and methods.

[0057] The empirical cumulative distribution function F m+n Random samples of (x) are generated by the improved Bootstrap method:

[0058] Generate a random number η uniformly distributed between [0,1] using a computer; let β = (n-1)η, i = |β| + 1, where |β| is β rounded down; take any random number a uniformly distributed between [0,1 / (1-α)], and let: Take any random number b uniformly distributed in the interval [0,1 / (1-α)] and let: Repeat the above steps as needed;

[0059] where x i with x i+1 For the original data x1,x2,…,x n After sorting from small to large, the i-th and i+1-th data, and This is the sample data of freshmen;

[0060] (4) Calculation of reliability evaluation indicators;

[0061] The two-parameter Weibull distribution is used to describe the product life, and the malfunction function or distribution function is:

[0062]

[0063] The mean time between failures (MTBF) is: MTBF = η·Γ(1+1 / m);

[0064] Where m is the shape parameter of the Weibull distribution and η is the scale parameter;

[0065] The small sample is expanded by Bootstrap to obtain large sample data that meets the reliability assessment requirements. The expanded large sample data is sorted and the expanded data is fitted with the least squares linear fit. The life of the test equipment is converted into lnt as the horizontal axis and the life of the test equipment is converted into lnt as the horizontal axis. The data points are fitted with a least square linear fit for the ordinate;

[0066] The Arrhenius model reflects the relationship between life and temperature under temperature stress, and only considers the influence of a single temperature stress. Its equation is: L = A0exp(E a / kT);

[0067] Where L represents the characteristic life of the product; k represents the Boltzmann constant, usually k = 8.617 × 10 -5 eV / °C; T is the absolute temperature; A0 is a constant greater than 0; Ea is the activation energy of the failure mechanism. Based on engineering experience, Ea is usually set to 0.7eV and is a constant for the same failure mode of the same type of product.

[0068] According to the Arrhenius model, the calculation formula of the acceleration factor under temperature stress is:

[0069] Where, T 0K Indicates the temperature at room temperature, the unit is Kelvin, symbol K; T tK Indicates the temperature under test conditions, the unit is Kelvin, symbol K;

[0070] (5) Verification of reliability assessment results;

[0071] First, get the confidence interval. The steps are as follows:

[0072] ① Based on the failure data of the test samples obtained from the accelerated life test, the bootstrap and least squares method based on the BP neural network is used to obtain the distribution parameters m and η in the Weibull distribution after linear fitting;

[0073] ② Generate new data that obeys the Weibull distribution with shape parameter m and scale parameter η;

[0074] ③ Randomly extract a very small sample from the newly generated data;

[0075] ④ For the small sample, the Bootstrap method based on semi-empirical virtual augmentation and the least squares method are used to calculate m * and η * ;

[0076] ⑤ Repeat steps ③ and ④ N times to get m n and η n , where n = 1, 2, ..., N;

[0077] ⑥Then the p quantile can be calculated,

[0078] ⑦ General Arrange from small to large, and you can get the confidence interval of 1-α:

[0079] Determine whether the calculated MTBF value is within the above confidence interval. If so, it means that the calculation result is valid.

[0080] The technical effects and advantages of the present invention include: analyzing the DC charger's cooling system, determining the ventilation volume required when the DC charger is fully loaded, and determining the air volume of the cooling fan. The DC charger is then simulated and analyzed using the ANSYS Icepak thermal simulation platform, ultimately generating a temperature distribution cloud map of the DC charger under full load. Based on this temperature distribution cloud map, the DC charger's weaknesses are identified. Based on the thermal simulation results, the maximum internal temperature of the DC charger under full load is determined. This temperature serves as the test stress value for the newly developed DC charger, and an accelerated life test is designed to obtain test data. Taking into account the extremely small sample size of the test data, a BP neural network and the Bootstrap method are used to scale the small sample data. The least squares method is then used to estimate the distribution function parameters and calculate the mean time between failures reliability index. Finally, the Bootstrap method is again used for sampling to calculate confidence intervals, verifying the effectiveness of the scaling method and calculation results. Overall, this method establishes a reliability analysis method for newly developed DC chargers that can identify equipment weaknesses under temperature stress and predict equipment life when sufficient reliability data is lacking. BRIEF DESCRIPTION OF THE DRAWINGS

[0081] Figure 1 This is a flow chart of the DC charger reliability analysis method of the present invention;

[0082] Figure 2 This is a temperature field distribution diagram of the charger cross section of this embodiment;

[0083] Figure 3 This is the temperature distribution diagram of the charging module of this embodiment;

[0084] Figure 4 : is the trajectory distribution of the airflow inside the charger of this embodiment;

[0085] Figure 5 This is a graph showing the linear fitting results based on the Weibull distribution least squares method of this embodiment;

