High-reliability radar array surface power supply system health management method

Abstract

The invention discloses a health management method for a high-reliability radar array surface power supply system, and the method comprises the steps: collecting the voltage and current characteristic parameters of a power supply system through a voltage collection circuit and a current collection circuit, and carrying out the data cleaning, dimensionality reduction and normalization processing; extracting a fault feature vector, training an array power supply system health management model based on a hidden Markov model, taking the health state of the power supply system as a hidden state, taking the fault feature vector as an observation value, training the model by using historical data, and predicting the current health state; and performing judgment based on a set health state threshold value, when a health state prediction result exceeds a state judgment threshold value range, judging that maintenance is performed, and when the health state prediction result is within the threshold value range, judging that monitoring is performed. According to the invention, high-precision judgment and positioning of the power supply system fault are realized.

Description

Technical Field

[0001] This invention relates to radar technology, and more specifically to a health management method for a high-reliability radar array power supply system. Background Technology

[0002] The power supply system is a crucial component of a radar array. Failures or potential faults in the power supply system can lead to unpredictable and severe consequences. Therefore, health management of the array's power supply system to ensure stable and reliable power supply, and to proactively eliminate potential faults, is of paramount importance. Currently, health management of radar array power supply systems often relies on hard fault detection or single physical quantity detection for fault location and troubleshooting. These methods, to varying degrees, suffer from limitations in eliminating potential faults and the inherent limitations of the monitoring parameters themselves. In addition, some literature proposes using digital twin technology to digitally model the radar array power supply system and simultaneously monitor component characteristic parameters in real time, thereby meeting health management needs. While this method offers high accuracy in fault detection and lifespan prediction, its modeling is complex, resource-intensive, and not conducive to low-cost industrial application. Summary of the Invention

[0003] The purpose of this invention is to provide a health management method for a high-reliability radar array power supply system, so as to solve the problems of high reliability requirements, high maintenance costs, and complex operating environment of radar array power supply systems.

[0004] The technical solution for achieving the objective of this invention is: a health management method for a high-reliability radar array power supply system, comprising the following steps:

[0005] Data acquisition steps: The voltage and current characteristic parameters of the power supply system are acquired through the voltage and current acquisition circuit. The characteristic parameters include capacitor current, inductor voltage, input voltage, output voltage and output current.

[0006] Data preprocessing steps: Clean the collected data, delete abnormal data, and perform dimensionality reduction and normalization on the cleaned data;

[0007] Fault feature vector extraction steps: Combining the capacitor degradation model, inductor degradation model and power supply system equivalent model, extract fault feature vectors from the preprocessed data. The fault feature vectors include inductance value L, capacitance value C, mean value of inductor current curve Lmean, peak value of inductor current waveform Ltop, valley value of inductor current waveform Lbottom, mean value of capacitor voltage curve Cmean, peak value of capacitor voltage waveform Ctop, and valley value of capacitor voltage waveform Cbottom.

[0008] Health status prediction steps: Train a health management model for the power supply system based on a hidden Markov model, where the health status of the power supply system is taken as the hidden state, the fault feature vector is taken as the observation value, historical data is used to train the model, and the current health status is predicted.

[0009] Decision-making steps: Based on the health status prediction results, a judgment is made based on the pre-set health status threshold. If the health status prediction results exceed the status judgment threshold range, it is judged as maintenance; if the health status prediction results are within the threshold range, it is judged as monitoring.

[0010] If the fault is determined to be requiring repair, the fault component location step is executed: a power supply system fault classifier is trained based on a support vector machine, and the fault component is located based on the fault feature vector.

[0011] If the condition is determined to be under monitoring, the status assessment step is executed: the power supply system health status monitoring mechanism based on the hidden Markov model is used to assess the status of key components.

