Electric spark monitoring method for distribution box

Abstract

The invention relates to the technical field of data signal processing, in particular to a distribution box electric spark monitoring method which has the advantages of not depending on optical imaging conditions, adapting to complex environments and being capable of effectively distinguishing difference between mechanical vibration and discharge signals. According to the invention, non-intrusive monitoring of electric sparks in the distribution box can be realized, and the limitation of a traditional visual method in a closed space is overcome. And the fusion of the multi-dimensional acoustic features improves the recognition accuracy and reduces the false alarm rate. The introduction of environmental parameters enables the system to adapt to different working conditions, and improves the robustness of monitoring. The method does not need to reform the structure of the distribution box, can be widely applied to state monitoring of various low-voltage distribution equipment, and effectively prevents equipment faults and safety accidents caused by electric sparks.

Description

Technical Field

[0001] The present invention relates to the technical field of data signal processing, and in particular to a method for monitoring electric sparks in a distribution box. Background Art

[0002] Spark discharges in electrical equipment are a key risk for fires and equipment damage. Existing spark detection technologies primarily rely on visual imagery. For example, Chinese invention patent publication number CN120047901A proposes distinguishing sparks from interference sources by analyzing the brightness curve of a video stream. This method relies on the dynamic characteristics of image brightness changes. While it can reduce false alarm rates, it has significant limitations: it requires high-precision cameras and sufficient lighting, and fails in darkness, obscured light, or strong light interference. It also has high computational complexity and requires real-time processing of large numbers of image frames, placing heavy demands on hardware resources.

[0003] There are also technologies that use an end-to-end fully convolutional network (FCOS) to detect pantograph sparks, improving the recognition accuracy of small targets through an anchor-free design. However, model training relies on large amounts of labeled data, making actual deployment costly. They are still limited by optical imaging conditions, such as rain, fog, and lens contamination, and are unable to monitor non-visible areas, such as the enclosed space inside a distribution box.

[0004] Current spark monitoring solutions have shortcomings in applicable scenarios, anti-interference capabilities, and feature fusion mechanisms: The visual solution is limited by optical conditions and is difficult to cover enclosed equipment such as distribution boxes; Therefore, there is an urgent need for a monitoring method that integrates multi-dimensional acoustic emission characteristics and a dynamic compensation mechanism for environmental parameters to improve the detection robustness under complex working conditions. Summary of the Invention

[0005] (1) Technical issues to be resolved To solve the above problems, the present invention proposes a method for monitoring electric sparks in distribution boxes, which aims to solve the problem in the prior art that visual solutions are subject to optical conditions and are difficult to cover closed equipment such as distribution boxes.

[0006] (2) Technical solution A method for monitoring sparks in a distribution box according to the present invention comprises: Acquire the acoustic emission signal inside the distribution box; Acquiring characteristic data of the acoustic emission signal, the characteristic data comprising time domain characteristic parameters and frequency domain characteristic parameters extracted from the acoustic emission signal based on preset signal processing rules; Acquire environmental parameters of the distribution box, the environmental parameters including at least one of temperature, humidity, and current load, and The characteristic data of the acoustic emission signal and the environmental parameters are provided to an electric spark recognition model to perform electric spark state recognition.

[0007] Another computing device of the present invention comprises: at least one processor; and The memory stores instructions, and when the instructions are executed by the at least one processor, the at least one processor executes the distribution box electric spark monitoring method as described in any one of the above technical solutions.

[0008] Another non-transitory machine-readable storage medium of the present invention stores executable instructions, which, when executed, enable a machine to execute the distribution box electric spark monitoring method as described in any one of the above technical solutions.

[0009] (3) Beneficial effects Compared with the prior art, the present invention has the following beneficial effects: The present invention solves the problems of existing technologies such as reliance on optical conditions, inability to monitor enclosed spaces, and insufficient anti-interference capabilities through the fusion analysis of acoustic emission signal characteristic data and environmental parameters, combined with dynamic thresholds and adaptive weight adjustment mechanisms. It has the advantages of not relying on optical imaging conditions, adapting to complex environments, and being able to effectively distinguish between mechanical vibrations and discharge signals. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0011] Figure 1 Schematic diagram of the framework structure of the distribution box electric spark monitoring method; Figure 2 The following is a curve diagram comparing the characteristics of electric sparks and acoustic emission signals of interference sources (time domain characteristics); Figure 3 The following is a curve diagram comparing the characteristics of electric sparks and acoustic emission signals from interference sources (frequency domain characteristics); Figure 4 A schematic diagram of the execution device framework.

[0012] 1. Processor, 2. Memory, 3. Communication interface, 4. Communication bus. DETAILED DESCRIPTION

[0013] Example 1

[0014] Traditional spark monitoring technology relies on optical sensors to capture dynamic brightness characteristics. However, signal acquisition is limited by the light penetration capability of enclosed spaces and physical obstruction of equipment, making it ineffective for monitoring scenarios inside distribution boxes. While deep learning models based on fully convolutional networks can improve the accuracy of small target recognition, they require continuous acquisition of high-resolution image data streams, which puts edge computing devices under real-time processing pressure and storage resource bottlenecks. Environmental interference factors directly affect the signal-to-noise ratio of optical signals. For example, electromagnetic interference generated by the metal casing of a distribution cabinet can couple to the video transmission link, causing image distortion and feature drift, reducing the reliability of spark feature extraction.

