A mobile power supply fault detection method and system based on intelligent diagnosis

Abstract

The application relates to the technical field of electric fault detection, in particular to a mobile power supply fault detection method and system based on intelligent diagnosis, which comprises the following steps: collecting voltage and current data of a mobile power supply under a normal operation state, generating an electric performance baseline data set by recording the time stamp and corresponding electric performance indexes of the data; and extracting time sequence characteristics of the voltage and the current based on the electric performance baseline data set to obtain electric performance characteristic data. In the application, the voltage and current data of the mobile power supply under the normal state are collected to establish a baseline data set covering the time stamp and the electric performance indexes, time sequence characteristics such as peak value, average value and standard deviation are extracted to capture detailed information of electric performance changes; meanwhile, frequency spectrum characteristics are accurately extracted by means of fast Fourier transform to master the intensity and phase information of frequency components, and then wavelet transform is adopted to decompose the signal to obtain more abundant electric performance abnormal information under multiple scales.

Description

Technical Field

[0001] This invention relates to the field of electrical fault detection technology, and in particular to a mobile power bank fault detection method and system based on intelligent diagnosis. Background Technology

[0002] Power bank fault detection methods refer to the detection and analysis methods used when portable energy storage devices such as power banks (portable power banks) experience abnormal charging, insufficient power, abnormal discharging, unstable output, or other functional faults.

[0003] Current technologies focus solely on diagnosing abnormal phenomena after a power bank malfunction, primarily relying on manual or simple instrument-based localized analysis. They lack in-depth feature mining of abnormal voltage and current characteristics, resulting in insufficient precision in data feature extraction and making it difficult to effectively detect subtle or early-stage anomalies. Furthermore, existing methods lack analysis of electrical performance characteristics from a frequency domain or multi-scale perspective, often failing to effectively handle hidden or intermittent abnormal fluctuations in signals, leading to the overlooking of potential problems and increasing the likelihood of missed or misdiagnosed faults. Therefore, improvements are needed. Summary of the Invention

[0004] The purpose of this invention is to address the shortcomings of existing technologies by proposing a mobile power bank fault detection method and system based on intelligent diagnostics.

[0005] To achieve the above objectives, the present invention adopts the following technical solution: a mobile power bank fault detection method based on intelligent diagnosis, comprising the following steps: Voltage and current data of the power bank under normal operating conditions are collected. By recording the timestamps of the data and the corresponding electrical performance indicators, an electrical performance baseline dataset is generated. Based on the electrical performance baseline dataset, the time series features of voltage and current are extracted to obtain electrical performance feature data. Based on the electrical performance characteristic data, the intensity and phase of each frequency component are analyzed using Fast Fourier Transform to extract key spectral features and generate spectral feature data. The spectral feature data is then decomposed into multi-scale data by wavelet transform, which deconstructs the signal into detailed and approximate data at multiple time scales to obtain multi-scale spectral data. By comparing the multi-scale spectral data with the corresponding multi-scale spectral features in the electrical performance baseline data, fluctuations and aperiodic anomalies deviating from the norm are identified, and preliminary fault diagnosis results are generated; geometric curvature analysis is performed on the preliminary fault diagnosis results to calculate the degree of geometric deformation between data points and obtain fault identification feature data; Based on the fault identification feature data, a fault prediction model is trained using a support vector machine. The fault prediction model is then applied to continuously monitor the power bank, resulting in fault monitoring and early warning results.

[0006] Preferably, the steps for obtaining the electrical performance baseline dataset are as follows: real-time monitoring and recording of the current and voltage data of the power bank, adding a timestamp to each data point to form a preliminary current and voltage time series record; Based on the preliminary current and voltage time series records, the instantaneous power, energy consumption, and efficiency of the current and voltage are calculated to obtain electrical performance index data. Based on the electrical performance index data, each index is integrated with a timestamp to construct an electrical performance baseline dataset.

[0007] Preferably, the steps for obtaining the electrical performance characteristic data are as follows: extracting voltage and current data from the electrical performance baseline dataset, calculating the rate of change between adjacent data points, smoothing the rate of change sequence, and simultaneously combining the extreme points of voltage and current data to calibrate the fluctuation amplitude and stable range of current and voltage, thereby generating processed voltage and current time series data. Based on the processed voltage and current time series data, the time series characteristic values ​​of voltage and current are calculated using the following formula: ; in, Represents the feature values ​​of the time series. For time points Voltage data, For time points Voltage data, For time points Current data, For time points Current data, The total number of data points in the time series. Time point Voltage data, For time points Current data; Based on the time series characteristic values, combined with the extreme values, average values ​​and trends of current and voltage, electrical performance characteristic data are obtained.

[0008] Preferably, the step of obtaining the spectral feature data is as follows: preprocessing the voltage and current time series in the electrical performance feature data, including removing noise and outliers, applying window function processing to reduce spectral leakage, and obtaining processed time series data; Based on the processed time series data, the comprehensive characteristics of each frequency are calculated using the following formula: ; in, Representing frequency The comprehensive characteristics, It is frequency The result of the fast Fourier transform at that point, yes conjugate, and Represent The real and imaginary parts, , indicating phase, It is an adjustment factor. It is a smoothing parameter; Based on the comprehensive features, key spectral features are selected and combined to generate spectral feature data.

[0009] Preferably, the step of acquiring the multi-scale spectral data is as follows: based on the spectral feature data, the wavelet coefficients at each scale are calculated using the following formula: ; in, Indicated in scale and location wavelet coefficients, It is the first of the spectrum data One point, It is the wavelet mother function. It is the total number of data points. Indicates time delay. These are Gaussian window parameters used to adjust the window width; Based on the wavelet coefficients, the signal is reconstructed to separate and identify details and approximations at each time scale, resulting in multi-scale spectral data.

[0010] Preferably, the steps for obtaining the preliminary fault diagnosis results are as follows: acquiring the multi-scale spectrum data and the electrical performance baseline dataset, standardizing the multi-scale spectrum data and the electrical performance baseline dataset to make the data format consistent, and obtaining standardized multi-scale spectrum data and electrical performance baseline dataset. Based on the standardized multi-scale spectral data and electrical performance baseline dataset, the spectral characteristics at each scale are compared to identify deviations in frequency response, energy distribution, and time series stability, thus obtaining deviation analysis results. Based on the deviation analysis results, each anomaly is evaluated and classified. The anomalies are sorted and marked according to the severity of their impact and the frequency of their occurrence. A predicted fault type is defined for each anomaly, and preliminary fault diagnosis results are obtained.