[0012] Furthermore, in the data acquisition step, the inductor current I is periodically calculated based on the equivalent small parameter method model. L and capacitor voltage V C The equivalent small-parameter method model is expressed as:

[0013] x=a 00 +(a 11 *e jτ +a 22 *e j2τ +a 32 *e j3τ +cc)

[0014] In the formula, x represents the inductor current and capacitor voltage, a 00 a 11 a 22 a 32 τ is the coefficient, cc is the time variable, and cc represents the conjugate complex number.

[0015] Furthermore, in the fault feature vector extraction step, the calculated inductor current I is used as the basis for... L and capacitor voltage V C Periodically calculate the inductance value L and the capacitance value C, where:

[0016] The inductance value L is obtained through the formula L=V L *△t / △I L calculate,

[0017] The capacitance value C is obtained using the formula C=I C *△t / △V C calculate,

[0018] In the formula, Δt is the charging time of the inductor and capacitor.

[0019] Furthermore, in the fault feature vector extraction step, 50% of the initial inductance value of the inductor and the initial capacitance value of the capacitor are set as the device failure threshold.

[0020] Furthermore, in the fault feature vector extraction step, the inductor current waveform and capacitor voltage waveform are periodically plotted based on the equivalent small parameter method model, and the following features are extracted from the waveforms:

[0021] Inductor current curve mean Lmean, inductor current waveform peak value Ltop, inductor current waveform valley value Lbottom;

[0022] The mean value of the capacitor voltage curve is Cmean, the peak value of the capacitor voltage waveform is Ctop, and the valley value of the capacitor voltage waveform is Cbottom.

[0023] Furthermore, in the data preprocessing step, the normalization process uses the min-max normalization method, with the following formula:

[0024] X nor =(xx min ) / (x max -x min )

[0025] In the formula, x min、 x max These are the minimum and maximum values ​​of the data, x and X, respectively. nor The data are shown before and after normalization, respectively.

[0026] Furthermore, in the fault component localization step, when training the power supply system fault classifier using a support vector machine, the main fault modes of the power supply system include inductor degradation F1, capacitor degradation F2, and both inductor and capacitor degradation F3. A dataset is constructed based on these three fault modes, and three support vector machine binary classification models F1(x), F2(x), and F3(x) are trained. Then, the data samples to be predicted are input into the classifier, and the inductor degradation F1, capacitor degradation F2, and both inductor and capacitor degradation F3 are located according to the maximum output result.

[0027] Furthermore, in the health status prediction step, the hidden Markov model subdivides the health status of the power supply system into four states: healthy, sub-healthy, critically healthy, and faulty.

[0028] Furthermore, in the health status prediction step, the hidden Markov model is trained using the Baum-Welch algorithm, and the health status prediction uses the Viterbi algorithm. The specific steps are as follows:

[0029] The fault feature vector is trained and adjusted based on the Baum-Welch algorithm. First, an initial estimate is set for the model parameters λ = (A, B, π). The state transition matrix A is initialized so that it is more likely to remain in a "healthy" state and has a smaller probability of transitioning to a "sub-healthy" state. The observation probability matrix B follows a Gaussian distribution with a mean of the rated voltage and a small variance. The initial state distribution π starts from the "normal" state, i.e., π = [1, 0, 0]. Then, the forward probability α and the backward probability β are calculated. Using the forward and backward probabilities, two key intermediate variables are calculated. Finally, the model parameters λ = (A, B, π) are re-estimated and iterated and converged.

[0030] Based on the fault feature vectors obtained by training the Baum-Welch algorithm, health status prediction is performed using the Viterbi algorithm. First, an initialization matrix is ​​defined, which is the state ψ(i-1) before the maximum probability δ(i) is reached. Then, for each state j at different times, the maximum probability of reaching j is calculated, and the path with the maximum probability is recorded. Next, the state with the highest probability at the final time is found as the endpoint of the path. Finally, starting from the endpoint, backtracking is performed to find the complete optimal state sequence.

[0031] Furthermore, the equivalent model of the power supply system is a BUCK topology model, constructed based on the equivalent small parameter method, wherein the time-varying differential equation of the BUCK converter is:

[0032] G1(P)f+ G0(P)x=u

[0033] In the formula, G0(P) and G1(P) are both coefficient matrices, u is a constant, and f is a function of x and u.

