An aviation alternating current series arc fault detection method based on coefficient of variation

Abstract

The application belongs to the technical field of arc detection, and relates to an aviation alternating current series arc fault detection method based on a variation coefficient, comprising the following steps: collecting a current signal, dividing the current signal into continuous time windows, and forming a current frequency data set; performing frequency component analysis on the current frequency data set to obtain a frequency distribution vector, calculating a standard deviation and a mean value, and obtaining a variation coefficient; forming a variation coefficient sample set based on historical current data collected under different loads, and performing double Weibull distribution fitting; selecting a numerical value of the variation coefficient corresponding to a preset quantile level as an adaptive threshold value; if the variation coefficient of a current time window is greater than the adaptive threshold value, it is determined that an alternating current series arc fault exists, and alarm information is output. The technical scheme of the application characterizes the current behavior in a form capable of reflecting frequency structure characteristics, and realizes stable identification under multiple loads and disturbance working conditions through the variation coefficient and the adaptive threshold value.

Description

Technical Field

[0001] This invention belongs to the field of arc detection technology, and specifically relates to a method for detecting AC series arc faults in aviation based on the coefficient of variation. Background Technology

[0002] In existing technologies, the detection of series arc faults in aviation AC power supply lines typically relies on features such as current waveform amplitude, harmonic content, wavelet energy coefficient, or time-domain rate of change. These features are assessed by setting fixed thresholds or employing pattern recognition algorithms to meet the requirements for fault state identification. However, existing detection methods exhibit shortcomings such as low adaptability and weak consistency in judgment under different load types, varying power supply stability states, and vibration disturbance environments. Especially in aviation power systems where frequencies are fixed and loads are variable, traditional features demonstrate large distribution dispersion and weak statistical regularity.

[0003] Detection methods based on fixed thresholds often rely on experience to set judgment boundaries. When the combination of resistive, inductive, and nonlinear devices changes, the judgment criteria corresponding to the threshold will shift significantly, resulting in weak stability of fault identification results and a high false trigger rate. On the other hand, technical solutions that use criteria such as wavelet energy ratio and harmonic energy ratio have high fluctuations in the extracted feature quantities and weak difference discrimination under high-frequency vibration, loose wire contact, and power supply fluctuation interference, and are weak in characterizing the random disturbance nature of the arc process.

[0004] Meanwhile, the technical approach for classifying and judging spectral mutations or temporal difference trends usually requires superimposing classifier models or multi-feature fusion mechanisms. The parameter training is highly dependent, and under actual aviation load switching, line aging and changes in the operating environment, the feature distribution is prone to drift, resulting in weak reliability and poor continuity of the judgment results output by the model. In long-term engineering operation, the maintenance cost is high and the model update frequency is high, which is not conducive to the stable deployment of aviation power equipment.

[0005] It is evident that existing technologies often suffer from problems such as weak stability in fault characteristic representation, high dependence on judgment criteria, and low consistency in identification under multiple loads and disturbances. These are the shortcomings of existing technologies.

[0006] In view of this, it is very necessary to provide an aviation AC series arc fault detection method based on the coefficient of variation to solve the above-mentioned defects in the prior art. Summary of the Invention

[0007] The purpose of this invention is to address the shortcomings of existing technologies, such as weak stability of fault feature expression, high dependence on judgment criteria, and low consistency in identification under multiple loads and disturbances. This invention provides a method for detecting aviation AC series arc faults based on the coefficient of variation, thereby solving the aforementioned technical problems.

[0008] To achieve the above objectives, the present invention provides the following technical solution:

[0009] This application provides a method for detecting AC series arc faults in aviation based on the coefficient of variation, including:

[0010] The current signal of the aviation AC power supply line is collected, and the current signal is divided into continuous time windows according to a fixed duration corresponding to the basic power supply frequency. A current frequency dataset is formed in each time window to characterize the current frequency distribution in that time window.

[0011] Within each time window, frequency component analysis is performed on the current frequency dataset to obtain a frequency distribution vector representing the probability distribution of each frequency component. The mean and standard deviation of the frequency distribution vector are calculated to obtain the coefficient of variation corresponding to that time window. The coefficient of variation is the ratio of the standard deviation to the mean.

[0012] Based on historical current signals collected under different aviation electrical load conditions, the historical current signals are divided into windows, frequency components are analyzed and coefficients of variation are calculated to obtain the coefficients of variation corresponding to multiple historical time windows, and a coefficient of variation sample set is formed. The coefficient of variation sample set is fitted with a double Weibull distribution to obtain the cumulative distribution function used to characterize the random distribution characteristics of the coefficient of variation.

[0013] Based on the cumulative distribution function, the coefficient of variation value corresponding to the preset quantile level is selected as the adaptive threshold. The coefficient of variation obtained in the current time window is compared with the adaptive threshold. When the coefficient of variation is greater than the adaptive threshold, it is determined that there is an aviation AC series arc fault in the current time window, and the fault alarm information is output.

[0014] As a preferred embodiment, the step of dividing the current signal into continuous time windows of a fixed duration corresponding to the fundamental frequency of the power supply includes: establishing a time reference based on the fundamental period of the power supply signal, and periodically synchronizing the current signal with the time reference to form continuous time windows, wherein each time window is synchronized with the fundamental frequency of the power supply.

[0015] Preferably, the step of performing frequency component analysis on the current frequency dataset within each time window to obtain a frequency distribution vector representing the probability distribution of each frequency component includes: performing frequency component analysis on the current frequency dataset within each time window, and dividing the spectral amplitude according to continuous frequency intervals to form a corresponding frequency distribution vector. The frequency component analysis includes spectral estimation.

[0016] Preferably, the frequency range division is set according to the distribution characteristics of harmonic sequences in the aviation AC power supply line, and the frequency range division method remains unchanged under different load conditions.

[0017] Preferably, the step of fitting the sample set of coefficients of variation with a double Weibull distribution includes:

[0018] The statistical consistency criterion based on the sample sequence is used to remove outliers from the coefficient of variation samples and fit them to a double Weibull distribution.

[0019] The parameters of the double Weibull distribution are solved using the maximum likelihood estimation method, and the convergence of the parameters is checked based on the deviation relationship between the empirical cumulative distribution function and the fitted cumulative distribution function. The parameters include shape parameters, location parameters and scale parameters.

[0020] Preferably, the preset quantile level is determined based on the tail characteristics of the cumulative distribution function of the coefficient of variation samples, and the adaptive threshold corresponds to the upper probability domain of the coefficient of variation sample distribution.

[0021] As a preferred embodiment, the step of selecting the coefficient of variation value corresponding to the preset quantile level as the adaptive threshold based on the cumulative distribution function includes: establishing a confidence interval based on the cumulative distribution function, and using the coefficient of variation value corresponding to the preset quantile level in the confidence interval as the adaptive threshold, wherein the confidence interval is used to describe the range of values ​​of the coefficient of variation.

[0022] Preferably, the step of comparing the coefficient of variation obtained in the current time window with the adaptive threshold includes: performing comparisons and recording the comparison results in adjacent time windows respectively, and outputting the judgment result when the comparison results are consistent at least once.

[0023] As a preferred method, the comparison of the coefficient of variation and the adaptive threshold is paused when the aviation AC power supply line is in the commutation, startup or steady-state transition range, and the comparison is resumed after the power supply status stabilizes.

[0024] Preferably, the method further includes: when a change in the load type of the aviation AC power supply line is detected, re-performing the double Weibull distribution fitting based on the new coefficient of variation sample and updating the adaptive threshold.