[0011] Preferably, the steps for obtaining the fault monitoring and early warning results are as follows: based on the fault identification feature data and multi-scale spectrum data, configure and train a support vector machine model, select a radial basis function and adjust the model penalty coefficient and kernel parameters to obtain a trained fault prediction model. Based on the trained fault prediction model, the trained fault prediction model is deployed to the mobile power supply monitoring system to analyze the collected power operation data in real time, assess fault risks, and obtain fault monitoring and early warning results.

[0012] This invention provides a mobile power bank fault detection system, comprising: The data acquisition module records the voltage and current of the power bank during operation, adds timestamps, and generates an electrical performance baseline dataset. The feature extraction module extracts the peak value, mean value, and standard deviation of voltage and current from the electrical performance baseline dataset to generate electrical performance feature data; The spectrum analysis module applies Fast Fourier Transform to the electrical performance characteristic data to analyze the intensity and phase of each frequency component, generating spectrum characteristic data. The spectrum characteristic data is then decomposed into multi-scale spectrum data through wavelet transform. The fault diagnosis module compares multi-scale spectral data with electrical performance baseline datasets, identifies fluctuations and non-periodic anomalies, generates preliminary fault diagnosis results, performs geometric curvature analysis on the preliminary fault diagnosis results, calculates the degree of deformation between data points, and obtains fault characteristic data. The fault prediction module trains a support vector machine model based on fault feature data, performs continuous monitoring, predicts and issues fault monitoring warnings, and generates fault warning results.

[0013] Compared with the prior art, the advantages and positive effects of the present invention are as follows: In this invention, based on the voltage and current data collection of a power bank under normal conditions, a baseline dataset covering timestamps and electrical performance indicators is established. By extracting time-series features such as peak value, mean, and standard deviation, detailed information on changes in electrical performance is captured. Simultaneously, fast Fourier transform is used to accurately extract spectral features, grasping the intensity and phase information of frequency components. Then, wavelet transform is used to decompose the signal, obtaining richer information on electrical performance anomalies at multiple scales. By comparing multi-scale data with the baseline state, non-periodic fault fluctuations deviating from the normal state are accurately identified, improving the sensitivity and accuracy of fault location. Geometric curvature analysis is then used to further refine and quantify the degree of data anomalies, uncovering more subtle abnormal feature changes and enhancing the detection capability of hidden faults inside the power bank. At the same time, fault prediction and real-time monitoring are achieved based on support vector machines, enabling early fault warning, reducing safety hazards caused by delayed fault diagnosis, and improving the overall service life and operational stability of the power bank. Attached Figure Description

[0014] Figure 1 This is a schematic diagram of the steps of the present invention. Detailed Implementation

[0015] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0016] Please see Figure 1 This invention provides a technical solution: a mobile power bank fault detection method based on intelligent diagnostics, comprising the following steps: The voltage and current data of the power bank under normal operating conditions are collected. By recording the timestamps of the data and the corresponding electrical performance indicators, an electrical performance baseline dataset is generated. Based on the electrical performance baseline dataset, the time series features of voltage and current are extracted to obtain electrical performance feature data. Based on electrical performance characteristic data, the intensity and phase of each frequency component are analyzed using fast Fourier transform to extract key spectral features and generate spectral feature data. The spectral feature data is then decomposed into multi-scale data by wavelet transform, which deconstructs the signal into detailed and approximate data at multiple time scales to obtain multi-scale spectral data. By comparing the multi-scale spectral data with the corresponding multi-scale spectral features in the electrical performance baseline data, fluctuations and aperiodic anomalies that deviate from the norm are identified, and preliminary fault diagnosis results are generated. Geometric curvature analysis is performed on the preliminary fault diagnosis results to calculate the degree of geometric deformation between data points and obtain fault identification feature data. Based on fault identification feature data, a fault prediction model is trained using a support vector machine. The fault prediction model is then applied to continuously monitor the power bank, resulting in fault monitoring and early warning results.

[0017] The steps for obtaining the electrical performance baseline dataset are as follows: monitor and record the current and voltage data of the power bank in real time, add a timestamp to each data point, and form a preliminary current and voltage time series record. Based on preliminary current and voltage time series records, the instantaneous power, energy consumption, and efficiency of current and voltage are calculated to obtain electrical performance index data. Based on electrical performance index data, each index is integrated with a timestamp to construct an electrical performance baseline dataset.

[0018] Specifically, current and voltage information is acquired using the data acquisition device installed on the power bank. A time stamp corresponding to the recording time is added with each recording. First, a current reference range of 0A to 5A and a voltage reference range of 0V to 24V are set. If a recorded measurement value exceeds this current or voltage range, the record is marked as an abnormal measurement and temporarily excluded from analysis. These range values ​​can be set through preliminary testing or empirical values. For example, a typical power bank sample is used for continuous measurement to statistically analyze the minimum and maximum current and voltage values. Then, the range is expanded by a certain proportion at both ends to obtain the currently used range values. The system records the real-time status of the power bank once per second and associates the actual collected data with the corresponding time markers. During the association process, the collected current and voltage measurement points are stored as sequential data according to the time sequence. If the time interval between adjacent samplings exceeds the pre-defined 1-second error range (this range can be determined by internal clock accuracy verification, for example, by setting ±0.05 seconds as the allowable deviation), the system marks the sampling as a time-series fluctuation and continues recording. Finally, after summarizing several sampling records, the current and voltage information are arranged in chronological order to obtain a preliminary current and voltage time series record.

[0019] Based on preliminary current and voltage time series records, the current and voltage data at each moment are calculated to obtain the instantaneous power. During the statistical phase, the corresponding energy consumption and efficiency are extrapolated by combining the cumulative duration of the recording intervals. This is first achieved through... Calculate instantaneous power, where The voltage value at that moment, To determine the reasonable power fluctuation range, the current value at that moment is used as a reference. First, multiple measurement data points under typical operating conditions are selected, and the average power is calculated. Then, a temporary power threshold is defined by adding 20% ​​above and below this average value. If the calculated instantaneous power exceeds this range, it is considered a possible anomaly and temporarily stored for later use. Subsequently, in the energy consumption calculation, the power at each moment is multiplied by the time interval to obtain the corresponding energy increment. The total energy consumption is obtained by summing all increments. For efficiency measurement, a reference efficiency benchmark can be set according to existing power bank specifications, such as 85%. Then, the calculated energy output is compared with the energy input to this benchmark. If there is a significant deviation, it is marked and subsequently checked to obtain the final electrical performance index data.

