Low-power SuperBuck converter failure discrimination method based on component degradation model

Abstract

The invention relates to a low-power Superbuck converter failure discrimination method based on a component degradation model, and the method comprises the following steps: S1, designing a Superbuck converter high-temperature degradation experiment, and obtaining degradation experiment data; s2, acquiring key component degradation data, preprocessing the key component degradation data, and iteratively training the CNN network and the LSTM network to obtain a key component degradation curve prediction model; and S3, obtaining actual degradation data, inputting the degradation curve prediction model of the key component to obtain an actual degradation curve, injecting a current degradation parameter prediction value of the key component in the actual degradation curve into the Superbuck converter, and carrying out failure judgment to obtain a converter failure moment. Compared with the prior art, the method has the advantages of solving the problem of difficult positioning at the failure moment of the converter and the like.

Description

Technical Field

[0001] The present invention relates to the technical field of encoder testing, and in particular to a failure judgment method for a low-power SuperBuck converter based on a component degradation model. Background Art

[0002] SuperBuck converters offer low power consumption and high efficiency. Due to these characteristics, they are widely used in long-life, high-reliability products. For example, power supply systems for submarine observation networks (operating for 25 years or more) and satellite power supply systems (operating for 10 years or more) require converters to provide reliable, high-quality power. Therefore, converter health is a prerequisite for the normal operation of these devices.

[0003] After long-term operation, SuperBuck converter component parameters may change or even degrade, affecting the reliability and stability of the power supply system. Therefore, monitoring and evaluating the converter's operating status and locating its failure point are crucial. Locating the converter's failure point to ensure stable and reliable power supply system operation presents the following challenges:

[0004] (1) The degradation of the converter's health status is a latent characteristic and difficult to observe. Before the system fails, the system may already be in a sub-healthy state. Although no fault has occurred, the stability has been reduced. At the same time, there are relatively few characteristic signals reflecting the degradation law, and the law of parameter changes is not clear.

[0005] (2) The degradation of components is uncertain, and it is difficult to accurately capture the degradation trend of components.

[0006] (3) The converter failure moment is variable and nonlinear, and the indicators that cause converter failure are uncertain, so it is difficult to accurately locate it. Summary of the Invention

[0007] The purpose of the present invention is to solve the problem of difficulty in locating the converter failure moment and to provide a low-power SuperBuck converter failure judgment method based on a component degradation model.

[0008] The purpose of the present invention can be achieved by the following technical solutions:

[0009] A method for determining failure of a low-power SuperBuck converter based on a component degradation model comprises the following steps:

[0010] S1. Taking the Superbuck converter as the object, design the Superbuck converter high temperature degradation experiment and obtain the degradation experimental data;

[0011] S2. Select and obtain degradation data of key components from degradation experimental data. The degradation data of key components is preprocessed and then used as input of the CNN network. The output of the CNN network is used as input of the LSTM network. The LSTM network outputs the degradation curve of the component. The CNN network and LSTM network are iteratively trained. The obtained CNN-LSTM network is used as the degradation curve prediction model of the key components.

[0012] S3. Obtain actual degradation data and input the key component degradation curve prediction model to obtain the actual degradation curve. Inject the current degradation parameter prediction values ​​of the key components in the actual degradation curve into the Superbuck converter to obtain the output voltage ripple of the Superbuck converter. If the output voltage ripple does not exceed the failure threshold, continue to inject the degradation parameter prediction value at the next moment until the degradation parameter prediction values ​​at all moments are injected. Otherwise, record the moment corresponding to the current degradation parameter prediction value as the converter failure moment.

[0013] Furthermore, a high-temperature degradation experiment of the Superbuck converter is designed. The specific steps to obtain degradation experimental data are as follows:

[0014] Create a temperature stress of 120°C in a constant temperature box;

[0015] Place the SuperBuck converter in a constant temperature box;

[0016] The parameters of the converter components are measured at regular intervals to obtain degradation experimental data.