[0025] As can be seen from the above technical solutions, the present invention has the following advantages:

[0026] This application provides a method for detecting series arc faults in aviation AC power supply lines based on the coefficient of variation. The method involves periodically dividing the current signal, extracting the frequency structure of the current data within a time window, and constructing a statistical expression based on the distribution relationship of each frequency component. The ratio of the standard deviation to the mean of the frequency distribution structure is used to construct an index characterizing the strength of random disturbances. A probability distribution model is built based on a sample set of this index formed under historical operating conditions. Values ​​with specific quantile meanings are selected from this probability distribution model as judgment criteria. Finally, the index is recalculated under the current operating condition and compared with the judgment criteria, ensuring a one-to-one correspondence between the judgment behavior and the statistical framework, avoiding discontinuities in results caused by changes in load or disturbance conditions. This method maintains high stability of expression, strong independence of judgment sources, and high consistency in fault identification under different operating conditions, improving the continuous reliability of series arc fault detection results in aviation AC power supply lines under vibration environments, multiple load changes, and fluctuating operating conditions.

[0027] Furthermore, the design principle of this invention is reliable, the structure is simple, and it has a very wide range of application prospects.

[0028] Therefore, it is evident that the present invention has outstanding substantive features and significant progress compared with the prior art, and the beneficial effects of its implementation are also obvious. Attached Figure Description

[0029] To more clearly illustrate the technical solution of the present invention, the accompanying drawings used in the description will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0030] Figure 1 This is a flowchart of an aviation AC series arc fault detection method based on the coefficient of variation provided by the present invention.

[0031] Figure 2 This is a frequency distribution histogram of the fault arc current signal of a certain load in an aviation AC series arc fault detection method based on the coefficient of variation provided by the present invention.

[0032] Figure 3 This is a schematic diagram of the coefficient of variation of a certain load within a certain time window in the aviation AC series arc fault detection method based on the coefficient of variation provided by the present invention. Detailed Implementation

[0033] Various embodiments of this disclosure are described more fully below with reference to the accompanying drawings. This disclosure may have various embodiments, and adjustments and changes may be made therein. However, it should be understood that there is no intention to limit the various embodiments of this disclosure to the specific embodiments disclosed herein, but rather this disclosure should be understood to cover all adjustments, equivalents, and / or alternatives falling within the spirit and scope of the various embodiments of this disclosure.

[0034] In the following, the terms “comprising” or “may include”, which may be used in various embodiments of this disclosure, indicate the presence of the disclosed functions, operations, or elements, and do not limit the addition of one or more functions, operations, or elements. Furthermore, as used in various embodiments of this disclosure, the terms “comprising,” “having,” and their cognates are intended only to indicate a particular feature, number, step, operation, element, component, or combination of the foregoing, and should not be construed as primarily excluding the presence of one or more other features, numbers, steps, operations, elements, components, or combinations of the foregoing, or the possibility of adding one or more combinations of the foregoing.

[0035] It should be noted that, in various embodiments of this disclosure, the expression "or" or "at least one of A and / or B" includes any combination or all combinations of the words listed simultaneously. For example, the expression "A or B" or "at least one of A and / or B" may include A, may include B, or may include both A and B.

[0036] To facilitate a clear description of the technical solutions in the embodiments of this application, some terms and technologies involved in the embodiments of this application will be briefly introduced below:

[0037] 1. Aviation AC power supply line

[0038] In the field of aviation electrical systems technology, aviation AC power supply lines generally refer to power supply circuits that provide AC power from aviation power units to airborne equipment. These circuits include power output terminals, feeders, cables, connectors, switching equipment, and electrical interfaces connecting to various aviation loads. Compared to conventional land-based AC power distribution lines, aviation AC power supply lines face stricter constraints in terms of rated voltage, current frequency, insulation level, electromagnetic compatibility, and wiring space. They typically require long-term stable operation in environments with high vibration, strong electromagnetic interference, and confined spaces.

[0039] 2. Aircraft AC series arc fault

[0040] Alternating Current Series Arc Fault (ACSAF) in aviation AC power supply lines typically refers to the AC arc discharge phenomenon that occurs at a gap in the current path when a conductor strand breakage, poor terminal contact, or loose connector is introduced. Electrically, this type of fault exhibits a strongly nonlinear conduction path resistance that varies intermittently over time, leading to distortion of the line current waveform, increased harmonic components, and enhanced transient fluctuations. It can also cause localized heating, insulation aging, and electromagnetic interference to surrounding avionics equipment. Compared to parallel arc faults, series arc faults often do not immediately trigger a large current short circuit, exhibiting greater concealment, weaker early characteristics, and a gradual impact on power supply reliability.

[0041] 3. Aviation electrical load conditions

[0042] Aviation electrical load conditions are typically used to describe the combined states of various loads supplied to aviation AC power lines during different operating phases. This includes load types (such as avionics, lighting loads, and motor-driven loads), load connection and disconnection status, load power levels, and the operating modes of the load's internal power electronic interfaces. As the aircraft transitions between different operational phases, such as ground power supply, engine start-up, takeoff, cruise, and landing, the aviation electrical load conditions change, resulting in variations in the current amplitude, waveform shape, and harmonic distribution characteristics of the power supply lines.

[0043] 4. Coefficient of variation

[0044] The coefficient of variation (CV) is commonly used to measure the relative dispersion of a random variable or sample sequence. It is the dimensionless ratio between the standard deviation and the mean.

[0045] For a set of samples, the mean is denoted as... The standard deviation is denoted as The corresponding coefficient of variation can be expressed as:

[0046]

[0047] in, Represents the coefficient of variation. This represents the sample mean. The coefficient of variation represents the standard deviation of a sample. Compared to using the standard deviation directly, the coefficient of variation is normalized by introducing the mean, which provides a more comparable characterization of the degree of fluctuation across different units or different mean levels.

[0048] 5. Weibull distribution and double Weibull distribution

[0049] The Weibull distribution is a type of continuous probability distribution widely used in reliability analysis, lifetime statistics, and extreme value modeling. It typically describes the distribution shape of random variables through shape parameters, scale parameters, and optional location parameters. The Weibull distribution can fit stochastic processes such as failure probabilities or load intensity changes over time quite well. A bi-Weibull distribution usually refers to a mixed distribution model formed by linearly superimposing two Weibull distributions with certain weights. It can be used to describe random phenomena generated by the superposition of two different statistical populations, or to fit sample distributions with bimodal or heavy-tailed characteristics. Compared to a single Weibull distribution, the bi-Weibull distribution offers greater flexibility in characterizing complex random distributions.

[0050] To address the problems of weak stability of fault feature representation, high dependence on judgment criteria, and low consistency of identification under multiple load and disturbance conditions in traditional AC series arc fault detection methods based on fixed thresholds or pattern recognition algorithms, this application discloses an AC series arc fault detection method based on the coefficient of variation. By introducing the coefficient of variation and an adaptive threshold that can be dynamically adjusted under different electrical loads, and by utilizing the distribution function and quantiles of the features instead of binarizing the feature image when setting the adaptive threshold, the method ensures statistical consistency of the judgment criteria and avoids inconsistencies caused by threshold dependence on empirical settings. This effectively improves the accuracy of arc fault detection and meets the requirement of maintaining stable identification results under multiple load and disturbance conditions.

[0051] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0052] like Figure 1 As shown in the figure, this embodiment provides a method for detecting aviation AC series arc faults based on the coefficient of variation, including:

[0053] Step S1: Collect the current signal of the aviation AC power supply line, divide the current signal into continuous time windows according to a fixed duration corresponding to the basic power supply frequency, and form a current frequency dataset to characterize the current frequency distribution within each time window.