[0015] For example, in the operating environment of industrial-grade low-voltage distribution boxes, the metal cabinet creates a Faraday cage effect, preventing external cameras from penetrating the box to capture internal discharges. Solutions using built-in micro-cameras face the risk of device overheating and failure due to insufficient heat dissipation space, and the mechanical vibrations generated by the operation of the circuit breaker inside the box can cause the lens to lose focus. When the load current exceeds 400A, electromagnetic radiation from the busbar causes pixel saturation in the image sensor, obscuring the brightness variation characteristics of actual sparks. While acoustic emission sensors can capture mechanical vibration signals within the box, traditional fixed-threshold detection methods cannot distinguish between sparks and contactor pull-in noise. When the humidity exceeds 85% RH, the sensitivity of piezoelectric ceramic sensors decreases, causing high-frequency signals to attenuate.

[0016] If these issues are not addressed, spark monitoring systems will continue to have blind spots, failing to warn of early discharges within enclosed spaces, leading to undetected insulation degradation. Increased false alarm rates will trigger redundant maintenance work orders, increasing equipment downtime and labor inspection costs. Undetected events can cause localized overheating and accelerate contact oxidation, ultimately leading to main circuit shorts or arc faults, resulting in cascading equipment damage. The lack of monitoring system reliability will also hinder the development of a condition-based maintenance system for intelligent distribution systems, hindering improvements in the safe operation of power supply networks.

[0017] Faced with these challenges, this application first considered overcoming the physical limitations of visual monitoring and experimented with non-optical sensing methods to collect spark signals. Given the impossibility of deploying optical equipment in enclosed spaces, the application explored the use of mechanical vibration waves as signal carriers, using acoustic emission sensors to capture the transient elastic waves generated by sparks. However, traditional acoustic detection is subject to interference from ambient noise, necessitating the development of an anti-interference feature extraction mechanism.

[0018] Further analysis found that the electric spark acoustic emission signal has high-frequency transient characteristics, while interference such as contactor pull-in is mostly low-frequency components. In this regard, the present application attempts to construct a time-frequency joint analysis framework to extract signal mutation characteristics from the time domain and filter high-frequency energy distribution from the frequency domain. However, fixed thresholds are difficult to adapt to background noise fluctuations under different working conditions. For example, when high humidity causes the sensor sensitivity to decrease, it is necessary to introduce dynamic correction judgment logic for environmental parameters. To resolve the above contradictions, the present application uses temperature, humidity, and current load as compensation variables to explore their coupling relationship with acoustic characteristics. By establishing a multi-dimensional joint judgment model, the characteristic threshold is adaptively adjusted with the environmental parameters to avoid false triggering of a single condition. Ultimately, a closed-loop detection system is formed in which acoustic signals and environmental parameters work together to suppress working condition interference while retaining the advantages of high-frequency characteristics.

[0019] like Figure 1-2 As shown, in order to address the above-mentioned defects, this disclosure proposes a method for monitoring sparks in a distribution box, which mainly includes the following core steps: S100: Acquire an acoustic emission signal inside a distribution box; S200, acquiring characteristic data of the acoustic emission signal, the characteristic data including time domain characteristic parameters and frequency domain characteristic parameters extracted from the acoustic emission signal based on preset signal processing rules; S300. Acquire environmental parameters of the distribution box, where the environmental parameters include at least one of temperature, humidity, and current load. Then, provide the characteristic data of the acoustic emission signal and the environmental parameters to a spark recognition model to perform spark state recognition.

[0020] Specifically, the acoustic emission signal refers to the mechanical wave signal generated by electric spark discharge inside the distribution box. It can be collected by an acoustic emission sensor, which converts mechanical vibration into an electrical signal through the piezoelectric effect to capture the high-frequency transient characteristics of the electric spark, breaking through the limitations of optical monitoring on closed spaces.

[0021] Characteristic data refers to the time domain characteristic parameters and frequency domain characteristic parameters extracted from the acoustic emission signal. Specifically, it can be calculated using short-time zero-crossing rate, rise time, peak amplitude, spectrum kurtosis and the energy proportion of a specific frequency band. It integrates the time domain waveform change rate and frequency domain energy distribution characteristics to enhance the ability to characterize the electric spark characteristics and overcome the defect that a single characteristic dimension is easily interfered by noise.

[0022] Environmental parameters refer to the temperature, humidity, and current load parameters inside the distribution box. Specifically, they can be collected in real time using temperature sensors, humidity sensors, and current transformers. They can be used to dynamically correct the judgment threshold of characteristic parameters or adjust the model weight coefficient to solve the misjudgment problem caused by environmental fluctuations in traditional methods.

[0023] The electric spark recognition model refers to a state recognition algorithm based on multi-dimensional joint criteria. Specifically, it can be implemented using a judgment rule that combines dynamic thresholds with fixed thresholds. By simultaneously meeting the threshold conditions of short-time zero-crossing rate, rise time, spectral kurtosis, and frequency band energy ratio, it can reduce the false alarm rate and improve the detection robustness under complex working conditions.

[0024] The core innovation of this application lies in the construction of a non-vision-dependent electric spark monitoring system through the combination of acoustic sensing technology, multi-dimensional feature fusion and dynamic compensation mechanism of environmental parameters, the use of acoustic emission signals to break through the monitoring limitations of closed spaces, the fusion of time domain and frequency domain features to improve anti-interference capabilities, and the introduction of environmental parameters to dynamically adjust model parameters to solve the problems of false alarms and missed alarms caused by optical conditions limitations, environmental sensitivity and single feature dimensions in existing technologies.

[0025] The working process and principle of this application are as follows: the distribution box spark monitoring method first uses an acoustic emission sensor to acquire acoustic emission signals from within the distribution box. Acoustic emission signals are caused by mechanical fluctuations generated during spark discharge and can propagate through the metal casing of the distribution box. Next, feature extraction is performed on the acquired acoustic emission signals, including time domain and frequency domain feature parameters. Time domain feature parameters reflect the transient variation characteristics of the signal, while frequency domain feature parameters characterize the spectral distribution of the signal. Feature extraction is performed based on preset signal processing rules to capture the typical acoustic characteristics of spark discharge. Simultaneously, environmental parameters of the distribution box are acquired, including at least one of temperature, humidity, and current load. These environmental parameters are used to compensate for variations in the acoustic emission signal characteristics under different operating conditions. Finally, the characteristic data of the acoustic emission signal and the environmental parameters are input into a spark recognition model. The model comprehensively considers the acoustic characteristics and environmental factors to perform spark status recognition. This multi-dimensional information fusion method can improve recognition accuracy and robustness.

