Aerial series arc fault detection method based on improved levene test and JS divergence

Abstract

The application discloses an aviation series arc fault detection method based on improved Levene test and JS divergence, which improves the calculation method of the conversion variable in the original Levene test process, solves the overlapping problem of P eigenvalues under the single-phase cross talk interference condition and the arc fault condition, and simultaneously proposes that the Jensen-Shannon divergence and the improved Levene test P value feature are jointly applied to the detection of the aviation alternating current arc fault in order to avoid the contingency of a single feature in identifying the arc fault. After analysis, the unified threshold of the two features is given, and the aviation electrical system arc fault diagnosis process is realized. The arc fault detection accuracy of the aviation electrical system reaches more than 97.5%, and the recognition accuracy is high.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of fault arc detection of aircraft electrical systems, and particularly to an aircraft series arc fault detection method based on improved Levene test and JS divergence. BACKGROUND

[0002] Because the aircraft wiring harness is in a harsh environment of vibration, high temperature, humidity and the like for a long time, the insulation layer will accelerate decomposition and aging, and the wear between the cable surfaces will also increase, which easily induces intermittent arc faults. The aircraft electrical system adopts a power supply of 115V / 400Hz, resulting in arc faults with characteristics of short duration and strong concealment. Moreover, because the wires of the aircraft electrical system are concentratedly bundled and placed in the frame, when an arc fault occurs in a certain line, it is easy to cause crosstalk interference to other lines. If physical quantity changes such as arc light, arc sound, arc temperature and electromagnetic radiation are measured to detect arc faults, the installation of sensors will be limited by space, which adversely affects the identification of arc faults.

[0003] In order to solve the above problems, the person skilled in the art detects the arc fault of the aircraft line by studying the time domain, frequency domain and time-frequency domain characteristics of the aircraft arc fault current or voltage signal. Among them, the frequency domain method mainly uses Fast Fourier Transform (FFT) to calculate the harmonic content of the arc current signal, and judges the arc fault by measuring the change of the harmonic content. The frequency domain feature can effectively reflect the change of the harmonic content of the arc current signal, and when harmonic interference occurs in the line, the current signal in the disturbed circuit will also change in harmonic, which often causes misjudgment. Compared with the frequency domain method, the time-frequency domain method can analyze the time domain and frequency domain information of the signal, and the main analysis methods include wavelet transform, wavelet packet transform and Empirical Mode Decomposition (EMD), etc. However, the effective frequency band of the wavelet transform reflecting the arc fault characteristics is different under different loads, which makes it difficult to select a suitable wavelet basis. Compared with the frequency domain and time-frequency domain methods, the time domain statistical quantity feature is easy to implement in engineering and can better meet the real-time detection requirements under the high frequency of 400Hz of the aircraft power supply. This kind of method regards the signal as a statistical sample, and quantifies the numerical characteristics of the arc fault signal by using statistical algorithm, such as kurtosis, coefficient of variation and skewness index, etc. These statistical quantity characteristics have good recognition effect on linear load, but they are not completely applicable to the arc fault diagnosis of the aircraft nonlinear load. The common arc fault diagnosis characteristics are very similar in the case of series arc fault and crosstalk interference (including single-phase and three-phase), which easily causes misjudgment. In SAE AS5692, it is clearly specified that the disturbed line in three-phase and single-phase crosstalk interference cannot be judged as an arc fault.

[0004] In order to solve the above technical problems, the technical personnel in the art put forward an aviation alternating current series arc fault detection method based on Levene test (Cui Rihua, Tong Desuan. Aviation alternating current series arc fault detection based on Levene test [J]. Transactions of Electrical Engineering Technology, 2021, 36(14): 3034-3042.), In this method, a method based on Levene test is proposed, and the significance level P value in the test process is extracted as the identification feature, which has good identification effect on series arc fault, and no misjudgment occurs under the condition of three-phase cross talk interference, but the P value characteristics of arc fault current and the P value characteristics of the current signal of the disturbed branch when the line occurs single-phase cross talk interference have overlap, which will produce misjudgment. Therefore, the technical personnel in the art need a new method that can detect the arc fault of the aviation line. SUMMARY

[0005] The purpose of the present application is to solve the above problems, and an aviation series arc fault detection method based on improved Levene test and JS divergence is designed.