[0034] Compared with existing technologies, the significant advantages of this invention are as follows: 1) By introducing fault feature values ​​such as inductance value L, capacitance value C, mean inductor current curve Lmean, peak inductor current waveform Ltop, valley inductor current waveform Lbottom, mean capacitor voltage curve Cmean, peak capacitor voltage waveform Ctop, and valley capacitor voltage waveform Cbottom, the degradation status of the power supply system is more comprehensively reflected. Using these data as observations and training a Hidden Markov Model, the deviation between the current values ​​and the healthy state values ​​is obtained, achieving a high-precision health status assessment. 2) Based on the support vector machine algorithm, fault location of power supply system components is performed, and the fault types are subdivided into three categories. High-precision fault determination and location of the power supply system are achieved by judging the output results of the classifier. Attached Figure Description

[0035] Figure 1 This is a flowchart of the method. Detailed Implementation

[0036] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0037] A health management method for a high-reliability radar array power supply system, the technical solution of which includes:

[0038] Voltage and current characteristic parameters of the power supply system are collected by voltage and current acquisition circuits. At the same time, abnormal, defective and duplicate data are removed through data cleaning. The relevant data are subjected to dimensionality reduction and normalization. Then, fault feature vectors are extracted from the processed data by combining capacitor degradation model, inductor degradation model and power supply system equivalent model. Using a large amount of extracted data, a health management model of the power supply system is trained based on a hidden Markov model. At this time, the health status of the power supply system is regarded as the hidden state, and the collected and processed fault feature vectors are used as monitoring values. A large amount of historical data is used for training and health status prediction in the system. Finally, based on the prediction results, it is determined whether the key components of the current power supply system need to be repaired or monitored. If it is to be repaired, the component is located; if it is to be monitored, its expected life is estimated.

[0039] Furthermore, the acquisition of capacitor current, inductor voltage, input voltage, output voltage, and output current of the power supply system is based on the equivalent small-parameter method model x=a. 00 +(a 11 *e jτ +a 22 *e j2τ +a 32 *e j3τ +cc) Periodic calculation of inductor current I L and capacitor voltage V C Based on L=V L *△t / △I L and C=I C *△t / △V CThe inductance value L and capacitance value C are periodically calculated. Simultaneously, 50% of the initial inductance and capacitance values ​​are set as the device failure threshold. Based on the equivalent small parameter method model, the waveforms of inductor current and capacitor voltage are periodically plotted. The mean, peak, and bottom values ​​of different curves are obtained. The inductance value L, capacitance value C, mean inductor current curve Lmean, peak inductor current waveform Ltop, valley inductor current waveform Lbottom, mean capacitor voltage curve Cmean, peak capacitor voltage waveform Ctop, and valley capacitor voltage waveform Cbottom are used as fault feature vectors for the power supply system. These fault feature vectors are then subjected to min-max normalization.

[0040] Furthermore, the faulty component localization is mainly based on the support vector machine training of the power supply system fault classifier. The main fault modes of the power supply system are set as inductor degradation F1, capacitor degradation F2, and both inductor and capacitor degradation F3. Data sets are constructed according to the above three fault modes to form three support vector machine binary classification models F1(x), F2(x), and F3(x). The data samples to be predicted are input into the classifier, and the component is located according to the maximum output result.

[0041] In summary, this invention improves the health management level of radar array power supply systems, enhances the monitoring accuracy of key components of the power supply system, and enables early prevention and replacement of potential faults, thereby significantly reducing the maintenance cost of the power supply system.

[0042] Example

[0043] To verify the effectiveness of the present invention, the following experimental design was conducted.

