Voltage-sensitive device damage diagnosis system combined with pulse test

Abstract

The invention relates to the technical field of electrical fault detection, in particular to a voltage-sensitive device damage diagnosis system combined with pulse test, which comprises the following modules: an original signal acquisition module for inputting pulse sequences to a voltage-sensitive device, respectively recording voltage response signals and current response signals under each pulse sequence, and generating a response signal set; and performing wavelet decomposition on the voltage response signal and the current response signal based on the response signal set. According to the method, the pulse sequences are input, and the voltage response signals and the current response signals under the pulse sequences are recorded, so that data acquisition covers different working states, and feature missing possibly caused by a single test point is reduced. A wavelet decomposition mode is adopted, a voltage response signal and a current response signal are divided into different frequency intervals, separation of high-frequency and low-frequency characteristics is facilitated, local abnormal signals are not covered by global characteristics, and the capability of analyzing short-time fault characteristics is enhanced.

Description

Technical Field

[0001] The present invention relates to the technical field of electrical fault detection, and in particular to a diagnosis system for damaged varistor devices combined with pulse testing. Background Art

[0002] The method for diagnosing damaged varistor devices is a fault detection technology for varistors, aiming to evaluate the damage type, failure mode, and performance degradation of varistor devices through the analysis of current and voltage response signals.

[0003] The existing technology is based on the simple analysis of current and voltage response signals, resulting in the difficulty of comprehensively extracting fault characteristics. Especially in the case of high-frequency signal disturbances or transient faults, it is vulnerable to noise interference, making it difficult to distinguish damage signs. At the same time, different frequency components of the signal cannot be separately analyzed, and local abnormal signals are easily masked by the overall trend. Especially in the initial stage of equipment damage, small characteristic changes are easily ignored, reducing the sensitivity of fault detection. Therefore, improvements are needed. Summary of the Invention

[0004] The purpose of the present invention is to solve the disadvantages existing in the prior art, and to propose a diagnosis system for damaged varistor devices combined with pulse testing.

[0005] To achieve the above purpose, the present invention adopts the following technical solution. The diagnosis system for damaged varistor devices combined with pulse testing includes the following modules:

[0006] The original signal acquisition module inputs a pulse sequence to the varistor device, respectively records the voltage response signal and current response signal under each pulse sequence, and generates a response signal set; based on the response signal set, wavelet decomposition is performed on the voltage response signal and current response signal to obtain a time-frequency feature matrix;

[0007] The signal wavelet decomposition module calculates the instantaneous frequency and phase value for the high-frequency coefficient group and low-frequency coefficient group respectively based on the time-frequency feature matrix, and generates a frequency-phase vector; based on the frequency-phase vector, the instantaneous frequency and phase value are reconstructed into a frequency sequence according to the time series, and Hilbert transform is performed on the frequency sequence to generate a fault feature spectrum;

[0008] The fault feature spectrum generation module segments the amplitude spectrum and phase spectrum on the time axis based on the fault feature spectrum, calculates the linear fitting parameters for each segment of the amplitude spectrum and phase spectrum, and generates a time-domain fitting parameter set; based on the time-domain fitting parameter set, the linear fitting parameters are combined with the voltage-current ratio within the current time period to establish a feature drift curve;

[0009] The fault diagnosis and mapping module, based on the feature drift curve, numerically combines the threshold drift amount with the voltage response signal and current response signal in the current time period to generate a combined feature value. Based on the combined feature value, it calculates the difference amount between the combined value and the standard threshold, discriminates and classifies the damage types of varistor devices according to the difference amount, and establishes a damage type mapping table.

[0010] Preferably, the steps for obtaining the response signal set are as follows:

[0011] Connect the varistor device to the pulse generator, start the emission of pulse sequences, record the voltage response signal and current response signal caused by each pulse sequence, and obtain preliminary response data;

[0012] Based on the preliminary response data, classify and organize the voltage response signal and current response signal to obtain comprehensive response data;

[0013] Based on the comprehensive response data, integrate all voltage response signals and current response signals to form a response signal set.