[0026] The following is a detailed description of this plan: A piezoelectric acoustic emission sensor with a sensitivity of -65dB is installed inside the distribution box. The acoustic emission signal is initially amplified by a preamplifier and then sampled by an analog-to-digital converter at a sampling frequency of 1MHz. The sampled digital signal is framed, with each frame consisting of 1024 points. Time domain characteristic parameters are extracted from each frame, including the short-term zero-crossing rate, signal rise time, and peak amplitude. The signal is subjected to a fast Fourier transform to extract frequency domain characteristic parameters, including spectral kurtosis and energy fraction in the 30-50kHz band. Environmental parameters are collected using a temperature and humidity sensor and a current transformer, with a sampling period of 1 minute. The spark recognition model uses an ensemble learning algorithm based on a decision tree. The model inputs include the aforementioned acoustic characteristics and environmental parameters. The model outputs a spark status judgment: normal, suspected, and confirmed. If the spark status is suspected or confirmed, the system will issue an alarm signal of the corresponding level.

[0027] In some of the above-mentioned solutions of this application, it is proposed to obtain characteristic data of acoustic emission signals for spark state identification. However, when extracting time domain characteristic parameters, there are problems such as unclear parameter definition and non-specific calculation logic, which leads to inaccurate signal feature representation and affects the accuracy of subsequent model identification.

[0028] In this regard, the present application further proposes that the time domain characteristic parameters include short-time zero-crossing rate, rise time and peak amplitude, wherein the calculation expression of the short-time zero-crossing rate is: ,in is the number of sampling points in the analysis window, is the discrete acoustic emission signal, is the sampling point number of the time series, is a symbolic function defined as: .

[0029] Rise time is defined as the time required for a signal to rise from 10% to 90% of its peak amplitude; peak amplitude is the maximum absolute amplitude of the signal within the analysis window.

[0030] The short-term zero-crossing rate is calculated by differentially calculating the sign function. For example, when the signs of adjacent sampling points are different, the absolute value of the difference is 2, and when the signs are the same, it is 0, thereby quantifying the signal fluctuation frequency. The number of sampling points in the analysis window can be set to 500 points, corresponding to a 25ms time window, covering the duration of a typical electric spark event. The 10%-90% interval of the rise time is intercepted by setting a dynamic threshold. For example, when the signal amplitude first exceeds 10% of the peak value, the starting point is marked, and when it reaches 90% of the peak value, the end point is marked to avoid baseline noise interference. The peak amplitude is calculated using sliding window extreme value detection. For example, the maximum absolute value is extracted within a 10ms subwindow to capture millisecond-level discharge events.

[0031] Specifically, the calculation expression for the short-time zero-crossing rate suppresses small-amplitude noise through a sign function differential mechanism. For example, when the signal amplitude fluctuation is less than the noise threshold, the sign function remains stable, and the calculated zero-crossing rate approaches zero, thereby distinguishing the high-frequency oscillations of the electric spark from background noise. The definition of the rise time eliminates signal saturation errors by intercepting the main energy interval. For example, when the signal amplitude reaches 100% of the peak value, sensor saturation may cause time measurement distortion, while the 10%-90% interval can accurately reflect the physical process of discharge channel establishment. The calculation of the maximum absolute value of the peak amplitude is directly related to the instantaneous discharge energy. For example, experimental data shows that its correlation coefficient with the current probe measurement value reaches 0.93, verifying the effectiveness of energy characterization.

[0032] The synergistic effect of these parameters is achieved through hierarchical parsing logic: the short-term zero-crossing rate is used to detect the presence of transient events in the signal. For example, when the calculated result exceeds a dynamic threshold, subsequent analysis is triggered; the rise time is used to confirm the event type. For example, the rise time of an EDM signal is typically less than 1ms, while the rise time of mechanical noise exceeds 5ms; the peak amplitude further quantifies the event intensity. For example, when the amplitude exceeds 10mV, it is determined to be a valid discharge. The analysis window duration is set to 25ms, covering the typical EDM duration of 10-20ms, ensuring the integrity of feature extraction. Through this combination of parameters, the characterization error of the time domain features is reduced from 18% with traditional methods to 3%, and the model recognition accuracy is improved to 92%.

[0033] As a preferred embodiment, this solution is specifically implemented as follows: In the distribution box spark monitoring method, the time domain characteristic parameters include short-time zero-crossing rate, rise time and peak amplitude.

[0034] Rise time is defined as the time it takes for a signal to rise from 10% to 90% of its peak amplitude. For example, for an acoustic emission signal with a sampling frequency of 100 kHz, if 50 sampling points are passed from 10% to 90% of its peak amplitude, the rise time is 0.5 ms.

[0035] The peak amplitude is the maximum absolute amplitude of the signal within the analysis window. For example, if the maximum positive amplitude sampled within a 25ms analysis window is 0.8V and the maximum negative amplitude is -0.6V, the peak amplitude is 0.8V.

[0036] Through the above technical solution, this application solves the problem of inaccurate time-domain feature extraction of acoustic emission signals. The short-term zero-crossing rate quantifies the frequency of signal fluctuations within a time window, distinguishing the transient characteristics of the spark from the stationary characteristics of background noise. The rise time reflects the steep changes in the spark signal. The peak amplitude captures the energy intensity of the spark. These parameters provide the spark recognition model with physically meaningful and quantifiable time-domain feature inputs, improving the accuracy of feature representation and, in turn, the accuracy of subsequent model recognition.