[0044] This invention provides a health management method for a high-reliability radar array power supply system, which mainly includes device model establishment, key feature parameter extraction, data cleaning, component fault location, and hidden Markov state monitoring. Based on the equivalent small-parameter method, a BUCK topology model in the array power supply system is constructed. First, the BUCK converter is expressed using time-varying differential equations, as shown in the following formula: where G0(P) and G1(P) are coefficient matrices, x is a state variable vector, u is a constant, and f is a function of x and u.

[0045] G1(P)f+ G0(P)x=u (1)

[0046] Based on the principle of the perturbation method, the above state vector x is expanded into a Fourier series, and a small parameter m is introduced. Based on the harmonic balance method and the perturbation method, the above formula 1 is modified into formula 2:

[0047] G1(P)F-m+ G0(P)x=-mG1(P)F' (2)

[0048] Based on the harmonic balance method, the above nonlinear equations are transformed into linear equations. The final linear equations consist of DC and AC components, and the corresponding expressions are shown in Formula 3 below:

[0049] x=a 00 +(a 11 *e jτ +a 22 *e j2τ +a 32 *e j3τ +cc) (3)

[0050] Based on the aging test model of electrolytic capacitors, an empirical degradation model of electrolytic capacitors is obtained. The corresponding degradation model calculation formula is as follows, and the failure criterion of electrolytic capacitors is defined as 50% of the initial value.

[0051] C=I C *△t / △V C (4)

[0052] Based on the inductance model, an empirical degradation model for inductance is obtained. The corresponding calculation formula is as follows, and the inductance failure criterion is defined as 50% of the initial value.

[0053] L=V L *△t / △I L (5)

[0054] Based on Formula 3 above, and combined with the periodic calculation of inductor current I... L and capacitor voltage V C It periodically generates inductor current waveforms and capacitor voltage waveforms.

[0055] Fault feature parameters were extracted from the above model. The main extracted and stored data included eight fault feature parameters: inductance value L, capacitance value C, mean inductor current curve Lmean, peak inductor current waveform Ltop, valley inductor current waveform Lbottom, mean capacitor voltage curve Cmean, peak capacitor voltage waveform Ctop, and valley capacitor voltage waveform Cbottom. Damaged and incomplete data were deleted. The filtered data was then standardized using the following min-max normalization formula.

[0056] X nor =(xx min ) / (x max -x min (6)

[0057] Based on data-driven principles, this paper describes a method for locating power supply system faults using support vector machines. The specific location steps are as follows:

[0058] Step 1: Extract the fault feature vectors for each mode of the power supply system: inductance value L, capacitance value C, mean value of inductor current curve Lmean, peak value of inductor current waveform Ltop, valley value of inductor current waveform Lbottom, mean value of capacitor voltage curve Cmean, peak value of capacitor voltage waveform Ctop, valley value of capacitor voltage waveform Cbottom.

[0059] Step 2: Perform data preprocessing on the fault feature vector, specifically including data cleaning, data dimensionality reduction, and other operations.

[0060] Step 3: Organize the processed fault feature vectors into a dataset and perform classification processing.

[0061] Step 4: Based on the support vector machine classification algorithm, the fault model of the power supply system is diagnosed and located. At the same time, the fault data is used as the training set, specifically the inductor degradation F1 dataset, the capacitor degradation F2 dataset, and the inductor and capacitor degradation F3 dataset, to train the support vector machine model, with the corresponding models being F1(x), F2(x), and F3(x), respectively.

[0062] Step 5: Input new fault data into the trained support vector machine model to determine and locate the fault.