[0006] The technical scheme of the present application to achieve the above purpose is an aviation series arc fault detection method based on improved Levene test and JS divergence, which comprises the following steps:

[0007] Step one, collect several current signal periods in the aviation electrical system, each current signal period contains several current signal samples, extract a normal current signal period data sample from several current signal periods as the comparison period of other current signal periods and as the test sample one;

[0008] Step two, select the sample data of other current signal periods as test sample two, use the improved Levene test model to test test sample one and test sample two and convert the variable Z, the statistic W and the value of the significance level P, and take P value as the distinguishing standard of normal current signal and arc fault current signal;

[0009] Step three, calculate the Jensen-Shannon divergence of the data samples of other current signal periods and the data samples of the comparison period, and take the characteristic value calculated by the Jensen-Shannon divergence as the basis for judging whether the arc fault occurs;

[0010] Step four, identify the fault telephone in the aviation electrical system, set u to represent the number of arc fault periods and the initial value of u is 0, Flag1 represents the flag bit of P value exceeding the threshold value, Flag2 represents the flag bit of characteristic value exceeding the threshold value calculated by JS divergence, and t is the detection time;

[0011] If Flag1 or Flag2 = 1, the value of u is added by 1, and when the value of u is greater than 4, the circuit breaker in the aviation electrical system is tripped;

[0012] If Flag1 or Flag2 ≠ 1, return to step two and repeat the above process;

[0013] If the value of u is not greater than 4 and t is greater than 100 ms, the value of u is equal to 0, and return to step two and repeat the above process;

[0014] If the value of u is not greater than 4 and t is not greater than 100 ms, return to step two and repeat the above process.

[0015] The current signal cycle collected in the step one includes three cases of normal current cycle, arc fault current cycle and crosstalk interference current cycle.

[0016] The process of testing test sample one and test sample two in the step two by using the improved Levene test model is as follows:

[0017] The current signal data in the test sample one and the test sample two to be tested are set as Y i , wherein i = 1, 2, Y ij represents the jth data point in the ith sample, so the calculation formula is:

[0018]

[0019] In the formula, is the arithmetic mean of the ith original data, and n is an integer greater than 1;

[0020]

[0021] In the formula, k is the number of sample groups, N i is the content of the ith sample, N is the sum of all samples, is the mean value of the ith sample group, is the total mean value of all data of each group;

[0022]

[0023] In the formula, the degrees of freedom v1 and v2 are (k-1) and (N-k) respectively, and the value of n is 50 or greater than 50.

[0024] The Jensen-Shannon divergence calculation process in the step three is as follows:

[0025] Set the current signal data in the comparison period as x1(m), the current signal data in other current signal periods as x2(m), m={1, 2, 3, …, 1024}, and convert them into probability distribution models p1(m), p2(m) respectively, and the calculation formula of JS divergence is shown in formulas (5) and (6),

[0026]

[0027]

[0028] Advantages

[0029] The aviation series arc fault detection method based on improved Levene test and JS divergence has the following advantages:

[0030] 1. The method is aimed at the problem that the Levene test significance level feature has poor anti-crosstalk interference ability in identifying series arc faults. Firstly, an improved Levene test significance level feature is proposed. After comparison with the original feature, it is found that the improved Levene test feature can effectively distinguish arc faults and has the ability of anti-crosstalk interference.

[0031] 2. In order to avoid the contingency of single feature, the Jensen-Shannon divergence feature is extracted from the perspective of current signal probability distribution. After analysis, the unified threshold of the two features is given and the arc fault diagnosis process is designed. The method has good accuracy and reliability and can be used as an effective reference for the research and development of aviation arc fault circuit breakers. BRIEF DESCRIPTION OF DRAWINGS

[0032] Figure 1 is a flowchart of the aviation series arc fault detection method based on improved Levene test and JS divergence;

[0033] Figure 2 is a flowchart of the significance level P value feature extraction; DETAILED DESCRIPTION

[0034] The embodiments of the present application will be specifically described below with reference to the accompanying drawings.

[0035] Embodiment

[0036] Firstly, ten current signal periods in the aviation electrical system are collected, each current signal period contains 1024 current signal samples, and the data samples of a normal current signal period are extracted from the ten current signal periods as the comparison period of other current signal periods and are set as the test sample one.