[0037] In some of the above-mentioned solutions of this application, time domain characteristic parameters are proposed to characterize the waveform characteristics of the acoustic emission signal. However, it is difficult to effectively distinguish the frequency domain distribution differences between electric sparks and background noise by relying solely on time domain characteristics. Especially under complex environmental noise interference, time domain parameters are easily affected by non-stationary signal fluctuations, resulting in the risk of misjudgment in electric spark state identification.

[0038] In this regard, the present application further proposes that frequency domain characteristic parameters include spectral kurtosis and 30-50kHz frequency band energy proportion, wherein the calculation expression of spectral kurtosis is: ,in is the spectrum amplitude after discrete Fourier transform, is the frequency point number, is the spectrum mean, is the frequency index; The energy ratio of the 30-50kHz frequency band is defined as: , is the sampling frequency of the signal, Frequency The power spectral density is represents the Nyquist frequency.

[0039] Spectral kurtosis is calculated as the ratio of the fourth-order central moment of the spectrum amplitude after the discrete Fourier transform to the square of the second-order central moment. Its physical nature reflects the steepness of the signal spectrum. For example, the transient impact characteristics of an electric spark discharge result in a peaked spectrum, where the kurtosis value is significantly higher than the smooth spectral distribution of background noise. The calculation of spectral kurtosis can be further normalized to eliminate the influence of signal amplitude fluctuations, ensuring comparability of the characteristics of sparks of different intensities.

[0040] The energy contribution in the 30-50kHz frequency band is calculated based on the integral of the power spectral density. Low-frequency interference is suppressed by concentrating the energy in a preset frequency band. Specifically, the primary energy of spark acoustic emissions is distributed within the 35±5kHz range, while the primary frequencies of switching arcs and mechanical vibrations are below 15kHz and 20kHz, respectively. Electromagnetic interference is concentrated above 100kHz. The frequency band selection was validated by experimental data. For example, in actual measurements of distribution boxes, the spark signal's energy contribution in the 30-50kHz frequency band exceeded 65%, while background noise contributed less than 12%.

[0041] Specifically, the combined application of spectral kurtosis and frequency band energy fraction forms a complementary noise reduction mechanism. Spectral kurtosis effectively captures the transient impact characteristics of electric sparks by quantifying the non-Gaussian nature of the spectrum, suppressing the interference of broadband stationary noise. For example, when fan vibration is present in a distribution box, its spectrum distribution is flat, with a kurtosis value close to 3, while the kurtosis value of electric sparks can reach over 5.3. Frequency band energy fraction filters out interference from low-frequency noise by extracting the energy concentration characteristics of a preset high-frequency band. For example, in scenarios where a sudden change in current load causes the time domain features to fail, the frequency band energy fraction can remain stable because the high-frequency energy distribution of electric sparks is not affected by signal propagation attenuation.

[0042] The combined judgment criteria of the two further achieve environmental adaptability through dynamic thresholds. For example, when the ambient temperature rises and the sensor sensitivity decreases, the judgment threshold of the frequency band energy ratio can be dynamically adjusted based on historical data to ensure detection stability. Experimental data shows that after introducing dual-frequency domain features, the false positive rate under electromagnetic interference decreases by 62%, and the missed detection rate in mechanical vibration scenarios is reduced to 3.2%. The synergy between frequency domain features and time domain parameters constructs a multidimensional feature space, improving recognition accuracy through interpretable mapping of physical properties. For example, spectral kurtosis is positively correlated with discharge instability, and the frequency band energy ratio reflects the discharge energy release efficiency, thus providing a reliable basis for graded warning of electric spark status.

[0043] As a preferred embodiment, this solution is specifically implemented as follows: The frequency domain characteristic parameters include spectrum kurtosis and energy proportion in the 30-50kHz band. The spectrum kurtosis is obtained by calculating the ratio of the fourth-order moment to the square of the second-order moment of the spectrum amplitude after discrete Fourier transform. Specifically, the acoustic emission signal is first subjected to discrete Fourier transform to obtain the spectrum amplitude. Then calculate the spectrum mean. Then calculate the fourth power of the difference between the amplitude of each frequency point and the mean, and sum them. At the same time, calculate the square of the difference between the amplitude of each frequency point and the mean, sum them and square them again. Finally, divide the former by the latter to get the spectrum kurtosis .

[0044] The energy proportion of the 30-50kHz frequency band is obtained by calculating the ratio of the power spectrum density integral in the 30-50kHz frequency band to the total power spectrum density integral. The specific steps are: first, estimate the power spectrum density of the acoustic emission signal and obtain Then calculate the power spectrum density integral in the 30-50kHz frequency band and the total power spectrum density integral from 0 to Nyquist frequency. Finally, divide the former by the latter to get the frequency band energy ratio. .

[0045] Therefore, spectral kurtosis can quantify the non-Gaussian characteristics of the signal spectrum and capture the transient impact characteristics of the EDM signal in the frequency domain. The energy proportion in the 30-50kHz frequency band can extract the high-frequency energy concentration of the EDM. These two frequency domain characteristics complement the time domain parameters, enhancing the anti-interference capability of EDM recognition.

[0046] Through the above technical solution, the present application solves the problem that it is difficult to effectively distinguish the frequency domain distribution differences between electric sparks and background noise by relying solely on time domain features. The spectral kurtosis can effectively capture the transient impact characteristics of the electric spark signal in the frequency domain, which is different from the smooth spectrum distribution of the background noise. The energy proportion of the 30-50kHz frequency band can suppress the interference of low-frequency environmental noise on feature extraction. These two frequency domain features complement the time domain parameters. Through the joint judgment of multi-dimensional features, the anti-interference ability of electric spark identification is enhanced, especially when there is mechanical vibration or electromagnetic noise inside the distribution box, the acoustic emission signal characteristics of the electric spark can be more accurately separated.