[0063] A power supply system health status monitoring mechanism is established based on a Hidden Markov Model (HMM). The power supply system health status is subdivided into four states: healthy, sub-healthy, critically healthy, and faulty. The power supply system is modeled from healthy to faulty states, and feature data from all states are collected and used as the training dataset for the HMM. The Baum-Welch algorithm is primarily used to obtain the HMM parameters. The newly collected feature parameters (L, C, Lmean, Ltop, Lbottom, Cmean, Ctop, Cbottom) are used as corresponding state observation parameters, and state prediction is performed based on the model and the Viterbi algorithm. The specific steps are as follows: Based on the Baum-Welch algorithm, the feature vector is trained and adjusted. First, an initial estimate is set for the model parameters λ = (A, B, π). The state transition matrix A is initialized so that a state from "healthy" is more likely to remain "healthy" and has a lower probability of transitioning to "sub-healthy". The observation probability matrix B follows a Gaussian distribution with a mean of the rated voltage and a small variance. The initial state distribution π starts from the "normal" state, i.e., π = [1, 0, 0]. Then, the forward probability α and the backward probability β are calculated. Next, using the forward and backward probabilities, two key intermediate variables are calculated. Finally, the model parameters λ = (A, B, π) are re-estimated, and iteration and convergence are performed. Based on the Viterbi algorithm, health state prediction is performed. First, an initialization matrix is ​​defined, i.e., the maximum probability δ(i) and the state ψ(i-1) before δ(i) reaches its maximum are defined. Then, for each state j at different times, the maximum probability of reaching j is calculated, and the path of this maximum probability is recorded from which state i. Next, find the state with the highest probability at the final moment, and use it as the end point of the path. Finally, start from the end point and backtrack to find the complete optimal state sequence.

[0064] Since the power supply topology of the array power supply system is relatively large, in actual health status monitoring, it is necessary to monitor multiple power supply units within the system. Therefore, it is necessary to establish equivalent small parameter models, inductor-capacitor models, support vector machine models, and hidden Markov models for multiple power supply units to achieve comprehensive monitoring of the power supply system.

[0065] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0066] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of this application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these modifications and improvements all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.

Claims

1. A health management method for a high-reliability radar array power supply system, characterized in that, Includes the following steps: Data acquisition steps: The voltage and current characteristic parameters of the power supply system are acquired through the voltage and current acquisition circuit. The characteristic parameters include capacitor current, inductor voltage, input voltage, output voltage and output current. Data preprocessing steps: Clean the collected data, delete abnormal data, and perform dimensionality reduction and normalization on the cleaned data; Fault feature vector extraction steps: Combining the capacitor degradation model, inductor degradation model and power supply system equivalent model, extract fault feature vectors from the preprocessed data. The fault feature vectors include inductance value L, capacitance value C, mean value of inductor current curve Lmean, peak value of inductor current waveform Ltop, valley value of inductor current waveform Lbottom, mean value of capacitor voltage curve Cmean, peak value of capacitor voltage waveform Ctop, and valley value of capacitor voltage waveform Cbottom. Health status prediction steps: Train a health management model for the power supply system based on a hidden Markov model, where the health status of the power supply system is taken as the hidden state, the fault feature vector is taken as the observation value, historical data is used to train the model, and the current health status is predicted. Decision-making steps: Based on the health status prediction results, a judgment is made based on the pre-set health status threshold. If the health status prediction results exceed the status judgment threshold range, it is judged as maintenance; if the health status prediction results are within the threshold range, it is judged as monitoring. If the fault is determined to be requiring repair, the fault component location step is executed: a power supply system fault classifier is trained based on a support vector machine, and the fault component is located based on the fault feature vector. If the condition is determined to be under monitoring, the status assessment step is executed: the power supply system health status monitoring mechanism based on the hidden Markov model is used to assess the status of key components.

2. The health management method for a high-reliability radar array power supply system according to claim 1, characterized in that, In the data acquisition step, the inductor current I is periodically calculated based on the equivalent small parameter method model. L and capacitor voltage V C The equivalent small-parameter method model is expressed as: x=a 00 +(a 11 *And jτ +a 22 *And j2τ +a 32 *And j3τ +cc) In the formula, x represents the inductor current and capacitor voltage, a 00 a 11 a 22 a 32 τ is the coefficient, cc is the time variable, and cc represents the conjugate complex number.