[0017] Furthermore, the component parameters include capacitor C3, C3 parasitic parameter RC3, capacitor C2 parasitic parameter RC2, inductor L2, L2 parasitic parameter RL2, inductor L3, L3 parasitic parameter RL3 and MOS parasitic parameter rds.

[0018] Furthermore, the key component degradation data includes degradation data of the capacitor C3, the parasitic parameter RC3 of C3, the parasitic parameter RC2 of capacitor C2, and the MOS parasitic parameter rds.

[0019] Furthermore, the specific steps for preprocessing the degradation data of key components are as follows:

[0020] The degradation data of key components are divided into multiple subsequences using a sliding window method, and preprocessed using the normalization method. The preprocessed sequences are used as the input of the CNN network.

[0021] Furthermore, the preprocessed sequence is:

[0022]

[0023] Among them, value is the preprocessed sequence, range is the normalized range, Data is the subsequence, col_max is the maximum value of each subsequence, and col_min is the minimum value of each subsequence.

[0024] Furthermore, the CNN-LSTM network consists of a convolutional layer, an LSTM layer, a fully connected layer, and an output layer.

[0025] Furthermore, the degradation curve prediction model is trained using the Adam optimizer.

[0026] Furthermore, the working principle of the Superbuck converter is:

[0027] When the PWM signal controlling the MOSFET is at a low level, the MOSFET is turned off, the diode D is turned on, and the current on the inductor L2 forms a loop through the capacitor C2, the diode D and the load R2;

[0028] When the PWM signal is high, the MOSFET is turned on, the diode D is turned off, the power supply forms a loop with the MOSFET and the load R2, and charges the inductor L2.

[0029] Furthermore, when iteratively training the CNN network and LSTM network, the degradation data of key components are divided into training set and test set in a ratio of 8:2.

[0030] Compared with the prior art, the present invention has the following beneficial effects:

[0031] The present invention is based on the analysis of the degradation model of the key components of the converter. Through the CNN-LSTM network, the degradation trend of key components can be predicted, which overcomes the problem of difficulty in accurately capturing the degradation trend of components. Since key components directly affect the health of the converter, the degradation model of the components and the future predicted values ​​of their degradation parameters are injected into the virtual simulation model. The output voltage ripple of the converter is collected and analyzed by the simulation model to determine the current operating state of the converter. This overcomes the problem of few characteristic signals reflecting the degradation law of the converter and the unclear parameter change law. Based on the above two points, it is finally possible to locate the moment of converter failure, thus solving the problem of difficulty in locating the moment of converter failure. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] Figure 1 SuperBuck topology and working principle diagram, where Figure 1 (a) is the topology of the SuperBuck converter. Figure 1(b) shows the working principle that when the PWM signal controlling the MOSFET is at a low level, the MOSFET is turned off, the diode D is turned on, and the current on the inductor L2 forms a loop through the capacitor C2, the diode D, and the load R2. Figure 1 (c) shows the working principle that when the PWM signal is high, the MOSFET is turned on, the diode D is turned off, the power supply forms a loop with the MOSFET and the load R2, and charges the inductor L2;

[0033] Figure 2 High temperature degradation experiment principle wiring diagram;

[0034] Figure 3 Equivalent circuit model diagram of capacitor;

[0035] Figure 4 SuperBuck converter virtual simulation model diagram;

[0036] Figure 5 Flowchart of converter failure determination method;

[0037] Figure 6 Loss function change graph;

[0038] Figure 7 Figure 1 shows the training results of the degradation model for key components of Sample 1;

[0039] Figure 8 Training results of the degradation model for key components of Sample 2. DETAILED DESCRIPTION

[0040] The present invention is described in detail below with reference to the accompanying drawings and specific embodiments. This embodiment is implemented based on the technical solution of the present invention, and provides a detailed implementation method and specific operation process, but the protection scope of the present invention is not limited to the following embodiments.