[0020] Based on electrical performance index data, the target values ​​are combined with the recording time. First, in the aggregation stage, the power, energy consumption, and efficiency of each record are time-series labeled. Then, the labeled data are organized in order. If duplicate or missing time labels are found for some records, they are corrected or supplemented at this stage. To determine whether the corrected data remains consistent, a time continuity judgment index can be introduced. The time difference between adjacent records is compared with a set allowable deviation threshold, for example, the allowable deviation threshold is defined as ±0.1 seconds. If it exceeds the threshold, it is recorded separately and excluded in subsequent analysis. After completing this step, the power and efficiency data are checked again by comparing with the previously determined power threshold and efficiency benchmark value. If there is a large range of deviation, these records are placed in the abnormal archive area and removed during the final integration. Finally, an electrical performance baseline dataset is constructed based on the obtained time and the valid records corresponding to each index.

[0021] The steps for obtaining electrical performance characteristic data are as follows: extract voltage and current data from the electrical performance baseline dataset, calculate the rate of change between adjacent data points, smooth the rate of change sequence, and combine the extreme points of voltage and current data to calibrate the fluctuation amplitude and stable range of current and voltage, thereby generating processed voltage and current time series data. Based on the processed voltage and current time series data, the time series characteristic values ​​of voltage and current are calculated using the following formula: ; in, Represents the feature values ​​of the time series. For time points Voltage data, For time points Voltage data, For time points Current data, For time points Current data, The total number of data points in the time series. Time point Voltage data, For time points Current data; Based on time series characteristic values, combined with the extreme values, average values ​​and trends of current and voltage, electrical performance characteristic data are obtained.

[0022] Specifically, voltage and current data are extracted from the electrical performance baseline dataset. The voltage and current values ​​at each sampling point are compared sequentially with adjacent acquisition times, and their rate of change is calculated. The rate of change can be calculated by first counting the difference between every two consecutive points and then dividing by the time interval. To avoid high-frequency jitter, an empirical smoothing threshold can be set, and multiple sliding average operations can be performed. For example, the rate of change values ​​in each batch of 10 can be averaged, and the average result is used as the smoothed output. Each smoothing operation compares the data to a pre-defined allowable range, such as a voltage range of 0V to 24V and a current range of 0A to 5A. If the local rate of change continuously deviates from this range, it is determined whether to remove that segment of data based on the actual recording situation. Then, the maximum and minimum voltage and current values ​​are labeled. By observing the range between high and low values, the fluctuation amplitude can be divided into several levels. For example, 0V to 8V can be used as low amplitude, 8V to 16V as medium amplitude, and 16V to 24V as high amplitude. Similarly, the current can be divided into segments such as 0A to 1.5A, 1.5A to 3.0A, and 3.0A to 5.0A. In this process, the stable range is identified by continuously comparing the fluctuation trends of voltage and current. Specifically, if the adjacent smoothed rate of change is at a relatively small value, it is marked as a relatively stable region. The data is continuously recorded and summarized at different time periods. Finally, all the smoothed rate of change data, the corresponding extreme value information, and the stable range markers are summarized to generate processed voltage and current time series data.

[0023] The advantage of the formula is that it comprehensively considers the difference between voltage and current changes as well as the absolute measure of the product of voltage and current, so that the time series characteristic values ​​not only reflect the degree of fluctuation of electrical performance over time, but also encompass the magnitude of the power level.

[0024] The steps to obtain it are as follows: Indicates the first Voltage data at specific time points, obtained from actual measurements at the external port of the power bank, can be acquired using a voltage acquisition device with a range of 0V to 30V and a resolution of 0.01V. When the measured voltage at a certain point is 12.12V and the calibration deviation is 0.02V, the voltage is calculated as follows: V.

[0025] The steps to obtain it are as follows: Representative and If the voltage data from the preceding moment is 11.96V at the 3rd second, then the voltage measured at the 2nd second is considered as... If the voltage at that moment is 11.90V after calibration, then It can be substituted into the difference operation to finally form The voltage difference.

[0026] The steps to obtain it are as follows: Used to indicate the first The current values ​​collected at each time point are obtained using a current sensor with a rated range of 0A to 5A and an accuracy of 0.001A. When the device measures a current of 3.002A at the 5th second and the deviation is calibrated to -0.002A, then the current is obtained. A.

[0027] The steps to obtain it are as follows: That is to The current value at the previous adjacent moment in the sampling sequence is recorded as 2.980A after calibration at the 4th second. Subtracting it from the 3.000A at the 5th second yields a difference of 0.020A.

[0028] The steps to obtain it are as follows: The number of difference or product operations in a time series is directly related to the number of valid data points collected. If a total of [number] valid data points are obtained during the monitoring period... If the actual voltage and current data are obtained, then... or Total time available If 10 valid data points are acquired within one complete acquisition cycle using the second difference calculation, then .

[0029] and The steps to obtain it are as follows: and They represent the first time. Voltage and current data at various time points.

[0030] Calculation process: The sampling frequency was set to 1Hz, and a total of 4 valid data points were collected, i.e. and The voltage data measured and calibrated at each moment are sequentially... V. V. V. V, the corresponding current data are as follows A, A, A, A. First, calculate the voltage difference and then average it: ; Here This refers to the starting time of this sequence. If the voltage value obtained before the start of monitoring is recorded as 12.00V, then: ; Then calculate the sum and average of the current differences: ; If the starting current data at second 0 is recorded as 2.000A, then: ; Then calculate the voltage-current product term: ; ; Substituting the above results into the formula, we get: ; The results show that within the selected sampling interval, the combined effect of the difference between voltage and current and the product level are ultimately converted into a characteristic value of approximately 5.06. When used in conjunction with subsequent electrical performance analysis tools, this value can be used to determine the relative relationship between fluctuation amplitude and typical power. When this value is too large, it indicates that the fluctuation is large and the product is high. When the value is relatively small, it indicates that the fluctuation is relatively smooth and the power is low.