[0086] Figure 6 This is a graph showing the linear fitting result based on the least squares method of exponential distribution in this embodiment;

[0087] Figure 7 This is a diagram showing the training results of the BP neural network data samples of this embodiment;

[0088] Figure 8 This is a comparison chart of the BP neural network prediction data and the actual test data in this embodiment;

[0089] Figure 9 This is the Bootstrap expansion result diagram of this embodiment;

[0090] Figure 10 This is the fitting result based on the least squares method after Bootstrap expansion in this embodiment. DETAILED DESCRIPTION

[0091] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0092] The present invention provides Figure 1-10 A reliability analysis method for a DC charger affected by temperature is shown, which specifically includes the following steps:

[0093] Step 1: Classification and selection of thermal cooling methods

[0094] According to different heat transfer methods and heat dissipation requirements, there are seven main heat dissipation methods for electronic components, as shown in Table 1.

[0095] Table 1 Heat flux density of different cooling methods

[0096]

[0097] The heat flux density of the heat dissipation surface in the steady state is calculated by the following formula:

[0098]

[0099] Where A is the total heat dissipation area of ​​the charger, in cm 2 .

[0100] The volume power density of the heat dissipation surface in steady state is calculated as follows:

[0101]

[0102] Taking an 80kW DC charger as the research object, the internal temperature distribution of the DC charger was studied when the charger was fully loaded and the ambient temperature was 30°C. The highest temperature area inside the charger was identified. All housings and internal baffles were made of galvanized steel with excellent thermal insulation properties. The overall dimensions of the DC charger are 1800×700×400mm. The entire charger consists of a charging module, DC bus, AC line switch, charging gun, auxiliary power supply, and casing.

[0103] When the charging pile is actually working at full load, the power of a charging module is 20KW and the working efficiency is 95%. The remaining 5% of the power is dissipated as heat, so the heat dissipation Q1 of the charging module is 1000W.

[0104] The heat inside the charging module mainly comes from the rectifier module and transformer. The 80kW DC charger has a total of 4 charging modules with a total heat dissipation of 4000W. Adding the heat dissipation of some other low-power components inside the charging module, the heat dissipation of the entire charger is about 4500W.

[0105] The charger dimensions can be calculated to obtain a surface area of ​​A = 45200 cm 2 , calculated by formula (1) The volume of the charger is V = 504000 cm 3 , calculated by formula (2)

[0106] Heat flux The heat flux density is greater than the value under natural cooling, indicating that natural cooling alone cannot meet the heat dissipation needs of the DC charger, and additional effective cooling methods are needed. Therefore, a forced air cooling solution with a cooling fan is adopted to cool the entire DC charger.

[0107] Step 2: Determine ventilation volume and cooling fan

[0108] When forced air cooling is used as the primary heat dissipation method, most of the heat in the system is dissipated from the charger to the air by convection. The air mass flow rate is calculated by the following formula:

[0109]

[0110] Where q m Indicates air mass flow rate in kg / s; C p It represents the specific heat capacity of air at a fixed pressure, in J / (kg·K). p =1005J / (kg·K); Q 总 Indicates the total heat dissipation, in W; Δt indicates the air temperature difference between the charger outlet and inlet, in °C;

[0111] The relationship between the volume flow rate of air and the mass and flow rate of air is expressed as follows:

[0112] q v =60q m γ -1 (4)

[0113] Where q v Indicates the air volume flow rate, m 3 / min; γ represents the air density, which is generally 1.23kg / m at normal temperature and pressure. 3 ;

[0114] Assuming that the ambient temperature of the air at the air inlet is 30°C, the internal temperature of the charger is higher than the air outlet temperature due to the operation of components, etc., which is set to about 50°C, then Δt = 20°C. In order to meet the heat dissipation requirements, the total ventilation volume is calculated by formula (3) and formula (4) to be approximately 16.5m 3 / min, considering that the ventilation system will be affected by the wind resistance of other components inside the machine, the total air volume is 28.36m 3 / min, four fans are used according to the overall layout of the charging module in the charger, and all fans use the largest air volume of 7.09m 3 / min fan.

[0115] Step 3: Establishment of DC charger simulation model

[0116] The heat dissipation of the entire charger mainly comes from the four charging modules. Therefore, the components other than the charging modules can be regarded as low-heat components and can be ignored during modeling. The charging module is mainly composed of a plug-in heat sink, a transformer, a three-phase rectifier coil and a capacitor. The circuit board is Figure 4 There are heat dissipation fans on the left and vents on the right. The shell material of the charging module is galvanized aluminum plate, which is modeled using plate to simulate the chassis shell. The dust filter at the air inlet is established using the Grill module. The velocity loss coefficient is equivalent to the resistance of the dust filter to the airflow. The heat dissipation fan is established based on the Fan module. The type of fan, air volume and pressure and other parameters are selected according to actual conditions. The radiator is modeled using the Heatsink plate. In order to reduce the amount of calculation, without affecting the calculation results, the present invention adopts a simplified radiator. The internal heat of the charging module is mainly generated by the IGBT module and the transformer module, which is equivalent to a heat source and is modeled using the Source heat source. For other devices that have a greater impact on the air path but are low-power, such as larger capacitors and coils, they are simplified into cubes, cylinders, etc. according to their appearance characteristics.