[0041] The present invention provides a method for determining failure of a low-power SuperBuck converter based on a component degradation model. The method comprises the following steps: performing a high-temperature degradation experiment on key components of the converter and collecting degradation data of the components; combining the collected degradation data with a deep learning prediction model to predict the degradation parameter values ​​of the components; injecting the predicted degradation parameter values ​​into a virtual simulation model of the converter, and determining whether the converter has failed and locating the moment of failure based on the output index characteristics of the converter virtual simulation model. The present invention does not affect the normal operation of the system and does not change the original circuit structure. It determines the failure state of the converter through external measurement. Compared with traditional determination schemes, the present invention is low-cost and easy to implement. The effectiveness of this method has been verified using multiple SuperBuck converter samples.

[0042] The present invention proposes a low-power SuperBuck converter failure judgment method based on a component degradation model, which can solve the problem of difficulty in locating the converter failure moment, thereby ensuring long-term and highly reliable operation of the power supply system.

[0043] In order to achieve the above object, the design scheme of the present invention includes the following steps:

[0044] (1) Design a high-temperature degradation experiment for the SuperBuck converter and collect the degradation data of converter components.

[0045] (2) Based on the collected degradation data, a deep learning prediction model is used to construct a component degradation model to predict the degradation parameter values ​​of the components and obtain a complete degradation curve of the components.

[0046] (3) Design a virtual simulation model of the SuperBuck converter, inject the parameter prediction values ​​in the degradation curve into the virtual simulation model, and determine whether the converter fails and locate the failure moment based on the output indicators of this model.

[0047] The present invention proposes a method for determining failure of a low-power SuperBuck converter based on a component degradation model, the method comprising the following steps:

[0048] S1. Taking the Superbuck converter as the object, design the Superbuck converter high temperature degradation experiment and obtain the degradation experimental data;

[0049] S2. Select and obtain degradation data of key components from degradation experimental data. The degradation data of key components is preprocessed and then used as input of the CNN network. The output of the CNN network is used as input of the LSTM network. The LSTM network outputs the degradation curve of the component. The CNN network and LSTM network are iteratively trained. The obtained CNN-LSTM network is used as the degradation curve prediction model of the key components.

[0050] S3. Obtain actual degradation data and input the key component degradation curve prediction model to obtain the actual degradation curve. Inject the current degradation parameter prediction values ​​of the key components in the actual degradation curve into the Superbuck converter to obtain the output voltage ripple of the Superbuck converter. If the output voltage ripple does not exceed the failure threshold, continue to inject the degradation parameter prediction value at the next moment until the degradation parameter prediction values ​​at all moments are injected. Otherwise, record the moment corresponding to the current degradation parameter prediction value as the converter failure moment.

[0051] The present invention is divided into four parts, which describe in detail the failure judgment method of a low-power SuperBuck converter based on a component degradation model: (1) designing a high-temperature degradation experiment scheme for the converter and collecting component degradation data; (2) realizing component degradation parameter value prediction based on a deep learning prediction model; (3) judging the converter failure state and locating the failure moment based on a converter virtual simulation model; (4) taking two different SuperBuck converter samples as examples, the practical application of the present invention is demonstrated and the feasibility of the method is verified.

[0052] Part 1:

[0053] The topology of the SuperBuck converter is as follows: Figure 1 As shown in (a), the output voltage is controlled by controlling the on and off of the MOSFET. Figure 1 As shown in (b), when the PWM signal controlling the MOSFET is low, the MOSFET is turned off, the diode D is turned on, and the current on the inductor L2 forms a loop through the capacitor C2, the diode D, and the load R2; when the PWM signal is high, the MOSFET is turned on, the diode D is turned off, the power supply forms a loop with the MOSFET and the load R2, and charges the inductor L2, as shown in Figure 2. Figure 1 (c) shown.