[0054] Step S2: Within each time window, perform frequency component analysis on the current frequency dataset to obtain the frequency distribution vector representing the probability distribution of each frequency component, and calculate the mean and standard deviation of the frequency distribution vector to obtain the coefficient of variation corresponding to that time window. The coefficient of variation is the ratio of the standard deviation to the mean.

[0055] Step S3: Based on the historical current signals collected under different aviation electrical load conditions, the historical current signals are divided into windows, frequency components are analyzed and the coefficient of variation is calculated to obtain the coefficient of variation corresponding to multiple historical time windows, and a coefficient of variation sample set is formed. The coefficient of variation sample set is fitted with a double Weibull distribution to obtain the cumulative distribution function used to characterize the random distribution characteristics of the coefficient of variation.

[0056] Step S4: Based on the cumulative distribution function, select the coefficient of variation value corresponding to the preset quantile level as the adaptive threshold, compare the coefficient of variation obtained in the current time window with the adaptive threshold, and when the coefficient of variation is greater than the adaptive threshold, determine that there is an aviation AC series arc fault in the current time window and output the fault alarm information.

[0057] By adopting the above technical solution, the current signal is formed into a stable statistical analysis unit under a time reference corresponding to the basic power supply frequency, so that the current behavior can be characterized in a form that reflects the frequency structure characteristics. The coefficient of variation value in the cumulative distribution function corresponding to the preset quantile level is used as an adaptive threshold to describe the random disturbance introduced by the series arc in a probabilistic sense. At the same time, the adaptive threshold can be adaptively adjusted under different electrical loads, so that the source of the judgment basis has statistical consistency, avoiding the inconsistency of the basis caused by the threshold relying on experience setting, and meeting the requirement of maintaining stable recognition results under multiple loads and disturbance conditions.

[0058] Hereinafter, steps S1 to S4 will be specifically described according to embodiments of this application.

[0059] In step S1, the core task is to acquire the current signal on the aviation AC power supply line with high time resolution. With the help of a time reference that is strictly synchronized with the basic power supply frequency, the long-term continuous sampling data is divided into continuous time windows of finite length. In each time window, a current frequency dataset that can characterize the current frequency distribution within that time window is constructed, thereby providing a unified and comparable feature input basis for subsequent aviation AC series arc fault detection based on the degree of frequency distribution dispersion.

[0060] In this embodiment of the application, in order to ensure the comparability of statistics between different load conditions and different sampling periods, this step uses a window division method locked with the fundamental frequency period to ensure that each time window covers the same power supply phase range and avoids the impact of phase drift on the statistical results.

[0061] Specifically, aviation AC power supply lines can exist in single-phase or three-phase aviation AC power supply configurations, and the basic frequency of the power supply signal is denoted as... For example, an aircraft's onboard 400Hz AC power supply, or a 50Hz power frequency supply used in ground simulation tests. The detection method in this application acquires a continuous sampling sequence of current changes over time on the monitored aircraft AC power supply line using a current sensor, denoted as a discrete-time sequence. ,in For discrete sampling index, Indicates the first The instantaneous current value at each sampling moment. The sampling frequency is denoted as . The sampling time interval is The total duration of continuous sampling is recorded as Then it can be within the time interval The length obtained inside is The sampling sequence, where, .

[0062] Based on this, the current signal of the aviation AC power supply line is collected, and the current signal is divided into continuous time windows according to a fixed duration corresponding to the basic power supply frequency, so that each time window corresponds to several complete or fixed proportion fundamental frequency cycles, avoiding statistical deviations caused by different sampling start and end phases.

[0063] In some embodiments of this application, in order to keep the time window strictly synchronized with the fundamental frequency of the power supply, a time reference can be established based on the fundamental period of the power supply signal, wherein the fundamental period of the power supply signal is denoted as... During the acquisition process, reference phase signals of the supply voltage or current can be acquired simultaneously. By using methods such as zero-crossing detection, phase-locked loop (PLL), or digital PLL, the starting times of adjacent fundamental cycles can be marked on the time axis. ,in Based on this time reference, the current signal sampling points can be synchronously segmented and truncated according to the alignment with the fundamental frequency period, forming continuous time windows. The duration of each time window is defined as follows: and order ,in, It is a positive integer or rational number used to represent the number of fundamental cycles or the proportion of cycles covered by a single time window.

[0064] Given observation duration Under the condition, the total number of windows It can be represented as:

[0065]

[0066] in, Indicates the continuous observation time. This indicates the length of time contained in each time window. This represents the number of time windows available within the observation period. Using the above relationship, the continuously acquired current signal can be precisely divided into multiple consecutive time windows of equal length, with each time window synchronized with the fundamental power supply frequency, thus avoiding statistical distortion caused by the drift of the window's start and end points over time.

[0067] In some embodiments of this application, the set of current sampling points within each time window can be denoted as... ,in, For time window indexing, For the first Index of the starting sampling point of each time window The number of sampling points included in each time window.

[0068] Furthermore, in order to extract a "frequency distribution" that characterizes the statistical properties of the current amplitude from each time window, this embodiment does not directly perform simple mean or peak value calculations on the time-domain waveform, but instead statistically analyzes the frequency of occurrence of each amplitude interval within the time window. For this purpose, several non-overlapping amplitude intervals can be pre-divided on the current amplitude axis, with the boundaries of the amplitude intervals denoted as... ,in and These represent the minimum and maximum boundaries of the current amplitude, respectively. This represents the number of amplitude intervals.

[0069] In some embodiments of this application, an equal-interval amplitude division method can be adopted, that is, the width of adjacent intervals is set... , and according to Construct the amplitude interval, where, For the first The first time window and the first For each amplitude interval, the count of samples falling into that interval within a time window is defined as:

[0070]

[0071] in" " indicates the number of elements in the set.

[0072] In other embodiments of this application, the count can be normalized to probability density or relative frequency, resulting in:

[0073]

[0074] in, Indicates the first The current amplitude falls within the first time window. The relative frequency of each interval is a dimensionless probabilistic quantity. Arranging the relative frequencies of each interval within the same time window in order yields the frequency distribution vector for that time window. , where superscript This represents the vector transpose. Through the above construction, a current frequency dataset characterizing the current frequency distribution within each time window can be formed. This dataset can be further structured into a sequence concatenated window by window along the time axis. .

[0075] In some exemplary embodiments, to enhance the separability of arc fault characteristics in frequency distribution, the division of amplitude or frequency ranges can be engineered by combining the load characteristics and harmonic sequence distribution in the aviation AC power supply line. For example, for a ground simulation test powered by a 50Hz power frequency, one cycle of the current signal can be divided into 5 time windows, with 6250 sampling points per cycle and approximately 1250 sampling points per time window. The current amplitude of each time window can then be divided into... Each interval is represented by a 50-dimensional vector. This characterizes the frequency distribution of the window. For example, in a certain test configuration, this could be applied to four different loads (e.g., 25...). Linear load, series solenoid valve load, with 47 Current signals from multiple power frequency cycles (such as those from nonlinear loads like capacitors) are collected. The time window division and frequency distribution vector construction process described above is performed on each cycle to obtain a frequency dataset across the load. Practice shows that when a series arc fault occurs, the frequency of the current amplitude range near zero in certain time windows increases significantly, causing the frequency distribution vector of the corresponding window to deviate significantly from the normal operating condition in certain dimensions, thus creating conditions for subsequent fault identification using the dispersion index.

[0076] It should be further explained that, compared to traditional methods that extract a single statistic only over a complete power frequency cycle or a longer time period, this step refines the long-term current sequence into multiple windows with uniform duration by dividing it into time windows synchronized with the basic power supply frequency. Within each window, a current frequency dataset is explicitly constructed, enabling the generation of a series of data under different load types and operating scenarios. The vectors maintain consistency in dimension and are continuous on the time axis, thus enabling them to more sensitively capture the local statistical feature changes introduced by series arc faults, thereby improving the separability and applicability of subsequent fault detection algorithms.