[0047] In some of the above-mentioned schemes of this application, it is proposed to identify electric sparks by combining the time domain and frequency domain characteristic parameters of the acoustic emission signal with environmental parameters. However, in this process, if a fixed threshold is used for state judgment, it cannot adapt to the dynamic changes of the background noise inside the distribution box and the influence of different environmental parameters on the sensitivity of the characteristic parameters, resulting in misjudgment or missed detection.

[0048] In this regard, the present application further proposes that the electric spark recognition model performs state recognition through a multi-dimensional joint criterion, and the multi-dimensional joint criterion is expressed as: in, is a dynamic threshold, which is adaptively generated based on the statistical characteristics of background noise. is the rise time upper threshold, is the lower threshold of spectrum kurtosis, is the frequency band energy ratio threshold, is the rise time.

[0049] The dynamic threshold is calculated using the moving average and standard deviation of the short-term zero-crossing rate within a time window. The moving average is used to track the baseline level of background noise, and the standard deviation is used to quantify the intensity of noise fluctuations. The weighting coefficient is dynamically adjusted based on temperature and humidity parameters, decreasing the weight of the moving average with increasing temperature and increasing the weight of the standard deviation with increasing humidity. The upper threshold for rise time is set to 500 nanoseconds, based on gas breakdown theory, to eliminate slow interference signals. The lower threshold for spectral kurtosis is set to 4.2, calibrated through experiments comparing electric sparks and mechanical noise in steel distribution boxes. The frequency band energy ratio threshold is set to 65%, based on the propagation attenuation characteristics of sound waves in metal boxes.

[0050] Specifically, during acoustic emission signal processing, the moving average and standard deviation of the short-term zero-crossing rate are first calculated. A dynamic threshold is then generated by combining the temperature compensation coefficient and humidity gain factor. When the short-term zero-crossing rate exceeds the dynamic threshold, subsequent feature parameter verification is triggered. The rise time is calculated by measuring the duration from 10% to 90% of the peak amplitude and comparing it to an upper threshold of 500 nanoseconds. The spectral kurtosis is calculated by taking the spectral amplitude after discrete Fourier transform. A kurtosis value below 4.2 is considered a non-spark signal. The energy fraction of the 30-50 kHz frequency band is calculated by integrating the power spectral density. If the fraction is below 65%, low-frequency interference is excluded. Four criteria must be met simultaneously to filter out stationary noise, slow interference, broadband noise, and low-frequency noise in sequence. Finally, a spark is determined only when all criteria are met. By combining dynamic thresholds with fixed physical thresholds, the short-term zero-crossing rate requirement is automatically relaxed when background noise increases due to high temperatures, while maintaining rigid constraints on essential characteristics such as rise time and spectral kurtosis. This balance between detection sensitivity and interference rejection is achieved under complex operating conditions. Experimental data show that the misjudgment rate of this method is reduced to 2.1% in the motor start-stop interference scenario, and the detection rate is still maintained at 88% in a 95% humidity environment.

[0051] In specific implementation, the acoustic emission signal from the distribution box is first acquired. Characteristic parameters such as the signal's short-term zero-crossing rate, rise time, spectral kurtosis, and energy percentage in the 30-50kHz frequency band are then extracted. Environmental parameters such as the distribution box's temperature and humidity are also acquired.

[0052] Next, adaptively generate based on the statistical characteristics of the background noise Dynamic threshold The extracted characteristic parameters are compared with the corresponding thresholds to determine whether all four conditions are met. If all conditions are met, it is determined to be an electric spark state; otherwise, it is determined to be a normal state.

[0053] This application further proposes that the dynamic threshold is generated by the following expression: in, is the moving average of the short-term zero-crossing rate in the time window, is the standard deviation of the short-time zero-crossing rate, are their respective weight coefficients. Among them, the weight coefficient The weight coefficient is limited to the range of 1.2 to 1.8. The moving average is used to characterize the steady-state level of background noise, and the standard deviation is used to quantify the intensity of transient fluctuations in noise. The lower limit value of ensures that the steady-state noise component accounts for no less than 40% in the threshold calculation, and the weight coefficient The upper limit of limits the impact of transient fluctuations on the threshold to no more than 83%. The constraint condition controls the coverage of the dynamic threshold within the theoretical noise fluctuation range of three standard deviations by maintaining the sum of the weight coefficients approximately equal to 3.

[0054] Specifically, the dynamic threshold generation process first calculates the moving average and standard deviation based on the short-term zero-crossing rate data within the time window. The moving average is updated in real time through the sliding window algorithm to reflect the long-term statistical characteristics of the background noise; the standard deviation is dynamically obtained through the variance calculation module to capture the fluctuation intensity of short-term noise. The weight coefficient is pre-set to a fixed interval value, where the weight coefficient of the steady-state component is The lower limit is set to 1.2 to ensure that steady-state noise dominates the threshold calculation; the weight coefficient of the transient component The upper limit is set to 2.5 to prevent sudden interference from excessively raising the threshold. The constraint condition forces the sum of the weight coefficients to be stable between 2.7 and 3.3 through a mathematical relationship, so that the final dynamic threshold can cover more than 99% of the background noise fluctuation range. By limiting the value range of the weight coefficients and their mutual relationship, this technical solution constructs a noise tolerance interval with probabilistic statistical significance. While ensuring the ability to control the steady-state noise baseline, it effectively suppresses the impact of transient interference on the threshold calculation, thereby improving the robustness and accuracy of EDM recognition.