[0117] Step 4: DC charger thermal simulation based on ANSYS Icepak

[0118] The computational domain fluid is set to air. Considering that the charger works outdoors, the ambient air temperature is set to 30°C. According to the previous analysis, the flow state of the fluid is selected as turbulent, the turbulence equation is selected as zero equation, the cooling method is selected as forced air cooling, the radiation calculation option is checked, and the simulation state of the charger is set to steady state.

[0119] The thermal conductivity of the materials is shown in Table 2:

[0120] Table 2 Thermal conductivity of materials

[0121]

[0122] The mesh is finely divided in the areas with complex structures such as cooling fans and dust filters. In other places, as long as the mesh can show the shape of the model, the mesh quality meets the requirements. Figure 2 The charger temperature distribution cloud diagram is shown.

[0123] The calculation and simulation results show the temperature distribution of the charger and the internal airflow trajectory distribution (reference Figure 3 and Figure 4 ),Depend on Figure 2 It can be seen that when the 80KW DC charger is working at full load and the external ambient temperature is 30°C, the maximum temperature of the heat source inside the charging module reaches 67.61°C. When the entire charger is working at full load, the highest temperature is concentrated in the heat source inside the charging module. The internal temperature of the charger gradually increases from the lower left air inlet along the direction of air flow.

[0124] The root cause is that the air flows through the charging module, and the air duct inside the body is long, and the air is heated by the hot charging module, so the air temperature at the air outlet is higher than that at the air inlet.

[0125] Depend on Figure 4 It can be seen that the air flows from the bottom air inlet through the charging module to the upper right cooling fan. The air flow trajectory gradually becomes more numerous and crowded, indicating that there are not enough cooling fans at the air outlet to completely extract the hot air in the machine, which is not conducive to air outflow, that is, not conducive to heat dissipation.

[0126] Step 5: Design an accelerated life test plan

[0127] From the simulation results, it can be seen that the charging module is the weak link of the DC charger. Therefore, targeted fault simulation tests on the charging module have become a necessary link to obtain reliability analysis data. During the operation of the charging module, the input and output voltages of charging and discharging are fixed, and in the process of charging electric vehicles, the temperature of the external environment and the temperature around the charging module are relatively stable. Therefore, the temperature stress that affects the reliability of the charging module can be regarded as constant. The results of the thermal simulation analysis of the charger show that the maximum internal temperature of the charger reaches 67.60℃ when it is working. The constant of the charger is determined. The stress level of the stress accelerated life test was selected as 67.60℃. As a newly developed equipment, the DC charger has the disadvantage of insufficient test samples. Only four prototypes can be provided as test equipment. Therefore, a constant truncation accelerated life test was selected. The four charging modules were placed in a 67.60℃ constant temperature test chamber. According to the control variable method, other external factors were kept consistent. A monitoring module was set at the output end of the charging module to monitor the output current and voltage of the charging module. If the output current of the charging module cannot meet the expected requirements, the charging module is considered to have failed. The failure time of the four charging module prototypes was recorded. The test was stopped after all four charging modules failed.

[0128] According to the above test plan design, the test data of the four charging module test prototypes are shown in the following table:

[0129] Table 3 Charging module test data

[0130]

[0131] Step 6: Charging module failure data sample expansion based on BP neural network

[0132] There are two main failure distribution models currently used for charging modules, one is the exponential distribution model and the other is the Weibull distribution model. Both distribution models are continuous distribution models. For the determination of the distribution model, the least squares method can be used to convert the nonlinear expression of the reliability function or distribution function into a linear form. Based on this characteristic, it can be determined what distribution the data obeys.

[0133] For the exponential distribution, the reliability function is:

[0134] R(t)=e -λt (5)

[0135] Where λ is the failure rate and t is the time.

[0136] Taking the logarithm on both sides of the equal sign can be transformed into the following form:

[0137]

[0138] If all the data are represented on the coordinate axis, and then a slant line that best fits the data trend is fitted according to the data trend, if the data trend and the straight line are basically consistent, it means that the sample follows an exponential distribution, and the slope of the slant line is the estimated value of λ:

[0139]

[0140] For the Weibull distribution, the reliability function is:

[0141]

[0142] Where m is the shape parameter and η is the scale parameter;

[0143] Similarly, taking the logarithm on both sides of the equal sign can be transformed into the following form:

[0144]

[0145] The data are then represented on the coordinate axis. The more the fitted slope line conforms to the direction of the data, the more it indicates that the data sample has the characteristics of Weibull distribution. The slope of the straight line is the estimated value of the shape parameter m, and the horizontal coordinate of the intersection of the straight line and the x-axis is the natural logarithm of the estimated value of the scale parameter η.