[0031] Based on time series characteristic values, combined with the extreme values, average values, and trends of current and voltage, the previously obtained characteristic value sequences are compared and organized. Initially, the maximum and minimum values ​​are selected from the collected characteristic values, and their corresponding times are checked to determine when the current or voltage reaches its highest and lowest points. Then, the trend of characteristic value changes is continuously observed. If it is found that characteristic values ​​at multiple times are concentrated in a relatively narrow numerical range, such as comparing characteristic values ​​in the 0-10, 10-20, and 20-30 ranges with known reference ranges and marking the cumulative number of times each recording point falls within which range, the stability benchmark set before monitoring is consulted. For example, if in the initial sampling, the characteristic values ​​of general equipment operation frequently fall within the 10-15 range, then the 10-15 range can be used as a reference range. Five intervals are recorded as an empirical interval. If it is detected that most values ​​in the current sequence are concentrated in this empirical interval, it indicates that the overall state remains stable. Then, the continuous trend of the characteristic values ​​is observed in chronological order. A threshold for trend judgment is set based on statistical calculations of past data. For example, in industrial production, the maximum and minimum characteristic values ​​in the past 50 recorded maintenance data can be averaged and then combined with their fluctuation ratios to determine a judgment line for distinguishing between upward and downward trends. For example, the average value plus or minus 1.5 can be set as a dividing line. When the latest characteristic value is greater than the average value plus 1.5, it is considered to be rising; when it is less than the average value minus 1.5, it is considered to be falling. Finally, based on these filtering and differentiation actions, the characteristic values ​​are correlated with time and current and voltage data to obtain electrical performance characteristic data.

[0032] The steps for obtaining spectral feature data are as follows: preprocessing the voltage and current time series in the electrical performance feature data, including removing noise and outliers, applying window functions to reduce spectral leakage, and obtaining the processed time series data; Based on the processed time series data, the comprehensive characteristics of each frequency are calculated using the following formula: ; in, Representing frequency The comprehensive characteristics, It is frequency The result of the fast Fourier transform at that point, yes conjugate, and Represent The real and imaginary parts, , indicating phase, It is an adjustment factor. It is a smoothing parameter; Based on comprehensive features, key spectral features are selected and combined to generate spectral feature data.

[0033] Specifically, the voltage and current time series data in the electrical performance characteristic data undergo preprocessing. First, each record is checked for time misalignment against the sampling frequency. If a record interval exceeds 0.05 seconds, it is marked as a timing anomaly and subsequently removed. Then, measured values ​​are compared one by one within the voltage range of 0V to 24V and the current range of 0A to 5A. Records outside this range are considered invalid measurements and excluded. To remove noise, the statistical results of the largest amplitude disturbance in the previously stored 100 operation records are retrieved. A fixed increment, such as 0.02V or 0.01A, is added to the mean of these values ​​as a noise removal threshold. The difference between the current sample and this threshold is monitored. If the difference exceeds the allowable deviation, the point is determined to be a noise point and deleted. Afterward... The remaining data needs to be processed using a window function. The window function length is set to 64 beforehand because it was found that this length provides good smoothness for power spectrum estimation when comparing multiple measurement results. Some overlap can be retained between adjacent segments to maintain transition. The same window function is applied to each segment, and the entire time series is cut and spliced ​​in a sequential manner. Next, the amplitude performance of voltage and current is checked segment by segment in the processing stage. If the amplitude difference of a segment exceeds a fixed threshold, such as 2V or 0.5A, it is recorded as a strong fluctuation segment and further investigated in subsequent steps. If the difference does not reach the threshold, it is directly retained. After completing all the above steps, the updated sequence data is integrated and output, which is the processed time series data.

[0034] The advantage of the formula lies in its comprehensive use of amplitude information, logarithmic amplitude information, and characteristic quantities of the larger real and imaginary parts, and in considering the influence of phase on the overall characteristics through an exponential factor, thus taking into account multiple aspects of amplitude and phase information in the frequency analysis stage.

[0035] The steps to obtain it are as follows: It is a non-negative number, mainly used to measure the proportion of amplitude in the overall characteristics. It needs to be determined based on actual operating data. First, during 120 minutes of continuous operation, perform FFT on multiple time series to obtain the amplitude range. Then, calculate the standard deviation of the amplitude distribution and compare it with the mean. Next, refer to the amplitude statistics generated from the typical operating voltage and current listed in the product manual. Summarize these data into a set of fluctuation metrics, denoted as [value missing]. After calculating the average of these volatility measures Next, a coefficient is selected to reflect the importance of the magnitude term, thereby determining... For example, Set as In the form of, a final actual measurement is performed and the stable value is recorded, such as when and When, can be calculated .

[0036] The steps to obtain it are as follows: The coefficients mainly control the logarithmic magnitude term. To determine the weight of the overall calculation result, it is necessary to measure the order-of-magnitude distribution of the amplitude values ​​across multiple execution phases. To quantify the order-of-magnitude range of the amplitude values, each FFT amplitude value can be converted to its natural logarithmic form, and its minimum and maximum logarithmic amplitudes can be recorded. If the logarithmic amplitude distribution spans a large range, then... The difference should be appropriately increased to emphasize its variability; if the distribution is relatively concentrated, then... It can be relatively smaller, which can make ,in This represents the range of the logarithmic amplitude within the measurement period. This represents the sum of the total amplitude ranges of voltage and current. Finally, this initial value is compared with the statistical data from multiple experiments to correct it to a constant value, for example, by recording the results after multiple tests. , ,but .

[0037] The steps to obtain it are as follows: To balance the contribution of the maximum value terms of the real and imaginary parts to the overall feature, it is necessary to first retrieve a batch of monitoring results. and The maximum achievable range is determined by selecting a significant portion and summarizing it, then calculating the average range between the two ranges. ,Will Set as ,here This represents the maximum range of the real or imaginary part across all observations. To obtain a fixed, usable value, it needs to be determined after multiple actual measurements. This value can be taken as a range within a certain range, such as 0.15 or 0.2, and then a secondary correction is performed based on precise data. For example, when... , At times .

[0038] The steps to obtain it are as follows: Indicates frequency The FFT result at this point will necessarily be a complex number, which can be obtained by performing a Discrete Fourier Transform on the processed time series. The specific procedure is to first determine the sampling length. Then perform the FFT operation, for the The complex number of the frequency output is denoted as .

[0039] The steps to obtain it are as follows: yes The complex conjugate, if ,but ,in and This parameter is retrieved from the complex portion of the aforementioned FFT output. To obtain this parameter, it is only necessary to find a match within the same sequence. The corresponding complex numbers are joined by taking the real part unchanged and the imaginary part negative. In actual programs, this can be done by... The real and imaginary parts are stored separately, and then the sign of the imaginary part is changed directly in the code to form a new quantity. .