[0037] Secondly, as Figure 2As shown, current signal samples from other periods (including normal current period, arc fault current period, and crosstalk interference current period) are compared with test sample one for testing calculation, and the significance level P-value of the two sets of current signal data samples is extracted as a feature. Let the two sets of current signal data to be tested be Y. i Where i = 1, 2, Y ij Let represent the j-th data point in the i-th sample. The formulas for calculating the transformation variable Z, the statistic W, and the significance level P are shown in (1), (2), and (3). Since the statistic W calculated for the arc fault signal, the interfered signal, and the normal signal are all greater than 10, the difference in the characteristics of the P-values ​​in the two cases is small, making misjudgment easy. Therefore, this application uses an improved algorithm that takes the square root of the transformation variable (i.e., formula (4)) of the original Levene test to obtain formula (1), that is, let

[0038] Transformed into:

[0039]

[0040] In the formula, Y i Y is the arithmetic mean of the i-th set of original data. ij Let represent the j-th data point in the i-th sample group, where n is an integer greater than 1.

[0041]

[0042] In the formula, k is the number of sample groups, and N i Let N be the sample size of the i-th group, and N be the sum of all samples. Let be the mean of the i-th sample group. This is the mean of all data in each group.

[0043]

[0044] In the formula, the degrees of freedom v1 and v2 are (k-1) and (Nk), respectively.

[0045] The interfered signal contains fewer arc fault features. By modifying the square root of the transformed variable, the difference between the interfered signal sample and the normal current signal sample can be reduced, lowering the statistical value W between the two samples to below 4, while keeping the p-value below 10. -2 Numerical values ​​appear on the order of magnitude. However, arc fault signals contain numerical characteristics such as flat shoulders and waveform distortion. Even after taking the square root of the transformed variable, the difference from the normal period remains significant, with the statistic W still greater than 10 and the p-value within 10. -2There is no obvious change in the order of magnitude. When the value of n is set to 1-100, the change of the statistic W tends to be flat when the number of roots n is greater than 50, so the number of roots n can be set to 50. The transformed variable after taking the root can improve the P value of the current signal in the single-phase cross-talk interference period, without significantly increasing the P value of the current signal in the arc fault period. Therefore, after taking the 50th root of the transformed variable, the P value of the current signal under the load in the arc fault can no longer overlap with the P value of the current signal under the single-phase cross-talk interference, thereby greatly improving the recognition accuracy of the P value for arc fault.

[0046] The characteristic value of P value under normal conditions is between 0.39-1, and after arc fault occurs, the characteristic value is between 0-0.018; before and after the occurrence of cross-talk interference, the P value changes, especially when single-phase cross-talk interference occurs, the P value characteristic has a significant decrease, but the P value is always between 0.074-1. Therefore, the P value characteristic after using the root algorithm still has high discrimination ability for normal and arc fault current signals. For the current signal of the disturbed branch in the cross-talk interference test, this characteristic can effectively distinguish it from the arc fault condition, and whether the P value is less than 0.019 can be used as the threshold for the occurrence of arc fault.

[0047] Secondly, Jensen-Shannon (JS) divergence calculation is performed, JS divergence is a variant of KL divergence, KL divergence is also called relative entropy, which is equivalent to the difference between the information entropy of two random probability distribution models, and is commonly used to describe the difference between the theoretical distribution and the actual distribution of random variables. Since the probability distribution model of the current signal will change before and after the arc fault occurs, the JS divergence can be calculated to distinguish between fault and normal conditions.

[0048] First, the 1024-point current signal per cycle is converted into a probability distribution model, and then a normal current signal cycle data is selected as the cycle to be compared, and the JS divergence of other current signal cycles is calculated with the selected current signal cycle.

[0049] The selected current signal data is set as x1(m), the other cycle current signal data is set as x2(m), m={1,2,3,…,1024}, and they are converted into probability distribution models p1(m), p2(m) respectively, and the JS divergence calculation formula is shown in formula (5) and (6),

[0050]

[0051]

[0052] In normal cases, the calculated JS divergence is between 0-0.012, and after arc fault occurs, the change range of the current signal JS divergence is 0.050-0.219. The change range of the disturbed signal JS divergence is 0.002-0.035, and does not overlap with the JS divergence of the arc fault signal, so whether the feature exceeds 0.049 can be used as the basis for judging whether the arc fault occurs.

[0053] Finally, series arc fault recognition is performed, and an arc fault recognition flowchart is shown in Figure 1 The figure shows that u represents the number of arc fault cycles, Flag1 represents a flag bit of the P value exceeding the threshold value, Flag2 represents a flag bit of the JS divergence exceeding the threshold value, and t is the detection time.