[0146]

[0147] When the Weibull distribution is selected as the data distribution model, the reliability error calculated according to the approximate median rank formula is the smallest. When the exponential distribution is selected as the data distribution model, the reliability error calculated using the mathematical expectation is the smallest.

[0148] Approximate median rank formula:

[0149]

[0150] Mathematical expectation formula:

[0151]

[0152] Where, t i is the failure data, N is the number of samples, R(t i ) is the reliability of the i-th sample.

[0153] The failure time t4 = [408, 367.2, 399.84, 391.68] in Table 3 is fitted with the least square curve based on the Weibull distribution model, with lnt as the horizontal axis and As the vertical axis, a linear fit is performed on the four failure data based on the exponential distribution with t as the horizontal axis. As the vertical axis, the four failure data are also linearly fitted, and the fitting results are as follows Figure 5 and Figure 6 shown.

[0154] Table 4 Least squares fitting correlation results

[0155]

[0156] From the least squares linear fitting graph, the data points are distributed near a straight line, and it is impossible to directly determine what distribution the failure data conforms to through the graph. The correlation coefficient ρ of the least squares curve fitting reflects the degree of conformity between the original data to be analyzed and the fitted distribution model. From the values ​​of the least squares fitting correlation coefficients in Table 4, it can be seen that the correlation coefficients of the two are very close. When the correlation coefficient ρ is closer to 1, the consistency of the data and the fitted distribution model is higher. Comparing the two correlation coefficients in Table 4, 0.9990 is shorter than 0.9940, so from the priority point of view, it can be considered that this set of data conforms to the Weibull distribution. Therefore, when the data sample conforms to the Weibull distribution, the calculation formula of the empirical reliability can choose the approximate median rank formula.

[0157] For the four test data in Table 3, calculate the reliability value R(t i ), and then the calculated reliability values ​​are combined into a reliability vector [R(t1), R(t2),…, R(t r )], so that the input data of the BP neural network can be determined as the reliability vector, and then the four test data are combined into a failure time vector [t1, t2,…, t r ], input it into the BP neural network, which is the data output by the BP neural network. The two sets of data are learned and trained in the BP neural network. After multiple learning and training, the weights and thresholds are optimized, and the trained BP neural network can be saved as a backup. Then this network can be called and then used to simulate and generate new reliability data that meets the expected requirements.

[0158] First, the reliability of the sample is calculated according to formula (11):

[0159] R(t i )=[0.84,0.61,0.39,0.16]

[0160] R(t i)=[0.84,0.61,0.39,0.16] is used as the input data of the BP neural network training data, and t4=[408,367.2,399.84,391.68] is used as the output of the learning data. After reading into the BP neural network, training is performed. At the same time, these four sets of input and output data are used as test data to confirm the accuracy of the training results. When the test data and the original data reach a certain accuracy, it can be considered that the predicted data has inherited the characteristics of the original data very well. At this time, it can be considered that the predicted value has a high degree of credibility, such as Figure 7 and as shown in Table 5.

[0161] Table 5 BP neural network training accuracy table

[0162]

[0163] from Figure 7 It can be seen that the predicted value inherits the characteristics of the original data very well. As can be seen from Table 5, the learning ability of the BP neural network is very strong. After training the original input and output, the simulation accuracy is extremely high, and its predicted value is almost exactly the same as the measured value. This shows that the training of this BP model is very successful. Therefore, this BP model can be used to expand the failure life of the charging module under small samples, and then the reliability analysis of the expanded samples will have a very high credibility.

[0164] To expand this quantity from 4 to 10, 6 more data are needed. First, according to the reliability calculated from the 4 test data in Table 3, the range of subsequent random reliability is determined. Therefore, the range of random reliability is set to 0.19-0.84. MATLAB is used to randomly generate 6 random reliabilities within this range. Then, these 6 random reliabilities are arranged in order from small to large as input data, and the trained BP neural network is called. Network simulation is used to predict 6 new data. These 6 new data can be regarded as new test data with the characteristics of the original test data, and together with the original 4 test data, 10 sample data are formed.

[0165] After BP neural network simulation, the prediction results are compared with the original values. Figure 8 As shown in the figure, the failure data predicted by BP neural network has basically the same trend as the actual test data, and the predicted data is still within the range of the actual test data, which once again proves that it is feasible and effective to use BP neural network to expand very small samples.