[0040] The steps to obtain it are as follows: yes The real part of the complex number needs to be extracted from the complex number output by the FFT. The specific process is to first record the result of the FFT operation, then read the real part value from the corresponding index and convert it into a floating-point form that can be used for numerical analysis. In order to obtain a reliable real part range, the original signal can be averaged and filtered before the FFT to reduce the DC offset before recording the real part value. Finally, the input item can be formed to participate in the formula calculation along with other parameters.

[0041] The steps to obtain it are as follows: for The imaginary component, similar to the real component extraction, is directly read from the imaginary field of the FFT result. If a significant phase shift occurs during the measurement period, large positive or negative values ​​may be observed in the imaginary component. Therefore, this anomaly is checked during post-processing. If no anomaly is found, the data is recorded according to the normal procedure. And incorporate it into each calculation.

[0042] The steps to obtain it are as follows: , which is the phase value at the current frequency, requires the previously obtained... and The values ​​are placed into the arctangent function in a specific order for computation. For cases where the value is too small or close to 0, a logical check will be added at the code level to avoid exceptions when the denominator is close to 0. After obtaining the phase, it can be stored in a phase list, and finally each... The corresponding phase values ​​are then substituted into the subsequent calculation process.

[0043] The steps to obtain it are as follows: The parameter representing the phase effect can be quantitatively evaluated based on the severity of phase changes at different frequencies. First, the range of phase variation with frequency is recorded over a relatively long period, and its mean square error is calculated. This mean square error is denoted as... Then let ,in This is a constant used to adjust the decay rate, which can be selected by referring to the results of multiple spectral analyses. A larger value indicates significant phase dispersion, and the value can be appropriately increased. With the smoothing phase effect, if the measurement is obtained ,and ,but .

[0044] Calculation process: Set at a certain frequency FFT complex values ​​obtained at Hz for ,but Therefore, we can obtain , Determined through the aforementioned methods , , , Then calculate the phase. radian.

[0045] First, calculate the amplitude. , Then the logarithmic magnitude is obtained. To simplify, this section will... Substitute logarithms into the calculation. The absolute value remains 0.84. Recalculate , Substitute these results into the parenthetical terms of the formula: ; Phase correction factor ; then The results indicate that at 50Hz, this comprehensive eigenvalue is approximately 0.886. A higher value indicates more significant amplitude, logarithmic amplitude, and real and imaginary characteristics at that frequency, while the phase influence is relatively smaller. At other frequencies... Repeat the above steps to obtain the entire frequency band. distributed.

[0046] After calculating the comprehensive characteristic values ​​at different frequencies based on the comprehensive features, it is necessary to check whether there are some frequencies whose corresponding characteristic values ​​are significantly higher than their adjacent frequency bands across the entire frequency band. It is also necessary to verify whether the actual voltage and current amplitudes or phases corresponding to these high-value points are above the previously defined anomaly judgment threshold. The threshold setting can be combined with the range of characteristic values ​​recorded in 30 consecutive tests to select an out-of-limit standard. For example, first calculate the mean and standard deviation of all characteristic values, and then add twice the standard deviation to the mean to set the critical standard. If the characteristic value at a certain frequency exceeds this standard, it is considered a suspicious frequency point. Then, confirm whether these suspicious points are concentrated at each frequency. If the number of high values ​​concentrated in a frequency band exceeds a certain counting threshold, the frequency band is determined to be a critical area. At the same time, it is necessary to further screen based on the fluctuation of characteristic values ​​after phase correction. If most of the high values ​​are concentrated on the side with smaller or larger phases, they can be classified separately and marked as a group with obvious phase tendencies. Finally, this type of characteristic value is separated from other normal interval values, and the most critical frequency points are retained according to the set screening strategy. The characteristics corresponding to these high-value points are recorded together to form key spectral characteristic data. After summarizing, the spectral characteristic data is obtained.

[0047] The steps for acquiring multi-scale spectral data are as follows: Based on the spectral feature data, calculate the wavelet coefficients at each scale using the following formula: ; in, Indicated in scale and location wavelet coefficients, It is the first of the spectrum data One point, It is the wavelet mother function. It is the total number of data points. Indicates time delay. These are Gaussian window parameters used to adjust the window width; Based on wavelet coefficients, the signal is reconstructed to separate and identify details and approximations at various time scales, resulting in multi-scale spectral data.

[0048] Specifically, the advantage of the formula lies in the wavelet mother function. In addition, a Gaussian window was introduced. By adjusting local regions, the multi-scale analytical characteristics of wavelet transform in terms of time and frequency can be reflected, while also taking into account local smoothness. This allows for a better balance between details and overall trends in the multi-scale analysis of spectral data.

[0049] The steps to obtain it are as follows: This is a scale parameter characterizing the scaling ratio; a larger value indicates a more pronounced stretching, meaning the wavelet function covers a wider range. In practice, the length of the previously obtained spectral data can be estimated first, for example, determining the number of discrete points collected within the 0Hz to 500Hz range. Then, a representative set of scales can be selected, such as... For example, during a review of industrial testing, it was found that when the data was primarily concentrated between 30Hz and 120Hz, using... and It can capture local details well, and then by converting the overall waveform peak and valley range recorded in the initial test table to a certain multiple, the corresponding value can be determined. List and input the results into the calculation.

[0050] The steps to obtain it are as follows: This represents the selected time (or position) offset on the signal sequence, requiring a point-by-point traversal of the entire sequence from beginning to end to locate local structures at different times. The value range is from 0 to ,in This represents the total number of data points. Each... For a local alignment point, it will make In The terms produce a relative displacement. In order to... The values ​​are set logically, and can be combined with previously obtained time or frequency indices to scan each index point. For example, if the data length... That will allow We select an integer value from 0 to 127 to cover the entire interval, thereby capturing the wavelet matching degree at different positions. For example, we can first... Substituting into the first calculation yields , and then beg Continue in sequence until... Ultimately, this forms a view of the entire signal at a certain scale. A comprehensive description.

[0051] The steps to obtain it are as follows: Is the spectrum data in the first The values ​​of each discrete point are extracted from the previous frequency domain analysis or FFT results. Each corresponds to a specific amplitude or some characteristic metric. Obtaining this data requires first performing an FFT on the original voltage and current information to obtain characteristic values ​​at several frequency points, which are then organized into... .

[0052] The steps to obtain it are as follows: It is the wavelet mother function. This represents a dimensionless variable obtained by combining sequence index with scale and position. When obtaining this variable, it is necessary to first determine which wavelet mother function to use, and then... Substitute the corresponding analytical expression. If Morlet wavelet is selected, it can be set as follows: One type of form can also use the Mexican hat wavelet or other types, but each generating function has its own analytical expression. During the calculation, first... and Give The specific value is then sent to... Complete the calculation. To avoid unnecessary truncation errors, you can... Values ​​are pre-calculated and stored in a table, then retrieved by index. For example, when using Morlet wavelets, for... exist to Discretize the interval step by step, and obtain the result by querying the table or interpolation. .