[0055] As a preferred embodiment, you can choose , , making , satisfying the constraints. Furthermore, the time window can be set to 1 second, and the moving average It can be obtained by calculating the short-term zero-crossing rate every 100 milliseconds and then taking the average of the last 10 calculation results. The calculation can be performed based on the same 10 calculation results. Thus, the dynamic threshold can adaptively reflect the statistical characteristics of the background noise.

[0056] In some of the above-mentioned schemes of the present application, it is proposed to adjust the weight coefficient through a dynamic threshold generation mechanism to optimize the adaptability of the spark recognition model. However, in this process, the adjustment of the weight coefficient is only based on a fixed reference value and cannot be dynamically compensated according to changes in actual environmental parameters, resulting in deviations that may not match the working conditions during the threshold generation process, thereby affecting the accuracy and robustness of spark state recognition.

[0057] In this regard, the present application further proposes a weight coefficient and weight coefficient Dynamic adjustment based on environmental parameters: in is the base weight, the default value is , is the temperature compensation coefficient, Moderate compensation coefficient are the attenuation factor and the gain factor respectively; The weight coefficient and the weight coefficient The process of dynamic adjustment according to environmental parameters is optimized by gradient descent algorithm: ,in The ideal threshold is deduced from historical missed detection / false alarm data. is the constraint penalty coefficient, The temperature compensation coefficient is constructed as a temperature-driven exponential decay function, and the specific expression can be set as ,in is the difference between the ambient temperature and the reference temperature. When the ambient temperature rises, the coefficient decays exponentially, making the weight coefficient The humidity compensation coefficient is constructed as a humidity-driven nonlinear gain function, and the specific expression can be set as: ,in is the difference between ambient humidity and reference humidity. When the humidity increases, the coefficient shows a linear growth trend. The cross term captures the synergistic effect of temperature and humidity, making the weight coefficient As the severity of the environment increases dynamically, it compensates for characteristic fluctuations caused by signal transmission attenuation.

[0058] The gradient descent optimization process is configured as a dual-objective loss function, which includes a threshold approximation term and a constraint penalty term. The former drives the weight coefficient to converge to the theoretical critical value inferred from historical missed detection / false alarm data by calculating the Euclidean distance between the dynamic threshold and the ideal threshold. The latter is achieved by constraining the penalty term coefficient. mandatory During the optimization process, the lower limit of the temperature compensation coefficient was set to 0.8, and the upper limit of the humidity compensation coefficient was set to 3.0 to prevent the parameters from getting out of control under extreme working conditions.

[0059] Specifically, during the dynamic threshold generation process, the temperature compensation coefficient and the humidity compensation coefficient are linked to the environmental parameters through a preset mathematical relationship. When the temperature sensor detects a rise in ambient temperature, the temperature compensation coefficient decays exponentially, reducing the short-term zero-crossing rate mean weight and suppressing the baseline drift caused by thermal noise. When the humidity sensor detects an increase in ambient humidity, the humidity compensation coefficient is increased by linearly superimposing cross-terms, which enhances the standard deviation weight and relaxes the tolerance for signal fluctuations. The gradient descent algorithm adjusts the compensation coefficient through iterative calculation, so that the dynamic threshold gradually approaches the ideal threshold determined by the minimum detectable threshold of historical missed samples and the maximum noise amplitude of false alarm samples, while maintaining the requirements of statistical principles through constraint terms. This mechanism is achieved through cross-term in the scenario of high temperature and high humidity synergy. The super-linear growth of the compensation coefficient is achieved, which effectively offsets the signal characteristic distortion caused by complex environmental factors, so that the dynamic threshold can still accurately distinguish between electric sparks and background noise under complex working conditions.

[0060] In the case of abnormal temperature, the mechanical vibration noise generated by the thermal expansion of metal structural parts causes the overall signal amplitude to rise. At this time, the formula Dynamically increase the peak amplitude judgment threshold, for example, when the temperature rises by 10℃, the threshold is increased by 20%, effectively avoiding misjudging thermal noise as electric sparks. In a high humidity environment, the air medium has an enhanced attenuation effect on the sound waves in the 30-50kHz frequency band. The decision weight of the spectral kurtosis parameter is strengthened to compensate for the weakening of features caused by high-frequency energy attenuation. Under overload conditions, the low-frequency noise generated by electromagnetic interference is removed by a 20kHz high-pass filter, retaining only pure high-frequency features for energy contribution calculation. For example, when the load reaches 150%, the 1-5kHz contactor jitter noise is directly blocked. These three compensation mechanisms form a multi-dimensional synergy with the dynamic weight coefficient optimization in the previous solution. The temperature compensation coefficient and the weight attenuation factor are jointly optimized through a gradient descent algorithm, ensuring that the electric spark recognition model maintains a comprehensive error rate below 6% under extreme conditions of 70°C high temperature, 95% humidity, and 150% load.

[0061] As a preferred embodiment, the solution of this application is specifically implemented as follows: Environmental parameters are used to dynamically adjust the decision logic for characteristic parameters. When the temperature exceeds the preset temperature threshold, the peak amplitude decision threshold is proportionally lowered. For example, the preset temperature threshold can be set to 50°C. When the temperature inside the distribution box reaches 55°C, the peak amplitude decision threshold is lowered from 100mV to 90mV. When the humidity exceeds the preset humidity threshold, the weight coefficient of the frequency domain characteristic parameters is increased. For example, the preset humidity threshold can be set to 80%. When the humidity inside the distribution box reaches 85%, the weight coefficient of the spectral kurtosis is increased from 1.0 to 1.2. When the current load exceeds the rated value, the extraction of frequency domain features below 20kHz in the acoustic emission signal is blocked. For example, the rated current load can be set to 100A. When the actual current load of the distribution box reaches 120A, the collected acoustic emission signal is subjected to a 20kHz high-pass filter, and only the frequency domain features above 20kHz are extracted for subsequent analysis.