[0166] The sample data after the charging module expansion is obtained as follows:

[0167] T=[367.20,374.63 380.05,387.60,391.83,394.16,397.91,399.84,408.00,412.22]

[0168] Step 7: Bootstrap-based failure data expansion

[0169] Assume that θ = θ(F) is a parameter in the population distribution F, F m+n Defined as the empirical distribution function of the sample, is the estimated value of θ, then the error between the estimated value and the original actual value is shown in formula (13):

[0170]

[0171] This article calculates the R(T,F) distribution characteristics and requires five basic steps:

[0172] 1. [t1,t2,…,t m ] represents the original real data set, [N1,N2,…,N n ] represents the new data set predicted by the BP neural network, and the expanded sample is T = [t1, t2,…, t m ,N1,N2,…,N n ], construct the empirical distribution function F m+n ,Right now:

[0173] After arranging the data in sample T in ascending order, a new data sample set is obtained. Then, the empirical cumulative distribution function of the sample is constructed based on the new data sample set as shown in formula (14):

[0174]

[0175] 2. From F m+n Perform independent random resampling with replacement, and let the sample set of the sample be:

[0176] 3. Calculate the Bootstrap statistic R according to formula (15) * (T * ,F m+n );

[0177]

[0178] In the formula It's T * The empirical distribution function, R m+n F m+n Bootstrap statistics;

[0179] 4. Repeat steps 2 and 3 above to obtain the Bootstrap statistic R * (T * ,F m+n ) a series of possible values;

[0180] 5. Use the Bootstrap statistic R of the sample * (T * ,F m+n ) to infinitely approach the distribution of the original data R(T,F), that is, using R m+n The distribution is approximately equal to T m+n distribution, and then we can get a series of possible values ​​of the parameter θ(F), and finally we can calculate the distribution of the parameter θ and its characteristic value based on statistical knowledge and methods.

[0181] In the above steps, the empirical cumulative distribution function F m+n The random samples of (x) need to be generated by simulation. The traditional Bootstrap method can only expand in one direction and cannot jump out of the original data interval. Therefore, the improved Bootstrap method is used to generate the sample that obeys the above empirical cumulative distribution function F m+n (x) The process of random sampling is as follows:

[0182] (1) Use a computer to generate a random number η that is uniformly distributed between [0,1].

[0183] (2) Let β = (n-1)η, i = |β| + 1, where |β| is β rounded down.

[0184] (3) Take any random number a uniformly distributed in the interval [0, 1 / (1-α)] and let:

[0185]

[0186] (4) Take any random number b uniformly distributed in the interval [0, 1 / (1-α)] and let:

[0187]

[0188] (5) Repeat the above steps as needed.

[0189] where x i with x i+1 For the original data x1,x2,…,x n After sorting from small to large, the i-th and i+1-th data, and This is the sample data for new students.

[0190] Arrange the expanded 10 samples from small to large. The 10 samples after arrangement are as follows: Y = [367.20, 374.63, 380.05, 387.60, 391.83, 394.16, 397.91, 399.84, 408.00, 412.22]. Use the above Bootstrap method to expand it into a large sample. The result is as follows Figure 9 shown.

[0191] Step 8: Calculation of reliability evaluation indicators

[0192] The two-parameter Weibull distribution is used to describe the product life, and the malfunction function or distribution function is:

[0193]

[0194] The mean time between failures (MTBF) is:

[0195] MTBF=η·Γ(1+1 / m) (17)

[0196] Where m is the shape parameter of the Weibull distribution and η is the scale parameter.

[0197] The small sample is expanded by Bootstrap to obtain large sample data that meets the reliability assessment requirements. The expanded large sample data is sorted and the expanded data is fitted with the least squares linear fit. The life of the test equipment is converted into lnt as the horizontal axis and the life of the test equipment is converted into lnt as the horizontal axis. The data points are fitted with least square linear fitting for the vertical coordinate, and the results are as follows Figure 10 As shown in the figure, it can be seen that most of the data after Bootstrap expansion are evenly distributed near the straight line.

[0198] The linear fitting straight line is: y=66.08x-393.08.

[0199] Combining formulas (9) and (10), we can get the estimated values ​​of the unknown parameters m and η in the Weibull distribution parameters, which are: m = 66.08, η = 383.20. The mean time between failures (MTBF) of the charging module at 67.60°C is:

[0200] MTBF=379.94h

[0201] The Arrhenius model reflects the relationship between life and temperature under temperature stress, considering only the influence of a single temperature stress. Its equation is:

[0202] L=A0 exp(E a / kT) (18)

[0203] Where L represents the characteristic life of the product; k represents the Boltzmann constant, usually k = 8.617 × 10 -5 eV / ℃; T is the absolute temperature; A0 is a constant greater than 0; Ea is the activation energy of the failure mechanism. Ea is usually taken as 0.7eV according to engineering experience and is a constant for the same failure mode of the same type of product.