[0077] Thus far, step S1, by introducing a time reference synchronized with the basic power supply frequency, continuously collects the current signal of the aviation AC power supply line and divides it into continuous time windows of fixed duration. Within each time window, a current frequency dataset composed of relative frequencies is constructed, forming an input data representation with a unified structure in both the time dimension and the statistical feature space. This provides a stable and reproducible data foundation for subsequent calculation of frequency distribution dispersion, extraction of coefficient of variation features, and construction of adaptive thresholds under multiple load conditions.

[0078] In step S2, the core task is to perform frequency component analysis on the current frequency dataset corresponding to each time window obtained in step S1, and construct a frequency distribution vector in the frequency domain that can represent the probability distribution of each frequency component, such as... Figure 2 As shown, the mean and standard deviation of the frequency distribution vector are calculated based on this, and the coefficient of variation corresponding to this time window is further obtained, where the coefficient of variation is defined as the ratio of the standard deviation to the mean. Through this process, the dispersion of the current frequency distribution can be characterized by a dimensionless statistic within each time window, providing a characteristic basis for subsequent fault sensitivity discrimination based on the coefficient of variation sequence.

[0079] Specifically, for the time window sequence already constructed in step S1, the first... The current sampling points within a time window are considered to have a length of [missing information]. The discrete time series, denoted as ,in To analyze the energy distribution of the current within this time window from the perspective of frequency components, a frequency component analysis of the discrete sequence can be performed using a spectral estimation-based approach.

[0080] In some embodiments of this application, the spectral amplitude can be divided into continuous frequency intervals to form a corresponding frequency distribution vector.

[0081] Specifically, the discrete Fourier transform can be used for spectrum estimation, and the window function is denoted as . Then the first A time window at discrete frequency index The complex spectrum at a certain point can be expressed as:

[0082]

[0083] in, The imaginary unit, , The number of frequency points selected.

[0084] The corresponding amplitude spectrum can be written as:

[0085]

[0086] in, Indicates the first Within the time window, corresponding to the first The amplitude at each discrete frequency point is used to reflect the intensity of the current component at that frequency. In this way, frequency component analysis can be performed on the current frequency dataset within each time window, and the amplitude sequence obtained from the spectrum estimation provides the basic data for subsequent frequency distribution statistics.

[0087] Furthermore, to construct a frequency distribution vector with probabilistic meaning from the spectral amplitudes, the frequency axis can be divided into continuous frequency intervals, and the spectral amplitudes within each frequency interval can be aggregated and statistically analyzed. Let the boundary of the frequency interval be denoted as... ,in and These are the lower and upper limits of the analysis bandwidth, respectively. This represents the number of frequency intervals. (The first...) Each frequency range can be represented as , For the first A time window can be used to define frequency ranges. The summation or superposition of the amplitudes at several discrete frequency points within a given range yields the comprehensive amplitude for that range.

[0088]

[0089] in, Indicates the first The actual frequencies corresponding to discrete frequency points. To eliminate the difference in overall amplitude scale between different time windows, the comprehensive amplitude of each frequency interval can be normalized and transformed into a dimensionless relative probability. Specifically, the ____ can be defined as follows: The frequency distribution vector components of each time window are:

[0090]

[0091] here Indicates the first Within the time window, the frequency falls into the [number]th [period]. The relative proportion of each frequency range satisfies ,and Through the above construction, a frequency distribution vector representing the probability distribution of each frequency component can be obtained within each time window. The frequency distribution vector of each time window can be represented as:

[0092]

[0093] Where, vector Each component is the probability distribution of each frequency component.

[0094] In some embodiments of this application, the division of frequency ranges is not arbitrarily chosen, but rather set with full consideration of the distribution characteristics of harmonic sequences in aviation AC power supply lines. Specifically, the current spectrum of aviation AC power supply lines under normal operating conditions often exhibits a significant fundamental component and several harmonic components related to the load type. For example, rectifier loads will form characteristic harmonic peaks near the 5th, 7th, and 11th orders, while operating conditions with motor loads will form strong narrowband concentrated energy near the fundamental frequency. By using these typical harmonic locations as interval boundaries or interval centers when dividing the frequency ranges, the frequency distribution vector can be made more sensitive to changes in the harmonic structure.

[0095] Meanwhile, to maintain comparability between different load conditions, in this embodiment, when designing the frequency range division method, a set of fixed interval boundaries is determined on the frequency axis by statistically analyzing the harmonic sequence based on typical load combinations that may occur on aviation AC power supply lines. Furthermore, the frequency range division method remains unchanged under different load conditions. In this way, the coefficient of variation retains the same physical meaning across load conditions and has good horizontal comparability.

[0096] In some embodiments of this application, a discrete random variable over a frequency interval can be constructed to define the mean and standard deviation of the frequency distribution vector from a statistical perspective. As some examples, the first... The frequency interval index within a time window is considered as a discrete random variable. Its value is the center frequency of the frequency range. The probability is .at this time, The mathematical expectation can be written as:

[0097]

[0098] in, Indicates the first The average frequency position of the current frequency distribution within a time window.

[0099] Accordingly, standard deviation can be defined as:

[0100]

[0101] in, Indicates the first The degree of dispersion of the current frequency distribution relative to the average frequency within a time window.

[0102] The aforementioned forms of expectation and standard deviation can be understood directly from the perspective of random variables, or they can be viewed as statistics obtained by weighted summation of the components of the frequency distribution vector. It should be further noted that in some implementations, the probability components can also be directly... The degree of dispersion itself is measured, Treating the frequency distribution as a set of non-negative numerical sequences, we calculate its mean and standard deviation to obtain a statistical measure that characterizes the uniformity of frequency energy distribution across frequency intervals. Regardless of the implementation method used, this application measures the statistical characteristics of frequency components within a time window by calculating the mean and standard deviation of the frequency distribution vector.

[0103] In some embodiments, to eliminate the influence of overall scale differences in frequency distribution under different time windows and load conditions, a dimensionless coefficient of variation can be constructed based on the aforementioned mean and standard deviation. For example, the first... The coefficient of variation within each time window is denoted as . Defined as:

[0104]

[0105] in, and These are the aforementioned mean and standard deviation, respectively. From this definition, it can be seen that when the frequency distribution shows significant concentration in certain intervals and substantial attenuation in other intervals, the standard deviation... Relative to the average It will increase significantly, causing the coefficient of variation to... The coefficient of variation tends to be large; conversely, when the frequency energy distribution is relatively smooth across intervals, the ratio of the standard deviation to the mean is close to a small constant, and the coefficient of variation exhibits a smaller value. Through this construction, the dispersion of frequency distribution within different time windows can be uniformly mapped onto the single scalar of the coefficient of variation.

[0106] In other embodiments of this application, the above definition can be quantitatively illustrated using actual aviation AC series arc fault test data. For example, in a scenario powered by a 400Hz fundamental frequency, a sampling frequency of 10kHz can be used to divide one fundamental frequency cycle into five equal-length time windows. Spectral estimation is performed on each time window, and the 0–5kHz frequency range is divided into several consecutive frequency intervals, some of which are specifically used to cover the frequency band near the fundamental and lower harmonics. Under normal load conditions, the fundamental and lower harmonic frequency bands... The proportion is relatively high, corresponding to the average frequency The distribution is located near the weighting center of the fundamental and lower harmonic frequencies, with a standard deviation of [missing information]. At a low level, therefore the coefficient of variation The values ​​are concentrated in a narrow range; when an aviation AC series arc fault occurs, due to the intermittent conduction and re-breakdown process introduced by the arc, the high-frequency components and non-integer multiple harmonic components related to the arc increase significantly in the spectrum, and some high-frequency sections... A sharp increase caused the standard deviation to rise. Much greater than normal operating conditions, while the average value... The changes are relatively limited, thus leading to The fault time window is significantly larger than the normal time window. This difference gives the coefficient of variation a high sensitivity in identifying AC series arc faults in aviation.