[0062] In some of the above-mentioned schemes of the present application, it is proposed to realize the spark status identification through multi-dimensional joint judgment criteria and dynamic adjustment mechanism of environmental parameters. However, in this process, it is only possible to determine whether the spark exists, and it is impossible to respond in a graded manner according to the spark energy level and occurrence frequency differences. As a result, it is impossible to take differentiated early warning measures for sparks of different dangerous levels, which may cause the problem of missing high-risk hidden dangers or excessive triggering of low-risk alarms.

[0063] In this regard, the present application further proposes that the spark status identification includes graded warnings: when a spark with energy lower than 10μJ is detected, a general maintenance warning is triggered; when a spark with energy higher than 100μJ is detected, an emergency shutdown warning is triggered; when the frequency of sparks in unit time exceeds a threshold, a system maintenance warning is triggered.

[0064] The energy threshold is set based on an indirect calculation model and is calculated using the peak amplitude, rise time, and frequency band energy parameters of the acoustic emission signal. The specific calculation formula is: Among them, the calibration coefficient The frequency threshold is determined based on the material type of the distribution box. For steel boxes, it is 1.0, and for plastic boxes, it is 1.8. The frequency threshold is generated in real time in relation to the current load, and the dynamic adjustment formula is: When the current load reaches 150% of the rated value, the frequency threshold is automatically reduced to 60% of the original value.

[0065] Specifically, after detecting the acoustic emission signal, the peak amplitude and rise time in the time domain characteristic parameters are first extracted, combined with the energy proportion of the 30-50kHz frequency band in the frequency domain characteristic parameters, and the spark energy is calculated through a multi-dimensional joint judgment. When the ambient temperature exceeds 40°C, the peak amplitude judgment threshold is attenuated by a factor of 0.8 to compensate for the signal attenuation caused by high temperature. When the calculated energy is lower than 10μJ, a general maintenance instruction is generated and recorded in the equipment maintenance log, which the operation and maintenance personnel can handle during the next planned maintenance; when the energy is higher than 100μJ, the power supply to the distribution box is immediately cut off and an audible and visual alarm is triggered; when the spark frequency exceeds the dynamic threshold within 1 hour, a system maintenance work order is automatically generated and pushed to the management platform.

[0066] By combining energy and frequency thresholds, low-energy discharge events are classified as general hazards that can be addressed later, while high-energy single discharges are identified as urgent risks. Frequent anomalies indicate systemic failures. Experimental data shows that this classification mechanism has reduced unnecessary downtime by 78% without causing major accidents due to missed inspections. The introduction of dynamic frequency thresholds enables accurate identification of insulation degradation trends even under fluctuating load conditions, predicting breakdown risks up to seven days in advance.

[0067] As a preferred embodiment, during the distribution box spark monitoring process, the characteristic data of the acoustic emission signal and environmental parameters are collected in real time and input into the spark recognition model. When a spark event is identified, its energy value is estimated using a preset calculation model, where the calculation of the energy value involves a nonlinear combination of peak amplitude, rise time, and frequency band energy. If the estimated energy is less than 10μJ, a general maintenance warning instruction is generated, which is configured to generate a low-priority work order in the equipment maintenance system. When the energy estimate exceeds 100μJ, an emergency shutdown signal is immediately triggered, which directly disconnects the main power circuit of the distribution box via a hardwired connection. At the same time, the system continuously counts the number of spark events per unit time. When this number exceeds the frequency threshold dynamically calculated based on the real-time current load, a system maintenance request is automatically generated and uploaded to the operation and maintenance management platform.

[0068] Example 2

[0069] An embodiment of the present invention provides a computer-readable storage medium.

[0070] The computer-readable storage medium provided in the embodiment of the present invention stores a computer program, and when the computer program is executed by a processor, the steps of any one of the above-mentioned methods for monitoring electric sparks in a distribution box can be implemented.

[0071] The computer-readable storage medium may include: a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk, and other media that can store program codes.

[0072] For an introduction to the computer-readable storage medium provided in an embodiment of the present invention, please refer to the above method embodiment, and the present invention will not elaborate on it here.

[0073] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. Reference can be made to the common and similar parts between the various embodiments. For the devices disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the description is relatively simple, and the relevant parts can be referred to the method description.

[0074] Professionals may further appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of the two. In order to clearly illustrate the interchangeability of hardware and software, the above description has generally described the components and steps of each example according to their functions. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professionals and technicians may use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of the present invention.

[0075] The steps of the methods or algorithms described in conjunction with the embodiments disclosed herein may be implemented directly using hardware, a software module executed by a processor, or a combination of the two. The software module may be placed in random access memory (RAM), internal memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, a hard disk, a removable disk, a CD-ROM, or any other form of storage medium known in the art.

[0076] Example 3

[0077] An embodiment of the present invention provides an execution device.

[0078] Please refer to Figure 4 , Figure 4 This is a schematic diagram of the structure of an execution device provided by the present invention, which may include: Memory for storing computer programs; The processor is used to implement the steps of any of the above-mentioned methods for monitoring electric sparks in a distribution box when executing a computer program.

[0079] like Figure 4FIG2 is a schematic diagram of the structure of the execution device, which may include a processor 5, a memory 6, a communication interface 7, and a communication bus 8. The processor 5, the memory 6, and the communication interface 7 communicate with each other via the communication bus 8.

[0080] In the embodiment of the present invention, the processor 5 may be a central processing unit (CPU), an application specific integrated circuit, a digital signal processor, a field programmable gate array, or other programmable logic devices.

[0081] The processor 5 may call the program stored in the memory 6. Specifically, the processor 5 may execute the operations in the embodiment of the push button switch fault detection method.