[0054] The technical solution of the present application is characterized in that the calculation method of the conversion variable in the original levene test process is improved, and the overlapping problem of the P characteristic value under the load in the single-phase crosstalk interference condition and under the arc fault condition is solved. In order to avoid the occasionality of a single characteristic quantity in identifying arc faults, the Jensen-Shannon divergence and the improved Levene test P value feature are jointly applied to the detection of aviation alternating current arc faults, a unified threshold value of the two features is given through analysis, and the flow of arc fault diagnosis is designed.

[0055] It should be noted that in this paper, relationship terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between the entities or operations. Moreover, the term "includes", "contains" or any other variant thereof is intended to cover non-exclusive inclusion, so that the process, method, article or equipment including a series of elements not only includes those elements, but also includes other elements not explicitly listed or inherent to such process, method, article or equipment. Without more limitations, the statement "includes a limited element" does not exclude the presence of additional identical elements in the process, method, article or equipment including the element.

[0056] The above technical solution only embodies the preferred technical solution of the present application, and some variations of certain parts made by the person skilled in the art also embody the principle of the present application and are within the protection scope of the present application.

Claims

1. An improved Levene's test and Jensen-Shannon divergence based aerial series arc fault detection method characterized in that, The method includes the following steps: Step 1: Collect several current signal cycles from the aviation electrical system. Each current signal cycle contains several current signal samples. The current signal samples in each current signal cycle follow a probability distribution function. Extract a normal current signal cycle from several current signal cycles as a comparison cycle for other current signal cycles. At the same time, select a current signal sample from the normal signal cycle as test sample 1. Step 2: Select a current signal sample from other current signal cycles as test sample 2. Use the improved Levene test model to test test sample 1 and test sample 2 and evaluate the transformed variable Z, statistic W and significance level P. Use the P value as the distinguishing criterion between normal current signal and arc fault current signal. Step 3: Calculate the Jensen-Shannon divergence for the other current signal periods and the comparison period, and use the eigenvalues ​​obtained from the Jensen-Shannon divergence calculation as the basis for judging whether an arc fault has occurred. Step 4: Identify faulty telephones in the aviation electrical system. Set u to represent the number of arc fault cycles and the initial value of u is 0. Flag1 represents the flag bit when the P value exceeds the threshold. Flag2 represents the flag bit when the feature value obtained by JS divergence calculation exceeds the threshold. t is the detection time. If Flag1 or Flag2 = 1, the value of u is incremented by 1. When the value of u is greater than 4, the circuit breaker in the aviation electrical system trips. If Flag1 or Flag2 ≠ 1, then return to step two and repeat the above process; If the value of u is not greater than 4 and t is greater than 100ms, then the value of u is equal to 0, and the process returns to step two and repeats the above process. If the value of u is not greater than 4 and t is not greater than 100ms, then return to step two and repeat the above process.

2. The improved Levene's test and Jensen-Shannon divergence based aerial series arc fault detection method according to claim 1, wherein, The current signal period collected in step one includes three cases: normal current period, arc fault current period, and crosstalk interference current period.

3. The improved Levene's test and Jensen-Shannon divergence based aerial series arc fault detection method according to claim 1, wherein, The process of using the improved Levene test model to test test sample one and test sample two in step two is as follows: Setting the current signal data in the test sample one and the test sample two to be Y i where i = 1, 2, Y ij represents the jth data point in the ith group of samples, so that the calculation formula is: wherein is the arithmetic mean of the i-th group of raw data, n is an integer greater than 1; where W is the statistic, k is the number of sample groups, N i is the content of the i-th sample group, and N is the sum of all samples, is the mean value within the i-th sample group, is the total mean value of all data in each group. In the formula, the degrees of freedom v1 and v2 are (k-1) and (Nk) respectively, and t is the detection time.

4. The improved Levene's test and Jensen-Shannon divergence based aerial series arc fault detection method according to claim 3, wherein, In the formula (1), the value of n is 50.

5. The improved Levene's test and Jensen-Shannon divergence based aerial series arc fault detection method according to claim 1, wherein, The Jensen-Shannon divergence calculation process in step three is as follows: Let the current signal data in the comparison period be x1(m), and the current signal data in other current signal periods be x2(m), m={1,2,3,…,1024}, and convert them into probability distribution models p1(m) and p2(m) respectively. The formulas for calculating the JS divergence are shown in formulas (5) and (6).

Images