[0107] It should be noted that, compared with traditional fault detection methods that are characterized by changes in single harmonic amplitude, total harmonic distortion, or effective current value, this step constructs a frequency distribution vector in the frequency domain and defines the coefficient of variation based on the mean and standard deviation of the frequency distribution vector. This allows the relative energy distribution changes of the spectrum in each frequency interval to be reflected as a dimensionless statistical quantity. This retains the comprehensive perception of high-frequency disturbances caused by the arc and the phenomenon of spectrum flattening near the zero point, while reducing the influence of differences in overall amplitude and harmonic amplitude under different load conditions on the characteristic quantity. Thus, it maintains good separability and robustness for aviation AC series arc faults over a wide range of load variations.

[0108] Thus, step S2 analyzes the frequency components of the current frequency dataset within each time window using spectral estimation, and divides continuous frequency intervals on the frequency axis according to the distribution characteristics of the harmonic sequence of the aviation AC power supply line, constructing a frequency distribution vector representing the probability distribution of each frequency component. Then, based on the frequency distribution vector, the mean and standard deviation are calculated, and the coefficient of variation is defined as the ratio of the standard deviation to the mean. This process maps the dispersion of the current frequency distribution to a dimensionless scalar, ensuring that each time window corresponds to a coefficient of variation that reflects the sensitivity characteristics of aviation AC series arc faults. This lays a clear and unified characteristic foundation for establishing a statistical model of the coefficient of variation samples under multiple load conditions.

[0109] In step S3, the core task is to repeat the aforementioned sampling, time window division, and frequency component analysis process based on historical current signals collected under different aviation electrical load conditions, and calculate the coefficient of variation within each historical time window, such as... Figure 3As shown, the coefficients of variation corresponding to multiple historical time windows are obtained, and these coefficients of variation are summarized into a coefficient of variation sample set. On this basis, a statistical modeling method adapted to multiple load scenarios is used to fit the coefficient of variation sample set with a double Weibull distribution to obtain a cumulative distribution function used to characterize the random distribution characteristics of the coefficient of variation. This "compresses" a large number of historical coefficient of variation samples across loads and operating conditions into a continuous probability distribution curve, and uses this curve to characterize the overall random behavior of the coefficient of variation under normal operating conditions, providing a statistical basis for identifying abnormal coefficients of variation that deviate from this distribution in unknown load conditions.

[0110] In this embodiment, to form a representative sample set of coefficients of variation, historical current signals need to be collected under various load conditions representing typical operating states of aviation AC power supply lines. Multiple groups of aviation electrical loads, including linear and nonlinear loads, can be selected, such as resistive loads, inductive loads, loads with capacitive branches, and composite loads with nonlinear actuators such as solenoid valves. Under each load condition, the current of the aviation AC power supply line is continuously sampled using the same sampling equipment as described above, ensuring that the sampling frequency, observation duration, and time window length are consistent with those in the online detection phase. For example, referring to a vibration test scenario, 100 power frequency cycles, including normal and fault currents, are collected for each of the four loads, with 6250 sampling points per cycle. Each power frequency cycle is divided into 5 time windows, with each time window containing approximately 1250 sampling points. Then, 10 representative cycles are extracted from each load, and a frequency distribution vector is constructed for each time window according to the aforementioned steps, and the corresponding coefficient of variation is calculated. In this way, each load can obtain the coefficient of variation for 50 time windows, and a total of 200 coefficient of variation samples corresponding to the four loads can be obtained, forming a historical coefficient of variation sample set across loads.

[0111] After obtaining the historical current signal, the aforementioned time window division and frequency distribution calculation process can be repeated for the current sequence under each load condition. For the first... Each historical time window can be used to obtain its frequency distribution vector through frequency domain analysis. And calculate the average value of the vector. with standard deviation Then, the coefficient of variation corresponding to this time window is constructed. Under multi-load conditions, performing this process over all historical time windows yields a set of scalar samples. ,in, This represents the total number of historical time windows. From a statistical perspective, this set of coefficients of variation samples can be regarded as a sequence of observations generated by random variables under the background of "normal operation + electric arc disturbance". Its distribution characteristics are related to the load type and are also affected by the probability and intensity of electric arc occurrence.

[0112] In some embodiments of this application, to ensure the stability and reliability of the statistical modeling results for the coefficient of variation sample set, sample quality can be preprocessed. For example, outlier removal can be performed on the coefficient of variation samples based on the statistical consistency criteria of the sample sequences. For instance, all samples can be preprocessed first. Calculate the sample mean with sample standard deviation Then, a cutoff threshold of several times the standard deviation is set to mark and remove extreme samples that significantly deviate from the distribution of the vast majority of samples, possibly due to measurement anomalies or transient strong interference. Alternatively, methods such as box plot quartile rules and empirical quantile pruning can be used to ensure that the coefficient of variation data used for fitting generally conforms to the assumption of a unimodal, bimodal, or finite mixture distribution, thereby improving the stability of subsequent distribution fitting. Through the above processing, a coefficient of variation sample sequence after excluding outliers can be obtained, providing a clean data foundation for further performance of biweibull distribution fitting.

[0113] To find the probabilistic model that best describes the random behavior of the coefficient of variation in a statistical sense, a double Weibull distribution can be fitted to the sample set of the coefficient of variation. As some examples, the coefficient of variation can be considered as taking values ​​of Let the random variable be an integer, and assume that its probability density function (PDF) follows a double Weibull distribution. The double Weibull distribution is a widely used family of distributions in reliability engineering and life analysis, suitable for describing failure data or characteristic data with "bathtub curve" characteristics. Its probability density function can be expressed as:

[0114]

[0115] in, For the coefficient of variation sample, For scale parameters, For shape parameters, This is a position parameter used to characterize the offset distributed on the number line.

[0116] The corresponding cumulative distribution function (CDF) can be written in piecewise form:

[0117]

[0118] in, Define any given value of the coefficient of variation When the cumulative probability of the coefficient of variation not exceeding a given value is considered, the cumulative distribution function is used to characterize the random distribution properties of the coefficient of variation. Through appropriate selection... The numerical values ​​of the double Weibull distribution can exhibit complex shapes such as unimodal, bimodal, or wide-tailed distributions, making it suitable for fitting coefficient of variation samples after superimposing different load conditions and failure probabilities.

[0119] In some embodiments of this application, in order to obtain the optimal parameter combination of the double Weibull distribution from the coefficient of variation sample set, the maximum likelihood estimation (MLE) method can be used to solve for the parameters of the double Weibull distribution. Let the parameter vector to be estimated be... Let the aforementioned sample set of coefficients of variation after outlier removal be denoted as Then in the parameters The joint likelihood function of these samples can be expressed as:

[0120]

[0121] in, For the above double Weibull probability density function in The goal of the maximum likelihood estimation algorithm is to find a set of parameters. , making the likelihood function To obtain the maximum value. In practice, the numerical solution for the parameters is usually obtained by maximizing the log-likelihood function:

[0122]

[0123] For example, numerical optimization methods such as gradient ascent, quasi-Newton's method, or expectation maximization can be used to iteratively update the parameters until the change in the log-likelihood value is lower than a preset convergence threshold. Through this process, the fitted distribution can be made to statistically closely resemble the distribution of the actual coefficient of variation sample.