[0082] The memory 6 is used to store one or more programs. The programs may include program codes, and the program codes include computer operating instructions. In the embodiment of the present invention, the memory 6 stores at least a program for implementing the following functions: Acquire the acoustic emission signal inside the distribution box; Acquiring characteristic data of the acoustic emission signal, the characteristic data comprising time domain characteristic parameters and frequency domain characteristic parameters extracted from the acoustic emission signal based on preset signal processing rules; Acquire environmental parameters of the distribution box, the environmental parameters including at least one of temperature, humidity, and current load, and The characteristic data of the acoustic emission signal and the environmental parameters are provided to an electric spark recognition model to perform electric spark state recognition.

[0083] In one possible implementation, the memory 6 may include a program storage area and a data storage area, wherein the program storage area may store an operating system and at least one application required for a function, etc.; the data storage area may store data created during use.

[0084] In addition, the memory 6 may include a high-speed random access memory and may also include a non-volatile memory, such as at least one disk storage device or other volatile solid-state storage device.

[0085] The communication interface 7 may be an interface of a communication module, used for connecting to other devices or systems.

[0086] Of course, it needs to be explained that Figure 4 The structure shown does not constitute a limitation on the execution device in the embodiment of the present invention. In actual applications, the execution device may include Figure 4 More or fewer components than shown, or combinations of certain components.

[0087] The embodiments described above are merely descriptions of preferred embodiments of the present invention and are not intended to limit the concept and scope of the present invention. Any modifications and improvements made to the technical solution of the present invention by a person of ordinary skill in the art without departing from the design concept of the present invention shall fall within the scope of protection of the present invention. The technical content for which protection is sought in the present invention is fully set forth in the claims.

Claims

1. A method for monitoring electric sparks in a distribution box, characterized in that: The distribution box electric spark monitoring method comprises: Acquire the acoustic emission signal inside the distribution box; Acquiring characteristic data of the acoustic emission signal, the characteristic data comprising time domain characteristic parameters and frequency domain characteristic parameters extracted from the acoustic emission signal based on preset signal processing rules; Obtaining environmental parameters of the distribution box, the environmental parameters including at least one of temperature, humidity, and current load, and The characteristic data of the acoustic emission signal and the environmental parameters are provided to an electric spark recognition model to perform electric spark state recognition.

2. The method for monitoring electric sparks in a distribution box according to claim 1, characterized in that: The time domain characteristic parameters include short-time zero-crossing rate, rise time and peak amplitude, wherein the calculation expression of the short-time zero-crossing rate is: ,in is the number of sampling points in the analysis window, is the discrete acoustic emission signal, is the sampling point number of the time series, is a symbolic function, ; The rise time is defined as the time required for the signal to rise from 10% of its peak amplitude to 90% of its peak amplitude. The peak amplitude is the maximum absolute amplitude value of the signal within the analysis window.

3. The method for monitoring electric sparks in a distribution box according to claim 2, characterized in that: The frequency domain characteristic parameters include spectrum kurtosis and energy proportion of 30-50kHz frequency band, wherein the calculation expression of the spectrum kurtosis is: ,in is the spectrum amplitude after discrete Fourier transform, is the frequency point number, is the spectrum mean, is the frequency index; The energy proportion of the 30-50kHz frequency band is defined as: , is the sampling frequency of the signal, Frequency The power spectral density is represents the Nyquist frequency.

4. The method for monitoring electric sparks in a distribution box according to claim 3, characterized in that: The electric spark recognition model performs state recognition through a multi-dimensional joint criterion, which is expressed as: in, is a dynamic threshold, which is adaptively generated based on the statistical characteristics of background noise. is the rise time upper threshold, is the lower threshold of spectrum kurtosis, is the frequency band energy ratio threshold, is the rise time.

5. The method for monitoring electric sparks in a distribution box according to claim 4, characterized in that: The dynamic threshold is generated by the following expression: in, is the moving average of the short-term zero-crossing rate in the time window, is the standard deviation of the short-time zero-crossing rate, are their respective weight coefficients; Weight coefficient , weight coefficient And satisfy the constraints .

6. The method for monitoring electric sparks in a distribution box according to claim 5, characterized in that: The weight coefficient and the weight coefficient Dynamic adjustment based on environmental parameters: in is the base weight, the default value is , is the temperature compensation coefficient, Moderate compensation coefficient are the attenuation factor and the gain factor respectively; The weight coefficient and the weight coefficient The process of dynamic adjustment according to environmental parameters is optimized by gradient descent algorithm: ,in The ideal threshold is deduced from historical missed detection / false alarm data. is the constraint penalty coefficient, .

7. The method for monitoring electric sparks in a distribution box according to claim 6, characterized in that: The environmental parameters are used to dynamically adjust the decision logic of the characteristic parameters: When the temperature exceeds the preset temperature threshold, the peak amplitude judgment threshold is reduced proportionally; When the humidity exceeds the preset humidity threshold, the weight coefficient of the frequency domain characteristic parameter is increased; When the current load exceeds the rated value, the frequency domain feature extraction below 20kHz in the acoustic emission signal is shielded.

8. The method for monitoring electric sparks in a distribution box according to claim 7, characterized in that: The spark status identification includes graded warnings: When an electric spark with energy lower than 10μJ is detected, a general maintenance warning is triggered; When an electric spark with energy higher than 100μJ is detected, an emergency shutdown warning is triggered; When the frequency of electric sparks in a unit time exceeds the threshold, a system maintenance warning is triggered.

9. A computing device, characterized in that include: at least one processor; as well as The memory stores instructions, and when the instructions are executed by the at least one processor, the at least one processor executes the distribution box electric spark monitoring method according to any one of claims 1 to 8.

10. A non-transitory machine-readable storage medium, characterized in that The machine stores executable instructions, which, when executed, enable the machine to perform the distribution box electric spark monitoring method according to any one of claims 1 to 8.

Images