[0053] The steps to obtain it are as follows: This refers to the total number of data points, or the length of the spectral data used in the wavelet transform calculation. The specific value can be directly determined by statistically analyzing the previously generated spectral feature sequence. If a sampling length of 1024 was initially selected in the FFT stage, after truncating the useful frequency points, there might still be 512 valid points remaining. It is written as 512.

[0054] The steps to obtain it are as follows: This is the term in the formula used to characterize the time delay (or displacement) relationship; it is the index of the current count. With location index The result is obtained by subtraction. The acquisition process occurs within a program loop. Traversing from 0 to , Scan and calculate while fixed at the current position, when When changes occur, It will also change to ensure that the calculated result is accurate each time. Corresponding to different relative translations. If Negative numbers can still be substituted into the wavelet mother function for calculation; therefore, a sufficient domain needs to be defined to ensure that the wavelet mother function can still be correctly evaluated in the negative region. A specific example is shown below. , Sometimes, When performing interpolation or table queries, ensure that the index at index -3 is correct. The value has a valid definition.

[0055] The steps to obtain it are as follows: The parameters of the Gaussian window used to adjust the window width are as follows: the larger the value, the better the window function. Slower decay results in a wider coverage area, while a smaller coverage area leads to faster decay and greater limitations in the time domain. A "desired smoothness" can be calculated by combining statistics on the actual signal distribution and mapped to... The value is determined by several factors. For example, first, high-frequency noise analysis is performed on hundreds of industrial measurement samples to quantify the jitter in the high-frequency band. Then, this value is compared with the variance of the center frequency to obtain an average attenuation range, and finally, the value is determined. Alternatively, the values ​​can be fixed to representative values ​​within a range such as 1.0 or 2.0, and then the optimal values ​​can be determined through repeated trials in real-world scenarios. This results in a smoother transformation. The example, if summarized after multiple tests... It meets the requirements.

[0056] Calculation process: In a specific example, take ,Right now Five frequency points of data are used for calculation, making , , , , Let the scale ,Location Morlet wavelet mother function is selected. , Gaussian window parameters According to the formula: ; First, perform step-by-step calculations for each item: when hour, , ,then , Therefore, this contribution is approximately .

[0057] when hour, , , , Contribution .

[0058] when hour, , , , Contribution .

[0059] when hour, , , , Contribution .

[0060] when hour, , , , Contribution .

[0061] The sum of these contributions is obtained by adding them together: Multiply by ,have to .

[0062] This result indicates that at this wavelet scale With position Under these conditions, the weighted response of the signal to the Morlet generating function and the Gaussian window is approximately -0.236. If this value forms a distribution map after scanning at various locations, it can be further determined at which locations the local structure of the signal is most significant or weakest. When the value is greater than a certain threshold, it can be considered that the location contains a relatively obvious feature pattern, and when it is less than another threshold, it can be considered as stationary or without significant structure.

[0063] Based on wavelet coefficients, it is necessary to stack or assign corresponding coefficient values ​​to the appropriate positions point by point at each scale and reconstruct all obtained wavelet coefficient matrices layer by layer. For this purpose, the wavelet coefficients are first arranged from left to right within the same scale, and a judgment is made on whether the coefficient changes at adjacent positions exceed a specified range. A range detection method can be set here, such as obtaining the average and standard deviation of coefficients at the same scale in the last ten measurements beforehand, and then adding twice the standard deviation to the average as the upper limit. If the current coefficient significantly exceeds the upper limit, it is positioned as an abnormal coefficient and recorded in a separate list. The same judgment is then applied to other positions. After screening at this scale is completed, the same method can be used to process the next scale, merging all coefficients into a unified structure across multiple scales. Each scale has its own corresponding coefficient distribution graph, which is then analyzed during the reconstruction process. By arranging the detail layer and the approximation layer separately, it is possible to distinguish special fluctuation regions that may exist in different time and frequency regions. Then, the coefficient distribution at each scale is compared on the same time axis. If frequency bands with excessively large coefficients appear at multiple scales, it indicates that there are more obvious differences at that location. To avoid missing any cases, it is also necessary to check the positions of coefficients that are close to the edge but have not yet exceeded the limit. These coefficients are accumulated point by point to identify groups of coefficients that have a certain amplitude locally but have not reached the anomaly judgment threshold. If such groups appear continuously in time, they can be summarized into a weaker detail component. In this way, time-frequency structure diagrams corresponding to different resolutions are finally formed. Then, the detail information of several key segments is read from these data and merged into a complete spectrum reconstruction path so that the performance at different time scales can be merged in subsequent analysis to obtain multi-scale spectrum data.

[0064] The steps for obtaining preliminary fault diagnosis results are as follows: acquire multi-scale spectrum data and electrical performance baseline dataset, standardize the multi-scale spectrum data and electrical performance baseline dataset to make the data format consistent, and obtain standardized multi-scale spectrum data and electrical performance baseline dataset. Based on standardized multi-scale spectral data and electrical performance baseline datasets, the spectral characteristics at each scale are compared to identify deviations in frequency response, energy distribution, and time series stability, thus obtaining deviation analysis results. Based on the deviation analysis results, each anomaly is evaluated and classified. The anomalies are sorted and marked according to the severity of their impact and the frequency of their occurrence. A predicted fault type is defined for each anomaly, and preliminary fault diagnosis results are obtained.