[0124] In some exemplary embodiments, to evaluate the goodness of fit of the double Weibull distribution to the coefficient of variation samples and to verify the convergence and rationality of the parameter solution process, the convergence of the above parameters can be checked by combining the deviation relationship between the Empirical Cumulative Distribution Function (ECDF) and the fitted cumulative distribution function. Specifically, an empirical cumulative distribution function can be constructed based on the coefficient of variation samples. The function takes the value at each sample point equal to the proportion of observations not greater than that sample value. Meanwhile, the parameters obtained from the maximum likelihood estimation are substituted into the Weibull cumulative distribution function. The fitted cumulative distribution function curve is obtained. Then, the differences between the two cumulative distribution curves can be quantitatively evaluated using indicators such as the Kolmogorov-Smirnov (KS) test statistic, the sum of squared errors index, and the Akaike information criterion (AIC). For example, the sum of squared errors measures the sum of the longitudinal differences between the two curves; the AIC comprehensively considers the model's fitting accuracy and the number of parameters, with a smaller statistic indicating a better model; the KS test measures the consistency between the sample distribution and the fitted distribution through the maximum longitudinal distance. By comparing the above indicators for different candidate distributions (such as a single Weibull distribution, a log-normal distribution, etc.), it can be verified that the double Weibull distribution has the best fitting effect on the current coefficient of variation sample set, proving the statistical rationality of the selected model and parameters. This process can also be seen as a concrete manifestation of the convergence check of parameters based on the deviation relationship between the empirical cumulative distribution function and the fitted cumulative distribution function.

[0125] For example, in a scenario where statistical modeling is performed on 200 samples of coefficients of variation for four different loads, the shape parameter can be obtained by maximum likelihood estimation as approximately The position parameters are approximately The scale parameter is approximately The optimal solution is found, and the probability density function of the coefficient of variation can be expressed as:

[0126]

[0127] The corresponding cumulative distribution function can be written as:

[0128]

[0129] In some embodiments of this application, when a change in the load type of the aviation AC power supply line is detected, or a new aviation electrical load configuration is introduced, the above processing procedure can be repeated based on historical current signals collected under the new load conditions. This involves re-dividing the time window, performing frequency component analysis and calculating the coefficient of variation to form a new coefficient of variation sample set. The double Weibull distribution fitting process is then re-executed on this new sample set, and subsequent adaptive thresholds are updated. By performing parameter estimation and comparing empirical and fitted cumulative distributions on the new sample set, double Weibull distribution parameters and cumulative distribution functions suitable for the new load combination can be obtained. This allows the method to continuously update the statistical model as the load type and electrical system configuration evolve in actual aviation applications.

[0130] Thus, step S3 involves collecting historical current signals under different aviation electrical load conditions and processing them according to the aforementioned sampling, windowing, and frequency component analysis process. This yields the coefficients of variation corresponding to multiple historical time windows, forming a coefficient of variation sample set. By combining outlier removal and maximum likelihood estimation, the coefficient of variation sample set is fitted with a double Weibull distribution, ultimately obtaining the cumulative distribution function used to characterize the random distribution of the coefficient of variation. Furthermore, the double Weibull distribution fitting can be re-executed based on new samples when the load type changes, thereby constructing a coefficient of variation distribution model that adapts to multiple load conditions and has clear statistical meaning. This provides a complete and reproducible probabilistic description framework for subsequently selecting a preset quantile level and constructing an adaptive threshold based on this distribution.

[0131] In step S4, the core task is to select the coefficient of variation value corresponding to the preset quantile level in the upper probability domain of the cumulative distribution function of the coefficient of variation obtained in step S3 as an adaptive threshold, and compare the coefficient of variation obtained in the current time window during the online detection process with the adaptive threshold. By recording the comparison results in adjacent time windows and examining the continuity and consistency of the comparison results, under the premise of meeting the operating conditions constraints of the aviation AC power supply line, the determination of aviation AC series arc fault is completed and the fault alarm information is output, so that the coefficient of variation forms a separable criterion boundary with the aviation AC series arc fault in a statistical sense.

[0132] In this embodiment of the application, the cumulative distribution function of the coefficient of variation obtained in step S3 can be denoted as: , where random variables This represents the coefficient of variation under normal operation and typical load conditions. Let represent the value of the coefficient of variation. This cumulative distribution function characterizes the cumulative probability of the coefficient of variation occurring at different values. Based on this, a preset quantile level can be selected in the upper probability domain, and the coefficient of variation value corresponding to this quantile level can be used as the adaptive threshold. Let the preset quantile level be denoted as . ,satisfy The adaptive threshold is denoted as Then it can be solved by Determine At this time, when obey When the coefficient of variation does not exceed The probability is ,Exceed The probability is ,because By selecting a region close to 1, the adaptive threshold is naturally located in the upper probability domain of the coefficient of variation sample distribution, which keeps the probability of the coefficient of variation being greater than the threshold at a low level under normal fluctuation conditions.

[0133] In some embodiments of this application, the preset quantile level is not arbitrarily given, but determined based on the tail characteristics of the cumulative distribution function of the coefficient of variation samples, and the adaptive threshold corresponds to the upper probability domain of the coefficient of variation sample distribution. Specifically, the empirical cumulative distribution of the coefficient of variation samples can be analyzed first to observe its performance in the high-value range (e.g., The gradient change within the portion (greater than the median of the samples) indicates the presence of a large number of high coefficient of variation samples near the maximum value when the cumulative probability increases rapidly in the tail region. In this case, a slightly lower quantile level can be selected to balance detection sensitivity and false alarm rate. When the tail region is long and sparse, a higher quantile level can be selected to prevent a few extremely high-value samples from excessively raising the threshold. In engineering implementation, a set of candidate quantile levels (e.g., 0.95, 0.975, 0.99, etc.) can be set first. The false negative and false alarm rates at each quantile level can be statistically analyzed using offline experimental data. The quantile level that can cover most normal samples while also including obviously abnormal samples in the upper probability domain can be selected as the final quantile level. In this way, the threshold itself becomes statistically sensitive to tail-end anomalous behavior.

[0134] Specifically, to more intuitively describe the typical range of values ​​for the coefficient of variation under normal operating conditions, a confidence interval can be established based on the cumulative distribution function to describe the range of values ​​for the coefficient of variation. Let the confidence level be denoted as... Typically, values ​​of 0.90, 0.95, or 0.99 can be used, with the corresponding lower and upper quantiles denoted as _____. and They satisfy:

[0135]

[0136] Under this definition, the coefficient of variation falls within the interval The probability is approximately This interval can be considered as the "main range of values ​​for the normal coefficient of variation," meaning the confidence interval describes the range of values ​​for the coefficient of variation. Based on this, the preset quantile levels can be... Select in greater than Within the range, for example, take and in the interval Inside or with Solving for the vicinity of the upper bound Get the corresponding The adaptive threshold is obtained by using the coefficient of variation value corresponding to the preset quantile level within the confidence interval. This approach can use the confidence interval to exclude the excessive influence of extreme outliers on the threshold, while ensuring that the threshold is in the upper part of the normal range. This makes the adaptive threshold reflect the coefficient of variation level "higher than most normal windows" in a probabilistic sense.

[0137] In some embodiments of this application, after determining the adaptive threshold, it is necessary to compare the coefficient of variation of the current time window with the adaptive threshold during the online detection process, and to perform comparisons and record the comparison results in adjacent time windows respectively. The comparison results of multiple adjacent time windows are used to determine whether to output the judgment result of aviation AC series arc fault.