[0204] In accelerated life tests, the acceleration factor reflects the accelerating effect of accelerated stress on the product under test. It is a key indicator for characterizing accelerated stress and the product's service life, i.e., lifespan. The acceleration factor refers to the ratio of the product's life under normal stress to its life under accelerated stress. The acceleration factor is generally obtained using an existing acceleration model. The calculation formula for the acceleration factor under temperature stress based on the Arrhenius model is:

[0205]

[0206] Where, T 0K Indicates the temperature at room temperature, the unit is Kelvin, symbol K; T tK Indicates the temperature under test conditions, the unit is Kelvin, symbol K.

[0207] According to the acceleration factor calculation formula, when the room temperature is 25℃, the acceleration factor corresponding to 67.60℃ is calculated by formula (19):

[0208]

[0209] Since the expanded data is conducted under a high-temperature constant accelerated life test, the average life calculated using the expanded data is also the average life at high temperature. According to the definition of the acceleration factor, the acceleration factor is defined as the difference between the average life of the product under two different stress levels. The average life at high temperature is converted to the life under normal stress based on the acceleration factor. The life under normal stress can be calculated as:

[0210] MTBF=11493.18h.

[0211] That is, in the absence of any faults, the charging module can operate normally for 11493.18 hours under normal stress conditions.

[0212] Step 9: Verification of reliability assessment results

[0213] The main steps to obtain a confidence interval are as follows:

[0214] 1. Based on the failure data of the test samples obtained from the accelerated life test, the values ​​of the distribution parameters m and η in the Weibull distribution are obtained after linear fitting using the bootstrap and least squares method based on the BP neural network;

[0215] 2. Generate new data that follows a Weibull distribution with shape parameter m and scale parameter η;

[0216] 3. Randomly extract a very small sample from the newly generated data;

[0217] 4. For the small sample, the bootstrap method based on semi-empirical virtual augmentation and the least squares method are used to calculate m * and η * ;

[0218] 5. Repeat steps 3 and 4 N times to get m n and η n , where n = 1, 2, ..., N;

[0219] 6. Then the p quantile can be calculated,

[0220] 7. Arrange from small to large, and you can get the confidence interval of 1-α:

[0221] The above method is used to calculate the confidence interval of the test data, and the confidence interval with a confidence level of 95% at the percentile position is [10901, 11721], the confidence interval with a confidence level of 95% at the quartile position is [10964, 11707], and the confidence interval with a confidence level of 95% at the binary position is [10873, 11718].

[0222] It can be seen from the confidence interval that the calculated MTBF=11493.18h value is within the range of the above confidence interval, which means that the calculation result is valid, that is, under normal circumstances, the DC charger can work normally for 11493.18h.

[0223] Therefore, the present invention first analyzes the heat dissipation system of the DC charger, determines the ventilation volume when the DC charger is fully loaded, and determines the air volume of the cooling fan. Then, based on the ANSYS Icepak thermal simulation platform, the DC charger is simulated and analyzed, and a temperature distribution cloud map of the DC charger when fully loaded is finally obtained. Based on the temperature distribution cloud map, the weak links of the DC charger are analyzed. Based on the thermal simulation results, the maximum internal temperature of the DC charger when fully loaded is obtained. This temperature is used as the test stress value for the newly developed DC charger, and an accelerated life test is designed to obtain test data. Then, considering the characteristics of the test data presenting extremely small samples, the BP neural network and the Bootstrap method are used to expand the small sample data. The least squares method is combined to estimate the distribution function parameters and calculate the mean time between failures reliability index. Finally, the Bootstrap method is used again for sampling to calculate the confidence interval, verifying the effectiveness of the expansion method and calculation results.

[0224] Finally, it should be noted that the above is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art can still modify the technical solutions described in the aforementioned embodiments or make equivalent substitutions for some of the technical features therein. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A reliability analysis method for a DC charger under temperature influence, characterized by: The specific steps are as follows: Step 1: Analyze the heat dissipation system of the DC charger and select an appropriate heat dissipation method; Step 2: Determine the ventilation volume of the DC charger when it is fully loaded, and determine the layout of the heat dissipation equipment corresponding to the selected heat dissipation method based on the ventilation volume; Step 3: Perform simulation analysis on the DC charger to obtain a temperature distribution cloud diagram when the DC charger is fully loaded. Step 4: Analyze the weak links of the DC charger based on the temperature distribution cloud map. Then, based on the thermal simulation results, determine the maximum internal temperature of the DC charger when operating at full load. Step 5: Use the highest temperature as the test stress value of the newly developed DC charger and design an accelerated life test to obtain test data; Step 6: Use BP neural network and Bootstrap method to expand the small sample data, combine with least squares method to estimate the distribution function parameters, calculate the mean time between failures reliability index, and finally, use Bootstrap method for sampling again to calculate the confidence interval and verify the effectiveness of the expansion method and calculation results.