3. The health management method for a high-reliability radar array power supply system according to claim 1, characterized in that, In the fault feature vector extraction step, the calculated inductor current I is used as the basis for the extraction. L and capacitor voltage V C Periodically calculate the inductance value L and the capacitance value C, where: The inductance value L is obtained through the formula L=V L *△t / △I L calculate, The capacitance value C is obtained using the formula C=I. C *△t / △V C calculate, In the formula, Δt is the charging time of the inductor and capacitor.

4. The health management method for a high-reliability radar array power supply system according to claim 3, characterized in that, In the fault feature vector extraction step, 50% of the initial inductance value of the inductor and the initial capacitance value of the capacitor are set as the device failure threshold.

5. The health management method for a high-reliability radar array power supply system according to claim 2, characterized in that, In the fault feature vector extraction step, the inductor current waveform and capacitor voltage waveform are periodically plotted based on the equivalent small parameter method model, and the following features are extracted from the waveforms: Inductor current curve mean Lmean, inductor current waveform peak value Ltop, inductor current waveform valley value Lbottom; The mean value of the capacitor voltage curve is Cmean, the peak value of the capacitor voltage waveform is Ctop, and the valley value of the capacitor voltage waveform is Cbottom.

6. The health management method for a high-reliability radar array power supply system according to claim 1, characterized in that, In the data preprocessing step, the normalization process uses the min-max normalization method, with the following formula: X nor =(x-x min ) / (x max -x min ) In the formula, x min、 x max These are the minimum and maximum values ​​of the data, x and X, respectively. nor The data are shown before and after normalization, respectively.

7. The health management method for a high-reliability radar array power supply system according to claim 1, characterized in that, In the fault component localization step, when training the power supply system fault classifier using a support vector machine, the main fault modes of the power supply system include inductor degradation F1, capacitor degradation F2, and both inductor and capacitor degradation F3. A dataset is constructed based on these three fault modes, and three support vector machine binary classification models F1(x), F2(x), and F3(x) are trained. Then, the data samples to be predicted are input into the classifier, and the inductor degradation F1, capacitor degradation F2, and both inductor and capacitor degradation F3 are located according to the maximum output result.

8. The health management method for a high-reliability radar array power supply system according to claim 1, characterized in that, In the health status prediction step, the Hidden Markov Model subdivides the health status of the power supply system into four states: healthy, sub-healthy, critically healthy, and faulty.

9. The health management method for a high-reliability radar array power supply system according to claim 1, characterized in that, In the health status prediction step, the hidden Markov model is trained using the Baum-Welch algorithm, and the health status prediction uses the Viterbi algorithm. The specific steps are as follows: The fault feature vector is trained and adjusted based on the Baum-Welch algorithm. First, an initial estimate is set for the model parameters λ = (A, B, π). The state transition matrix A is initialized so that it is more likely to remain in a "healthy" state and has a smaller probability of transitioning to a "sub-healthy" state. The observation probability matrix B follows a Gaussian distribution with a mean of the rated voltage and a small variance. The initial state distribution π starts from the "normal" state, i.e., π = [1, 0, 0]. Then, the forward probability α and the backward probability β are calculated. Using the forward and backward probabilities, two key intermediate variables are calculated. Finally, the model parameters λ = (A, B, π) are re-estimated and iterated and converged. Based on the fault feature vectors obtained by training the Baum-Welch algorithm, health status prediction is performed using the Viterbi algorithm. First, an initialization matrix is ​​defined, which is the state ψ(i-1) before the maximum probability δ(i) is reached. Then, for each state j at different times, the maximum probability of reaching j is calculated, and the path with the maximum probability is recorded. Next, the state with the highest probability at the final time is found as the endpoint of the path. Finally, starting from the endpoint, backtracking is performed to find the complete optimal state sequence.

10. The health management method for a high-reliability radar array power supply system according to claim 2, characterized in that, The equivalent model of the power supply system is the BUCK topology model, constructed based on the equivalent small-parameter method, wherein the time-varying differential equation of the BUCK converter is: G1(P)f+ G0(P)x=u In the formula, G0(P) and G1(P) are both coefficient matrices, u is a constant, and f is a function of x and u.

Images