[0138] Let the current time window number be... Its coefficient of variation is Adaptive threshold is A binary comparison result can be defined as follows:

[0139]

[0140] A comparison result of 1 indicates that the coefficient of variation for the current time window has fallen above the adaptive threshold, suggesting a high risk of an aviation AC series arc fault; a comparison result of 0 indicates that the coefficient of variation for the current time window is still within the normal statistical range. Furthermore, to suppress misjudgments caused by transient noise or external interference within a single time window, the comparison results of adjacent time windows can be statistically analyzed over a period of time. For example, a length of [missing information] can be set. The sliding observation window examines the most recent Comparison result sequence of time windows When there is at least one consecutive occurrence in the sequence When all comparison results are 1 (where...) The system determines that an AC series arc fault has occurred and outputs a fault alarm message. If the comparison result remains 0 for a certain period of time, it can be determined that no AC series arc fault has occurred. Through this mechanism, the judgment result is output when the comparison result is consistent at least once, thereby effectively reducing the probability of false alarms caused by single, occasional disturbances while avoiding the omission of persistent arc faults.

[0141] Furthermore, to avoid erroneously triggering aviation AC series arc fault alarms when the aviation AC power supply line is in the commutation, startup, or steady-state transition range, an operational status identification and pause decision mechanism can be introduced into the comparison logic. The power supply status can be classified by combining information such as power supply voltage phase, frequency change rate, and load operation commands: when a switch from bypass to main power is detected, when the aviation power supply undergoes commutation, or when a high-power load performs a startup / shutdown process, the current waveform will exhibit a sudden change in morphology and spectral broadening similar to an arc fault within a short period. If the comparison of the coefficient of variation with the adaptive threshold continues during these time periods, normal commutation, startup, or steady-state transition processes are easily misjudged as fault states. Therefore, when the operational status is identified as the commutation, startup, or steady-state transition range, the comparison logic can be temporarily marked as "frozen," pausing the update of the comparison results. The comparison of the coefficient of variation with the adaptive threshold is not performed. Once the power supply status is detected to have stabilized (e.g., frequency and voltage amplitude remain within the steady-state range for several consecutive cycles), the comparison logic resumes normal operation. This approach effectively reduces misjudgments caused by normal operating condition switching processes by pausing the comparison of the coefficient of variation with the adaptive threshold when the aviation AC power supply line is in the commutation, startup, or steady-state transition range, and resuming the comparison once the power supply status stabilizes.

[0142] In other embodiments of this application, to adapt to scenarios where load types frequently switch in aviation AC power supply systems, the load type of the aviation AC power supply line can be continuously monitored during online detection. When a change in load type is detected, the double Weibull distribution fitting is re-executed based on the new coefficient of variation samples, and the adaptive threshold is updated. Specifically, the aviation power loads currently in the access state can be classified using load identification algorithms or information provided by the load management system. When it is detected that the load type has switched from a linear load to a rectifier load, from a single load to multiple loads in parallel, or from a certain power level to another power level, a certain number of historical time window coefficient of variation samples can be re-accumulated under the new load conditions. These new samples are then added to or replaced in the coefficient of variation sample set, and the double Weibull distribution fitting process described in step S3 is executed again to obtain the updated cumulative distribution function. and the corresponding adaptive threshold After the update is completed, subsequent online comparison processes will use the new adaptive threshold. This will enable the double Weibull distribution fitting to be re-executed and the adaptive threshold updated based on the new coefficient of variation samples when a change in the load type of the aviation AC power supply line is detected. This ensures that the aviation AC series arc fault detection method maintains reasonable detection sensitivity and false alarm level even when the load structure changes.

[0143] Thus, step S4 selects the coefficient of variation value corresponding to the preset quantile level in the upper probability domain of the cumulative distribution function of the coefficient of variation as the adaptive threshold, and clarifies the typical value range of the coefficient of variation within the confidence interval framework. It then combines an online decision strategy that records the comparison results within adjacent time windows and examines the continuity and consistency of the comparison results; and a working condition management mechanism that suspends the comparison during commutation, startup, or steady-state transition intervals and refits and updates the adaptive threshold when the load type changes. In this way, a set of aviation AC series arc fault detection criteria that takes into account both statistical rigor and engineering robustness is constructed, enabling the aviation AC series arc fault detection method based on the coefficient of variation to maintain stable fault identification capability and output reliable fault alarm information in the complex and ever-changing aviation electrical environment.

[0144] In summary, this method constructs the current frequency distribution and extracts the coefficient of variation features within a time window synchronized with the basic power supply frequency. It combines the fitting of the double Weibull distribution of the coefficient of variation samples under different aviation power load conditions with the adaptive threshold selection of the tail preset quantile level. Ultimately, it achieves accurate characterization and statistical judgment of the abnormal increase in spectral dispersion caused by electric arc under multi-load and multi-state backgrounds. It can improve the sensitivity and robustness of aviation AC series arc fault detection while suppressing false alarms during normal transition processes such as commutation and startup, and reduce the dependence on threshold setting and human experience intervention. It provides a unified and scalable feature measurement and threshold construction framework for online safety monitoring and fault early warning of aviation power supply lines.

[0145] It should be noted that, although the embodiments in this application are based on... Figure 1 Steps S1 to S4 are described sequentially, but this does not mean that steps S1 to S4 must be performed in a strict order. The reason this embodiment follows this order is... Figure 1 The order in which steps S1 to S4 are described is provided to facilitate understanding of the technical solutions of the embodiments of this application by those skilled in the art. In other words, in the embodiments of this application, the order of steps S1 to S4 can be appropriately adjusted according to actual needs.

[0146] In some specific embodiments of this application, the aviation AC series arc fault detection method based on the coefficient of variation can be applied to the online monitoring scenario of aviation AC power supply lines.

[0147] Let the fundamental frequency of the power supply be denoted as . The fundamental period is denoted as The current sampling frequency is denoted as The sampling time interval is The method runs in the software environment of the processing unit, and completes a full fault detection process through sampling, time window division, frequency distribution construction, coefficient of variation calculation, statistical modeling, and online decision-making.

[0148] A complete implementation process may include the following steps:

[0149] Step 1: Install current acquisition channels on the target aircraft's AC power supply line and select the sampling frequency. Make ,in This refers to the highest frequency component that needs to be monitored. The fundamental power supply frequency is acquired synchronously during the sampling process. Or fundamental phase information, using the fundamental period Constructing a time reference series This provides a reference for synchronizing subsequent time windows with the basic power supply frequency.

[0150] Step 2: Discretize the current sequence based on the time reference sequence. Based on fixed window length ( (If the number is positive) is divided into continuous time windows. Each time window contains a set of sampling points. ,in Within each time window, The time window sequence serves as the current frequency dataset for this window. They are arranged continuously on the time axis and kept synchronized with the fundamental power supply frequency.

[0151] Step 3: Perform frequency component analysis on the current frequency dataset within each time window. (The last sentence appears to be incomplete and requires further context.) Taking a window as an example, Amplitude spectrum is obtained by performing spectral estimation. Then, a set of fixed frequency ranges is set according to the harmonic distribution in the aviation AC power supply line. Amplitude aggregation is performed on each interval to obtain... , and according to Normalization, forming a frequency distribution vector , is used to represent the probability distribution of each frequency component within the time window.

[0152] Step 4, based on frequency distribution vector Calculate the mean and standard deviation for this time window. The center of the frequency interval can be... Treated as the values ​​that a random variable can take, in order to Define the average value ,by Define standard deviation Therefore, a time window is constructed. coefficient of variation The time-varying coefficient sequence was obtained. It is used to quantify the degree of dispersion of frequency distribution within each time window.