[0054] The Superbuck converter was used as the experimental object, and 120℃ high temperature was used as the accelerated degradation condition for experimental testing. The wiring diagram of the high temperature degradation experiment is as follows: Figure 2 The experimental equipment used is shown in Table 1.

[0055] Table 1 Experimental equipment

[0056]

[0057] The specific experimental steps are as follows: Step 1: Create a 120°C temperature stress in a constant temperature chamber; Step 2: Place the SuperBuck converter in the constant temperature chamber; Step 3: Measure the converter's key component parameters every hour and record the values. Capacitor C2 is the input-side capacitor and has minimal impact on the converter, but the influence of its parasitic parameter RC2 must be considered. Therefore, the component parameters that need to be measured include C3, RC3, RC2, L2, RL2, L3, RL3, and rds. The degradation of the SuperBuck converter's key component parameters is shown in Table 2.

[0058] Table 2. Parameter degradation of key components of SuperBuck converter

[0059]

[0060] Part II:

[0061] The main factor in the performance degradation of DC / DC converters is the degradation of components. Therefore, building a component degradation model can better analyze the performance status of the converter and measure its failure time. Taking electrolytic capacitors (output capacitors) as an example, electrolytic capacitors are one of the most critical components in DC / DC converters, and their health status directly affects the quality of DC power. At the same time, electrolytic capacitors are also one of the most fragile components in the power transmission part of the converter, and have the most obvious aging characteristics. The equivalent circuit model of a capacitor includes a capacitor (C) and an equivalent series resistor (ESR), such as Figure 3 shown.

[0062] Extensive experiments have shown that the key components that most significantly impact the health of the SuperBuck converter are the output capacitor C3, C3 parasitic parameter RC3, C2 parasitic parameter RC2, and MOS parasitic parameter rds. Therefore, four component degradation models were constructed.

[0063] Degradation data obtained from degradation experiments is discrete, but predicting future trends requires a continuous mathematical model. Therefore, it's necessary to abstract this discrete data set into a mathematical description. Component degradation data is treated as time series data. The CNN-LSTM time series prediction model can learn and store a large number of input / output mapping relationships without the need for pre-defined mathematical equations describing these mapping relationships.

[0064] First, the original data is divided into multiple subsequences using a sliding window method. The sliding window length is set to 30, which is the number of time steps of each input subsequence. The normalization method is used to standardize the signal data of different measurement ranges, limiting the output of the data to [0, 1], as shown in formula (1).

[0065]

[0066] Where value is the normalized value, range is the normalized range, Data is the input data, col_max is the maximum value of each subsequence, and col_min is the minimum value of each subsequence.

[0067] The processed data is used as the input of the CNN network, and the output of the network is used as the input of the LSTM network, and the output is the degradation curve of the component. All data are divided into training set and test set in a ratio of 8:2.

[0068] The CNN-LSTM network consists of a convolutional layer (each convolutional layer consists of a convolution operation and an activation function), an LSTM layer, a fully connected layer, and an output layer. The output layer outputs a time series prediction value. The specific parameters of the network are shown in Table 3. The activation function uses the ReLU activation function, and the specific expression is shown in (2). The training error is backpropagated using the Adam optimizer, and the learning rate is set to 0.0001. betas = (0.5, 0.999) is used to set the momentum parameter of the Adam optimizer. The predicted value is denormalized to restore it to the scale of the original data, and finally the complete prediction result value is obtained.

[0069] f(x)=max(0,x) (2)

[0070] Table 3 Parameters of CNN-LSTM network

[0071]

[0072] Part III:

[0073] A high-fidelity virtual simulation model of the SuperBuck converter is built in Simulink. Figure 4 As shown in Figure 2, extensive simulation experiments demonstrate that the outputs of the virtual simulation model are similar to observations of the physical converter, indicating that the two models exhibit the same mechanism. Therefore, the virtual simulation model can be used to monitor and predict the future operating state of the SuperBuck converter without interrupting its operation.