[0065] Specifically, after acquiring multi-scale spectrum data and electrical performance baseline datasets, the recording methods of the two are compared, and differences in acquisition frequency and sampling length are checked. The index frequencies in the multi-scale spectrum data are mapped to the time or voltage / current indicators in the electrical performance baseline dataset. If it is found that the index frequencies and time axes cannot be matched one-to-one, a unified indexing system needs to be introduced. Multiple records are uniformly labeled using a preset reference sequence. Repeated or skipped moments are checked in the records and filled in or removed. An alignment accuracy threshold is then set according to the previous operating conditions of the equipment or the application scenario. For example, the accuracy threshold can be determined by multiplying the average interval obtained by dividing the total recording time by the number of sampling points by a ratio. For example, when the average interval is 1 second and the ratio is 0.05, the accuracy threshold is 0.05 seconds. Next, all recording points in both types of data are traversed, and the alignment deviation is checked to see if it is less than the accuracy threshold. If the deviation exceeds the threshold, the record is marked as an abnormal index and excluded. Subsequently, numerical normalization operations are performed on the parameter columns in the multi-scale spectrum data and electrical performance baseline datasets respectively. Typically, the minimum and maximum values ​​of each column are calculated first, and then... The data is converted to the 0 to 1 range. If some columns contain negative values, an offset is required before normalization to correct the minimum value to 0 so that the normalized result will not have underflow or invalid results. After normalization, the field order of the two sets of data is checked to ensure that the frequency items in the multi-scale spectrum data correspond one-to-one with the time period or corresponding measurement number in the electrical performance baseline dataset. If a missing match is detected in some segments, null values ​​are added and marked as missing. After the entire process is completed, a set of standardized multi-scale spectrum data and electrical performance baseline dataset can be obtained.

[0066] Based on standardized multi-scale spectral data and electrical performance baseline datasets, the spectral characteristic values ​​within each scale range are traversed, and these characteristic values ​​are compared with the voltage and current records for the corresponding time periods in the baseline. Considering the variations in frequency response and energy distribution, the multi-scale spectral data is first divided into several segments according to frequency bands. The characteristic values ​​within each segment are statistically analyzed, and it is checked whether these statistical values ​​are within the previously determined baseline interval. For example, if the baseline data shows that the main energy of most devices is concentrated in the 20Hz to 80Hz range, the average spectral amplitude in this range can be compared with the average amplitude of the baseline to see if the deviation is greater than twice the root mean square error. If it is, this segment is recorded as a suspicious band, and the process continues. Synchronously compare electrical performance information for the same period to check if the voltage exceeds a threshold calculated based on historical data. For example, define the voltage threshold within a range extending 5% from both ends of the equipment's rated value, or similarly extend the current threshold. If multiple consecutive exceedances of these thresholds are detected, a deviation warning is issued. By summarizing the results, it can be found at which scales and frequency bands there are significant deviations. Then, check the stability of the time series. When comparing the fluctuations of characteristic values ​​in adjacent periods, set a fluctuation limit, such as adding or subtracting a certain number of standard deviations from the average of the existing baseline, to determine whether the fluctuation is significant. Through this series of comparison operations, the deviation results corresponding to each scale can be obtained, forming the deviation analysis results.

[0067] Based on the deviation analysis results, each detected abnormal event is listed and numbered individually. First, the occurrence time of all anomalies is correlated with the multi-scale spectral characteristic values ​​at the same time to identify whether there are simultaneous abnormal high points or continuous deviations in frequency distribution or energy amplitude when these anomalies occur. Then, these are compared with the operating parameters specified in the electrical performance baseline dataset to check whether the anomaly corresponds to excessive voltage, excessive current fluctuation, or other indicators exceeding preset limits. These preset limits are typically calculated by combining the average value of statistical equipment during stable operation with the maximum fluctuation amplitude, such as over the past hundred... The highest voltage recorded in this measurement is 22V. Therefore, 23V can be set as an upper limit for severity judgment, 1V higher than this value. The frequency of occurrence and the scope of impact are taken into consideration. Anomalies are classified as having a greater impact or occurring more frequently and placed in a priority position, forming an anomaly list sorted by severity. For each anomaly, an estimated fault label can be given, such as identifying it as line instability, power supply aging, or short-term load overload. This list allows for the categorization of each anomaly. Finally, all anomalies and their classification information are integrated into the corresponding fault type to obtain preliminary fault diagnosis results.

[0068] The steps for obtaining fault monitoring and early warning results are as follows: Based on fault identification feature data and multi-scale spectrum data, configure and train a support vector machine model, select radial basis function and adjust model penalty coefficient and kernel parameters to obtain a trained fault prediction model; Based on the trained fault prediction model, the trained fault prediction model is deployed into the mobile power bank monitoring system to analyze the collected power operation data in real time, assess fault risks, and obtain fault monitoring and early warning results.

[0069] Specifically, based on fault identification feature data and multi-scale spectrum data, each corresponding record in both is first read, and the parameters such as voltage, current, and previously obtained fault type labels contained in the record are compared. All data are then organized into a sample set that can be used for support vector machine training. To determine the values ​​of the penalty coefficient and kernel parameter, candidate values ​​are selected within a limited grid range and tested one by one. For example, the penalty coefficient is increased from 0.1 to multiple intervals of 10, and the kernel parameter is increased from 0.01 to multiple intervals of 1. Subsequently, cross-validation is performed on each combination of penalty coefficient and kernel parameter, and the accuracy and misclassification rate of each category in the training set are calculated. These statistical indicators are compared with the previously established acceptable range. For example, an accuracy of less than 80% can be considered acceptable. To ensure accuracy meets the requirements, and based on known operational case requirements, the optimal balance between accuracy and false positive rate is determined. Then, with a sufficient sample size, the dataset is divided into 80% training and 20% validation. A support vector machine classifier is built using radial basis functions, and the values ​​of each dimension in the fault identification feature data are input as features. If the accuracy in the validation set exceeds the aforementioned acceptable range and the false positive rate remains below 5% under different fault types, the current parameter combination is considered optimal. These parameter values ​​are recorded, and a complete training run is performed. After completion, the model's classification results for all training samples are checked to see if the consistency with the actual fault type labels still meets the requirements. If it does, the search is stopped, and the model is solidified as a trained fault prediction model.

[0070] Based on the trained fault prediction model, the previously obtained model parameters and support vector machine classifier need to be deployed in the mobile power bank monitoring scenario and connected to real-time sampled voltage and current data. To this end, while the device is running continuously, operating information such as voltage and current is collected at a frequency of one second or higher, and existing preprocessing rules are used to calibrate the instantaneous sampled values. If the sampled values ​​exceed the collection calibration threshold, the sample is recorded as abnormal and will be compared again in the future. Then, the calibrated data is converted into the same fault identification feature form as during training and fed into the support vector machine model. The model will output a judgment score based on the classification hyperplane and kernel function determined by training. If the fault probability corresponding to the score exceeds the pre-set risk threshold, for example, the threshold can be defined as a decision value critical line obtained from the analysis of previous training samples, or different fault types can be assigned their own score thresholds. If the real-time score reaches or exceeds the threshold, the risk level at the current moment is recorded on the monitoring interface and a fault is marked. When multiple high scores are generated in the same period, they are regarded as higher-level fault risks. Finally, the judgment results of each time period are sorted out and a fault monitoring and early warning result is formed.