2. The reliability analysis method of a DC charger under the influence of temperature according to claim 1, characterized in that: Step 1 specifically includes the following steps: according to The formula is used to calculate the heat flux density of the heat dissipation surface of the DC charger in steady state. Where A is the total heat dissipation area of ​​the charger, in cm 2 ; Compare the calculated heat flux density with the heat flux density value under natural cooling and select the appropriate cooling method.

3. The reliability analysis method of a DC charger under the influence of temperature according to claim 1, characterized in that: Step 2 specifically includes the following steps: (1) According to Formula to calculate air mass flow rate; Where q m Indicates air mass flow rate in kg / s; C p It represents the specific heat capacity of air at a fixed pressure, in J / (kg·K). p =1005J / (kg·K);Q 总 Indicates the total heat dissipation, in W; Δt indicates the air temperature difference between the charger outlet and inlet, in °C; (2) According to q v =60q m γ -1 Formula to calculate volume flow rate of air; Where q v Indicates the air volume flow rate, m 3 / min; γ represents the air density; (3) Determine the number and parameters of the heat dissipation equipment corresponding to the selected heat dissipation method based on the air volume flow value.

4. The reliability analysis method of a DC charger under the influence of temperature according to claim 1, characterized in that: Step three specifically includes the following steps: (1) Establish a DC charger simulation model; The DC charger's charging module housing is modeled using a plate to simulate the chassis. The dust filter at the air inlet is created using the Grill module, with the velocity loss coefficient equivalent to the dust filter's resistance to airflow. The heat dissipation device is created using the Fan module, with the fan type, air volume, and air pressure parameters selected based on actual conditions. The heat sink is modeled using a Heatsink plate. The internal heat of the charging module is modeled using a Source heat source. Other low-power devices that significantly impact the airflow are simplified into corresponding cubes based on their physical characteristics. (2) Thermal simulation of DC charger based on ANSYS Icepak; Set the computational domain fluid to air, the ambient air temperature to 30°C, the flow state to turbulent, the turbulence equation to zero equation, select the corresponding cooling method, check the radiation calculation option, and set the simulation state to steady state; By performing thermal simulation analysis on the DC charger, the temperature field distribution diagram of the charger section is obtained, and the temperature distribution diagram of the charger and the airflow trajectory distribution diagram inside the charger are obtained by calculation and simulation.

5. The reliability analysis method of a DC charger under the influence of temperature according to claim 1, characterized in that: Step 5 The specific steps include: The maximum temperature T max Determine the stress level of the constant stress accelerated life test for the charger and select the constant truncated accelerated life test; Place the charging modules of several newly developed DC chargers in T max In the constant temperature test chamber, according to the control variable method, other external factors are kept consistent. A monitoring module is set at the output end of the charging module to monitor the output current and voltage parameters of the charging module. If the output current of the charging module cannot meet the expected requirements, the charging module is considered to have failed. The failure time of each charging module prototype is recorded. The test is stopped when all charging modules fail.