[0074] Output voltage ripple is one of the most critical output specifications for DC / DC converters. Excessive ripple can reduce converter efficiency and generate surge voltages or currents, potentially damaging electrical equipment. Therefore, analyzing changes in the converter's output voltage ripple can reveal the converter's current operating status.

[0075] The main idea behind the converter failure identification method proposed in this paper is to inject the degradation parameter prediction values ​​of key components from the previously trained degradation curve into the converter virtual simulation model, collect its output indicators (output voltage ripple) and determine whether the indicators exceed the failure threshold. The specific identification steps are as follows:

[0076] a. If the indicator does not exceed the failure threshold, inject the subsequent degradation parameter value;

[0077] b. If the indicator exceeds the failure threshold, the time T corresponding to the currently injected degradation parameter value is recorded.

[0078] In Simulink, the output ripple of the converter is collected when each set of C3, RC3, RC2, and rds parameters changes. The output voltage ripple when C3 and RC3 degrade is shown in Table 4. Generally, an output ripple voltage exceeding 10% to 20% of the nominal value may trigger an alarm, and exceeding 30% to 50% may be considered a failure. Time T corresponds to the failure moment of the physical entity converter, ultimately achieving the failure state judgment of the low-power SuperBuck converter and the location of the failure moment. The complete flow chart of this judgment method is shown in the figure below. Figure 5 shown.

[0079] Table 4 Output voltage ripple size when C3 and RC3 of SuperBuck converter are degraded

[0080]

[0081] Part 4:

[0082] A complete experimental test was conducted using two different SuperBuck converter samples to verify the effectiveness of the converter failure identification method. 500 sets of degradation experimental data were collected for each sample for subsequent training.

[0083] The model training experimental platform and environment are as follows: CPU is i5-10200H, running environment is Python 3.9. The training batch size is 4, the learning rate is 0.0001, and the training is 100 rounds. The mean square error (MSE) loss function is used to evaluate the performance of the model. The change of the loss function during the training process is shown in the figure below. Figure 6 As shown in the figure, the training results of the key component degradation models of the two SuperBuck converters are as follows: Figure 7 、 Figure 8 As shown in Table 5, the mean square errors of the key component degradation models after training are shown in Table 5 and Table 6 respectively.

[0084] Table 5 Mean square error of key components of sample 1 after training

[0085]

[0086] Table 6 Mean square error of key components of sample 2 after training

[0087]

[0088] Tables 5 and 6 show that the present invention has a good performance in predicting degradation trends for different SuperBuck converter key component parameters. The training mean square error for key components C3, RC2, and rds in both samples is less than 3%, while the training mean square error for RC3 is slightly higher.

[0089] The degradation conditions of key components of two SuperBuck converter samples were collected when the output voltage ripple exceeded the nominal value by 10%, 20%, 30%, 40%, and 50%, respectively, as shown in Tables 7 and 8.

[0090] Table 7 Degradation of key components when the output voltage ripple of sample 1 exceeds the nominal value by 10% to 50%

[0091]

[0092]

[0093] Table 7 Degradation of key components when the output voltage ripple of sample 2 exceeds the nominal value by 10% to 50%

[0094]

[0095] As can be seen from Tables 7 and 8, the present invention solves, to a certain extent, the problem of difficulty in locating the moment of SuperBuck converter failure. Without affecting the normal operation of the system or changing the original circuit structure, it provides technicians with a converter failure warning method. This method can obtain early warnings before the converter health status deteriorates significantly and promptly capture the moment of converter failure. This has guiding significance for the application of long-term, high-reliability products.

[0096] The above describes in detail the preferred embodiments of the present invention. It should be understood that those skilled in the art can make numerous modifications and variations based on the concepts of the present invention without inventive effort. Therefore, any technical solutions that can be derived by those skilled in the art through logical analysis, reasoning, or limited experimentation based on the concepts of the present invention and the prior art should be within the scope of protection defined by the claims.