[0071] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention in any other way. Any person skilled in the art may make changes or modifications to the above-disclosed technical content to create equivalent embodiments that can be applied to other fields. However, any simple modifications, equivalent changes, and modifications made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the protection scope of the present invention.

Claims

1. A mobile power bank fault detection method based on intelligent diagnosis, characterized in that, Includes the following steps: Collect voltage and current data of the power bank under normal operating conditions, and generate an electrical performance baseline dataset by recording the timestamps of the data and the corresponding electrical performance indicators; Based on the aforementioned electrical performance baseline dataset, time-series features of voltage and current are extracted to obtain electrical performance feature data; Based on the electrical performance characteristic data, the intensity and phase of each frequency component are analyzed using Fast Fourier Transform to extract key spectral features and generate spectral feature data. The spectral feature data is then decomposed into multi-scale data by wavelet transform, which deconstructs the signal into detailed and approximate data at multiple time scales to obtain multi-scale spectral data. Based on the multi-scale spectral data and the electrical performance baseline dataset, the spectral characteristics at each scale are compared to identify fluctuations and non-periodic anomalies that deviate from the norm, and preliminary fault diagnosis results are generated. Geometric curvature analysis is performed on the preliminary fault diagnosis results to calculate the degree of geometric deformation between data points and obtain fault identification feature data. Based on the fault identification feature data, a fault prediction model is trained by a support vector machine, and the fault prediction model is applied to continuously monitor the power bank to obtain fault monitoring and early warning results. The steps for obtaining the electrical performance characteristic data are as follows: extract voltage and current data from the electrical performance baseline dataset, calculate the rate of change between adjacent data points, smooth the rate of change sequence, and combine the extreme points of voltage and current data to calibrate the fluctuation amplitude and stable range of current and voltage, and generate processed voltage and current time series data. Based on the processed voltage and current time series data, the time series characteristic values ​​of voltage and current are calculated using the following formula: ； in, Represents the feature values ​​of the time series. For time points Voltage data, For time points Voltage data, For time points Current data, For time points Current data, The total number of data points in the time series. Time point Voltage data, For time points Current data; Based on the time series characteristic values, combined with the extreme values, average values ​​and trends of current and voltage, electrical performance characteristic data are obtained; The steps for obtaining the spectral feature data are as follows: preprocessing the voltage and current time series in the electrical performance feature data, including removing noise and outliers, applying window function processing to reduce spectral leakage, and obtaining the processed time series data; Based on the processed time series data, the comprehensive characteristics of each frequency are calculated using the following formula: ； in, Representing frequency The comprehensive characteristics, It is frequency The result of the fast Fourier transform at that point, yes conjugate, and Represent The real and imaginary parts, , indicating phase, It is an adjustment factor. It is a smoothing parameter; Based on the comprehensive features, key spectral features are selected, and key spectral features are combined to generate spectral feature data. The steps for acquiring the multi-scale spectral data are as follows: Based on the spectral feature data, the wavelet coefficients at each scale are calculated using the following formula: ； in, Indicated in scale and location wavelet coefficients, It is the first of the spectrum data One point, It is the wavelet mother function. It is the total number of data points. Indicates time delay. These are Gaussian window parameters used to adjust the window width; Based on the wavelet coefficients, the signal is reconstructed to separate and identify details and approximations at each time scale, resulting in multi-scale spectral data.

2. The mobile power bank fault detection method based on intelligent diagnosis according to claim 1, characterized in that, The steps for obtaining the electrical performance baseline dataset are as follows: real-time monitoring and recording of the current and voltage data of the power bank, adding a timestamp to each data point to form a preliminary current and voltage time series record; Based on the preliminary current and voltage time series records, the instantaneous power, energy consumption, and efficiency of the current and voltage are calculated to obtain electrical performance index data. Based on the electrical performance index data, each index is integrated with a timestamp to construct an electrical performance baseline dataset.

3. The mobile power bank fault detection method based on intelligent diagnosis according to claim 1, characterized in that, The steps for obtaining the preliminary fault diagnosis results are as follows: acquiring the multi-scale spectrum data and the electrical performance baseline dataset, standardizing the multi-scale spectrum data and the electrical performance baseline dataset to make the data format consistent, and obtaining standardized multi-scale spectrum data and electrical performance baseline dataset. Based on the standardized multi-scale spectral data and electrical performance baseline dataset, the spectral characteristics at each scale are compared to identify deviations in frequency response, energy distribution, and time series stability, thus obtaining deviation analysis results. Based on the deviation analysis results, each anomaly is evaluated and classified. The anomalies are sorted and marked according to the severity of their impact and the frequency of their occurrence. A predicted fault type is defined for each anomaly, and preliminary fault diagnosis results are obtained.

4. The mobile power bank fault detection method based on intelligent diagnosis according to claim 1, characterized in that, The steps for obtaining the fault monitoring and early warning results are as follows: Based on the fault identification feature data and multi-scale spectrum data, configure and train the support vector machine model, select the radial basis function and adjust the model penalty coefficient and kernel parameters to obtain the trained fault prediction model. Based on the trained fault prediction model, the trained fault prediction model is deployed to the mobile power supply monitoring system to analyze the collected power operation data in real time, assess fault risks, and obtain fault monitoring and early warning results.

5. A mobile power bank fault detection system, used to implement the mobile power bank fault detection method based on intelligent diagnosis as described in any one of claims 1-4, characterized in that, include: The data acquisition module records the voltage and current of the power bank during operation, adds timestamps, and generates an electrical performance baseline dataset. The feature extraction module extracts the peak value, mean value, and standard deviation of voltage and current from the electrical performance baseline dataset to generate electrical performance feature data; The spectrum analysis module applies Fast Fourier Transform to the electrical performance characteristic data to analyze the intensity and phase of each frequency component, generating spectrum characteristic data. The spectrum characteristic data is then decomposed into multi-scale spectrum data through wavelet transform. The fault diagnosis module compares the spectral characteristics at each scale based on the multi-scale spectral data and the electrical performance baseline dataset, identifies fluctuations and non-periodic anomalies that deviate from the normal state, and generates preliminary fault diagnosis results. Geometric curvature analysis is performed on the preliminary fault diagnosis results to calculate the degree of geometric deformation between data points and obtain fault identification feature data. The fault prediction module, based on the fault identification feature data, trains a fault prediction model using a support vector machine, applies the fault prediction model to continuously monitor the power bank, and obtains fault monitoring and early warning results.

Images