6. The reliability analysis method of a DC charger under the influence of temperature according to claim 1, characterized in that: Step six specifically includes the following steps: (1) Use the least squares method to transform the nonlinear expression of the reliability function or distribution function into a linear form to determine the failure distribution model that the test data obeys; For the exponential distribution, the reliability function is: R(t) = e -λt ; Where λ is the failure rate and t is the time; Taking the logarithm on both sides of the equal sign can be transformed into the following form: The data are represented on the coordinate axis, and then a slant line that best fits the data trend is fitted according to the data trend. If the data trend and the straight line are basically consistent, it means that the sample follows an exponential distribution, and the slope of the slant line is the estimated value of λ. The vertical and horizontal coordinate data are: For the Weibull distribution, the reliability function is: Where m is the shape parameter and η is the scale parameter; Similarly, taking the logarithm on both sides of the equal sign can be transformed into the following form: The data are expressed on the coordinate axis. The more the fitted slope line matches the trend of the data, the more it indicates that the data sample has the characteristics of Weibull distribution. The slope of the straight line is the estimated value of the shape parameter m, and the horizontal coordinate of the intersection of the straight line and the x-axis is the natural logarithm of the estimated value of the scale parameter η. The vertical and horizontal coordinate data are: When the Weibull distribution is selected as the data distribution model, the reliability error calculated based on the approximate median rank formula is the smallest; when the exponential distribution is selected as the data distribution model, the reliability error calculated using the mathematical expectation is the smallest; Approximate median rank formula: Mathematical expectation formula: Where, t i is the failure data, N is the number of samples, R(t i ) is the reliability of the i-th sample; (2) Expansion of charging module failure data samples based on BP neural network; Calculate the reliability value R(t i ), and then the calculated reliability values ​​are combined into a reliability vector [R(t1), R(t2),…, R(t r )], as the input data of BP neural network, and then the test data is combined into failure time vector [t1, t2,…, t r ] and input it into the BP neural network, which is the data output by the BP neural network; then the two sets of data are trained in the BP neural network. After multiple trainings, the weights and thresholds are optimized, and then the BP neural network is used to simulate and generate new reliability data that meets the expected requirements; If you want to expand the sample size N to M (M>N), first set the range of random reliability, then use MATLAB to randomly generate MN random reliabilities within this range, and then arrange these MN random reliabilities in ascending order as input data, call the trained BP neural network, and use network simulation to predict MN new data; (3) Bootstrap-based failure data expansion; Assume that θ = θ(F) is a parameter in the population distribution F, F m+n Defined as the empirical distribution function of the sample, is the estimated value of θ, then the error between the estimated value and the original actual value is: Calculating the R(T,F) distribution characteristics includes the following steps: ①[t1,t2,…,t m ] represents the original real data set, [N1,N2,…,N n ] represents the new data set predicted by the BP neural network, and the expanded sample is T = [t1, t2,…, t m ,N1,N2,…,N n ], construct the empirical distribution function F m+n , that is: after arranging the data in sample T in ascending order, a new data sample set is obtained, and then the empirical cumulative distribution function of the sample is constructed based on the new data sample set as follows: ②From F m+n Perform independent random resampling with replacement, and let the sample set of the sample be: ③According to Calculating Bootstrap Statistics in R * (T * ,F m+n ); ④ Repeat ② and ③ to obtain the Bootstrap statistic R * (T * ,F m+n ) a series of possible values; ⑤Use the Bootstrap statistic R of the sample * (T * ,F m+n ) to infinitely approach the distribution of the original data R(T,F), that is, using R m+n The distribution is approximately equal to T m+n The distribution of θ(F) can be obtained, and then a series of possible values ​​of the parameter θ(F) can be obtained. Finally, the distribution of the parameter θ and its characteristic value can be calculated based on statistical knowledge and methods. The empirical cumulative distribution function F m+n Random samples of (x) are generated by the improved Bootstrap method: Generate a random number η uniformly distributed between [0,1] using a computer; let β = (n-1)η, i = |β| + 1, where |β| is β rounded down; take any random number a uniformly distributed between [0,1 / (1-α)], and let: Take any random number b uniformly distributed in the interval [0,1 / (1-α)] and let: Repeat the above steps as needed; where x i with x i+1 For the original data x1,x2,…,x n After sorting from small to large, the i-th and i+1-th data, and This is the sample data of freshmen; (4) Calculation of reliability evaluation indicators; The two-parameter Weibull distribution is used to describe the product life, and the malfunction function or distribution function is: The mean time between failures (MTBF) is: MTBF = η·Γ(1+1 / m); Where m is the shape parameter of the Weibull distribution and η is the scale parameter; The small sample is expanded by Bootstrap to obtain large sample data that meets the reliability assessment requirements. The expanded large sample data is sorted and the expanded data is fitted with the least squares linear fit. The life of the test equipment is converted into lnt as the horizontal axis and the life of the test equipment is converted into lnt as the horizontal axis. The data points are fitted with a least square linear fit for the ordinate; The Arrhenius model reflects the relationship between life and temperature under temperature stress, and only considers the influence of a single temperature stress. Its equation is: L = A0exp(E a / kT); Where L represents the characteristic life of the product; k represents the Boltzmann constant, usually k = 8.617 × 10 -5 eV / ℃; T is absolute temperature; A0 is a constant greater than 0; E a is the activation energy of the failure mechanism, E a According to engineering experience, 0.7eV is used, and it is a constant for the same failure mode of the same type of product; According to the Arrhenius model, the calculation formula of the acceleration factor under temperature stress is: Where, T 0K Indicates the temperature at room temperature, the unit is Kelvin, symbol K; T tK Indicates the temperature under test conditions, the unit is Kelvin, symbol K; (5) Verification of reliability assessment results; First, get the confidence interval. The steps are as follows: ① Based on the failure data of the test samples obtained from the accelerated life test, the bootstrap and least squares method based on the BP neural network is used to obtain the distribution parameters m and η in the Weibull distribution after linear fitting; ② Generate new data that obeys the Weibull distribution with shape parameter m and scale parameter η; ③ Randomly extract a very small sample from the newly generated data; ④ For the small sample, the Bootstrap method based on semi-empirical virtual augmentation and the least squares method are used to calculate m * and η * ; ⑤ Repeat steps ③ and ④ N times to get m n and η n , where n = 1, 2, ..., N; ⑥Then the p quantile can be calculated, ⑦ General Arrange from small to large, and you can get the confidence interval of 1-α: Determine whether the calculated MTBF value is within the above confidence interval. If so, it means that the calculation result is valid.

Images