[0153] Step 5: In the offline modeling phase, historical current signals are collected under various aviation electrical load conditions. For each load, following the procedures in Steps 1 to 4, the coefficient of variation is calculated for all time windows divided into several power frequency cycles, forming a historical coefficient of variation sample set. Statistical consistency tests and outlier removal are performed on the sample set to retain samples that represent normal fluctuations and typical weak fault disturbances, thus preparing input data for distribution fitting.

[0154] Step 6: Process the coefficient of variation sample set. Perform a double Weibull distribution fitting. Treat the coefficient of variation as a random variable. Assuming its cumulative distribution function is ,in, , , These are the scale parameter, shape parameter, and position parameter, respectively. Maximum likelihood estimation is used to evaluate the parameter vector. Solve by maximizing the log-likelihood function. Obtain parameter estimates and based on accumulated experience distribution and The deviation between the two values ​​is used as a basis for convergence and goodness of fit to obtain the cumulative distribution function that can characterize the random distribution of the coefficient of variation. .

[0155] Step 7: Based on the cumulative distribution function, set the preset quantile level according to the tail characteristics of the coefficient of variation. (For example Solve Obtain the adaptive threshold This threshold corresponds to the upper probability domain of the sample distribution of the coefficient of variation. If necessary, in Construct confidence intervals ,make and in Select near the upper end that satisfies of This enables threshold tuning that adapts to the statistical properties of the sample.

[0156] Step 8: During the online detection phase, the real-time acquired current signal is processed according to steps 1 to 4 to obtain the coefficient of variation for the current time window. .Will With adaptive threshold Comparison, defining a binary decision quantifier In length of The sequence examined within the sliding window When at least one of them occurs consecutively ( )indivual In the event of this situation, an aviation AC series arc fault is determined to exist and a fault alarm message is output; when If the value remains at 0 for an extended period, it is determined that no series arcing fault has occurred. During power supply commutation, load startup, or steady-state transition phases, updates are paused via operational status monitoring. The decision process resumes once the power supply stabilizes; when a change in load type is detected, the coefficient of variation sample is re-collected and updated. and This ensures that the threshold always matches the current load conditions.

[0157] Through the complete implementation process described above, this method uses the frequency distribution vector within a time window. and coefficient of variation Using this as the core feature, the current spectrum dispersion of aviation AC power supply lines under different loads and operating conditions is uniformly mapped to a scalar feature space; simultaneously, it utilizes a double Weibull distribution. With tail fraction threshold Constructing an adaptive decision boundary can maintain high sensitivity to series arc faults while suppressing false alarms from commutation, startup, and short-term interference. It can also maintain the stability of detection performance by refitting and updating thresholds when the load type changes, reducing manual threshold setting and reliance on experience, and improving the reliability, adaptability, and engineering application value of arc fault detection in aviation AC power supply lines.

[0158] It should be understood that the sequence number of each step in the above embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application.

[0159] The above-disclosed embodiments are merely preferred embodiments of the present invention, but the present invention is not limited thereto. Any non-creative variations that can be conceived by those skilled in the art, as well as any improvements and modifications made without departing from the principles of the present invention, should fall within the protection scope of the present invention.

Claims

1. A method for detecting AC series arc faults in aviation based on the coefficient of variation, characterized in that, include: The current signal of the aviation AC power supply line is collected, and the current signal is divided into continuous time windows according to a fixed duration corresponding to the basic power supply frequency. A current frequency dataset is formed in each time window to characterize the current frequency distribution in that time window. Within each time window, frequency component analysis is performed on the current frequency dataset to obtain a frequency distribution vector representing the probability distribution of each frequency component. The mean and standard deviation of the frequency distribution vector are calculated to obtain the coefficient of variation corresponding to that time window. The coefficient of variation is the ratio of the standard deviation to the mean. Based on historical current signals collected under different aviation electrical load conditions, the historical current signals are divided into windows, frequency components are analyzed and coefficients of variation are calculated to obtain the coefficients of variation corresponding to multiple historical time windows, and a coefficient of variation sample set is formed. The coefficient of variation sample set is fitted with a double Weibull distribution to obtain the cumulative distribution function used to characterize the random distribution characteristics of the coefficient of variation. Based on the cumulative distribution function, the coefficient of variation value corresponding to the preset quantile level is selected as the adaptive threshold. The coefficient of variation obtained in the current time window is compared with the adaptive threshold. When the coefficient of variation is greater than the adaptive threshold, it is determined that there is an aviation AC series arc fault in the current time window, and the fault alarm information is output.

2. The aviation AC series arc fault detection method based on the coefficient of variation as described in claim 1, characterized in that, The steps of dividing the current signal into continuous time windows of a fixed duration corresponding to the fundamental frequency of the power supply include: establishing a time reference based on the fundamental period of the power supply signal, and periodically synchronizing the current signal with the time reference to form continuous time windows, with each time window maintaining a synchronous relationship with the fundamental frequency of the power supply.

3. The aviation AC series arc fault detection method based on the coefficient of variation as described in claim 1, characterized in that, The step of performing frequency component analysis on the current frequency dataset within each time window to obtain a frequency distribution vector representing the probability distribution of each frequency component includes: performing frequency component analysis on the current frequency dataset within each time window, and dividing the spectral amplitude according to continuous frequency intervals to form a corresponding frequency distribution vector, wherein the frequency component analysis includes spectral estimation.

4. The aviation AC series arc fault detection method based on the coefficient of variation as described in claim 3, characterized in that, The frequency range division is set according to the distribution characteristics of harmonic sequences in aviation AC power supply lines, and the frequency range division method remains unchanged under different load conditions.

5. The aviation AC series arc fault detection method based on the coefficient of variation as described in claim 1, characterized in that, The steps for fitting a double Weibull distribution to the sample set of coefficients of variation include: The statistical consistency criterion based on the sample sequence is used to remove outliers from the coefficient of variation samples and fit them to a double Weibull distribution. The parameters of the double Weibull distribution are solved using the maximum likelihood estimation method, and the convergence of the parameters is checked based on the deviation relationship between the empirical cumulative distribution function and the fitted cumulative distribution function. The parameters include shape parameters, position parameters, and scale parameters.

6. The aviation AC series arc fault detection method based on the coefficient of variation as described in claim 1, characterized in that, The preset quantile level is determined based on the tail characteristics of the cumulative distribution function of the coefficient of variation samples, and the adaptive threshold corresponds to the upper probability domain of the coefficient of variation sample distribution.

7. The aviation AC series arc fault detection method based on the coefficient of variation as described in claim 1, characterized in that, The step of selecting the coefficient of variation value corresponding to the preset quantile level as the adaptive threshold based on the cumulative distribution function includes: establishing a confidence interval based on the cumulative distribution function, and using the coefficient of variation value corresponding to the preset quantile level in the confidence interval as the adaptive threshold. The confidence interval is used to describe the range of values ​​of the coefficient of variation.

8. The aviation AC series arc fault detection method based on the coefficient of variation as described in claim 1, characterized in that, The step of comparing the coefficient of variation obtained in the current time window with the adaptive threshold includes: performing comparisons and recording the comparison results in adjacent time windows, and outputting the judgment result when the comparison results are consistent at least once.

9. The method for detecting aviation AC series arc faults based on the coefficient of variation as described in claim 1, characterized in that, When the aviation AC power supply line is in the commutation, startup or steady-state transition range, the comparison of the coefficient of variation and the adaptive threshold is suspended and resumed after the power supply status is stable.

10. The method for detecting aviation AC series arc faults based on the coefficient of variation as described in claim 1, characterized in that, The method also includes: when a change in the load type of the aviation AC power supply line is detected, re-performing the double Weibull distribution fitting based on the new coefficient of variation sample and updating the adaptive threshold.

Images