Claims

1. A method for determining failure of a low-power SuperBuck converter based on a component degradation model, characterized in that: The method comprises the following steps: S1. Taking the Superbuck converter as the object, design the Superbuck converter high temperature degradation experiment and obtain the degradation experimental data; S2. Select and obtain degradation data of key components from degradation experimental data. The degradation data of key components is preprocessed and then used as input of the CNN network. The output of the CNN network is used as input of the LSTM network. The LSTM network outputs the degradation curve of the component. The CNN network and LSTM network are iteratively trained. The obtained CNN-LSTM network is used as the degradation curve prediction model of the key components. S3. Obtain actual degradation data and input the key component degradation curve prediction model to obtain the actual degradation curve. Inject the current degradation parameter prediction values ​​of the key components in the actual degradation curve into the Superbuck converter to obtain the output voltage ripple of the Superbuck converter. If the output voltage ripple does not exceed the failure threshold, continue to inject the degradation parameter prediction value at the next moment until the degradation parameter prediction values ​​at all moments are injected. Otherwise, record the moment corresponding to the current degradation parameter prediction value as the converter failure moment.

2. The method for determining failure of a low-power SuperBuck converter based on a component degradation model according to claim 1, wherein: The specific steps to design a high-temperature degradation experiment for a Superbuck converter and obtain degradation experimental data are as follows: Create a temperature stress of 120°C in a constant temperature box; Place the SuperBuck converter in a constant temperature box; The parameters of the converter components are measured at regular intervals to obtain degradation experimental data.

3. The method for determining failure of a low-power SuperBuck converter based on a component degradation model according to claim 2, wherein: The component parameters include capacitor C3, C3 parasitic parameter RC3, capacitor C2 parasitic parameter RC2, inductor L2, L2 parasitic parameter RL2, inductor L3, L3 parasitic parameter RL3 and MOS parasitic parameter rds.

4. The method for determining failure of a low-power SuperBuck converter based on a component degradation model according to claim 3, wherein: The degradation data of key components include the degradation data of capacitor C3, C3 parasitic parameter RC3, capacitor C2 parasitic parameter RC2 and MOS parasitic parameter rds.

5. The method for determining failure of a low-power SuperBuck converter based on a component degradation model according to claim 4, characterized in that: The specific steps for preprocessing the degradation data of key components are: The degradation data of key components are divided into multiple subsequences using a sliding window method, and preprocessed using the normalization method. The preprocessed sequences are used as the input of the CNN network.

6. The method for determining failure of a low-power SuperBuck converter based on a component degradation model according to claim 5, characterized in that: The preprocessed sequence is: Among them, value is the preprocessed sequence, range is the normalized range, Data is the subsequence, col_max is the maximum value of each subsequence, and col_min is the minimum value of each subsequence.

7. The method for determining failure of a low-power SuperBuck converter based on a component degradation model according to claim 1, wherein: The CNN-LSTM network consists of a convolutional layer, an LSTM layer, a fully connected layer, and an output layer.

8. The method for determining failure of a low-power SuperBuck converter based on a component degradation model according to claim 7, characterized in that: The degradation curve prediction model is trained using the Adam optimizer.

9. The method for determining failure of a low-power SuperBuck converter based on a component degradation model according to claim 1, wherein: The working principle of the Superbuck converter is: When the PWM signal controlling the MOSFET is at a low level, the MOSFET is turned off, the diode D is turned on, and the current on the inductor L2 forms a loop through the capacitor C2, the diode D and the load R2; When the PWM signal is high, the MOSFET is turned on, the diode D is turned off, the power supply forms a loop with the MOSFET and the load R2, and charges the inductor L2.

10. The method for determining failure of a low-power SuperBuck converter based on a component degradation model according to claim 1, wherein: When iteratively training the CNN network and LSTM network, the degradation data of key components are divided into training set and test set in a ratio of 8:2.

Images