[0014] Preferably, the steps for obtaining the time-frequency feature matrix are as follows:

[0015] Based on the response signal set, perform wavelet decomposition on the voltage response signal and current response signal respectively, extract the wavelet coefficients of each signal at different scales, classify the wavelet decomposition coefficients according to a preset frequency interval, and obtain a high-frequency coefficient group and a low-frequency coefficient group;

[0016] According to the high-frequency coefficient group and the low-frequency coefficient group, calculate the energy distribution value of each group. The calculation formula is:

[0017]

[0018] where E k represents the energy distribution value of the kth group, C k,m represents the absolute value of the mth wavelet decomposition coefficient in the kth group, D k,m represents the deviation term of the mth coefficient in the kth group, and QM is the number of wavelet coefficients;

[0019] Based on the energy distribution value, construct a time-frequency feature matrix and fill the energy distribution value of each group in the time-frequency feature matrix.

[0020] Preferably, the steps for obtaining the frequency-phase vector are as follows:

[0021] Based on the time-frequency feature matrix, extract the components of the high-frequency coefficient group and the low-frequency coefficient group respectively, calculate the instantaneous amplitude change rate for each component, and obtain the instantaneous change feature data by combining the time information;

[0022] Based on the instantaneous change characteristic data, calculate the instantaneous frequency and phase values, organize the time series information of each component, and associate the instantaneous frequency and phase change conditions of each component to obtain the instantaneous frequency data and phase data;

[0023] Based on the instantaneous frequency data and phase data, match and classify the data according to the corresponding frequency intervals, and combine the corresponding instantaneous frequency and phase values into a frequency-phase vector.

[0024] Preferably, the steps for obtaining the fault characteristic spectrum are as follows:

[0025] Based on the frequency-phase vector, extract the instantaneous frequency and phase values corresponding to each time point, sort them in the order of the time series, and perform interpolation and complementation. At the same time, smooth the frequency points with abnormal changes to obtain the reconstructed frequency series;

[0026] Perform Hilbert transform on the reconstructed frequency series, and calculate the instantaneous amplitude spectrum value after the transform. The calculation formula is:

[0027]

[0028] where A j is the instantaneous amplitude spectrum value corresponding to the jth frequency point, P j is the original component value in the frequency series, is the component after applying Hilbert transform to P j ;

[0029] Based on the instantaneous amplitude spectrum value, construct a feature matrix including amplitude information, classify the amplitude values corresponding to each time point according to the frequency distribution, and arrange the classified data in time sequence to form a fault characteristic spectrum.

[0030] Preferably, the steps for obtaining the time domain fitting parameter set are as follows:

[0031] Based on the fault characteristic spectrum, divide it into multiple time periods along the time axis, extract the amplitude data and phase data in each time period to form a segmented amplitude spectrum and phase spectrum;

[0032] According to the segmented amplitude spectrum and phase spectrum, calculate the linear fitting parameters of each time period. The calculation formula is:

[0033]

[0034] where L k represents the linear fitting parameter of the kth segment, p k,m represents the mth amplitude data point in the kth segment, q k,m represents the mth phase data point in the kth segment;

[0035] Based on the linear fitting parameters, arrange the fitting parameters of all time periods in chronological order, construct a complete set of time-domain fitting parameters, and correlate the fitting parameters of each time period according to the distribution relationship between the amplitude spectrum and the phase spectrum to generate a time-domain fitting parameter set.

[0036] Preferably, the steps for obtaining the characteristic drift curve are as follows:

[0037] Based on the time-domain fitting parameter set, extract the linear fitting parameters of each time period, obtain the voltage-current ratio corresponding to the time period, and at the same time use the sliding window method for smoothing processing to form time period matching data;

[0038] According to the time period matching data, calculate the threshold drift amount of each time period. The calculation formula is:

[0039]

[0040] where T k represents the threshold drift amount of the k-th time period, P k,m represents the m-th linear fitting parameter value within the k-th time period, R k,m represents the m-th voltage-current ratio within the k-th time period, and M represents the total number of data points within the k-th time period;

[0041] Based on the threshold drift amount, arrange the drift data of all time periods in chronological order and construct a characteristic drift curve including the drift characteristics of all time periods.

[0042] Preferably, the steps for obtaining the combined characteristic value are as follows:

[0043] According to the characteristic drift curve, calculate the combined characteristic value. The calculation formula is:

[0044]

[0045] where TC k represents the combined characteristic value of the k-th time period, U k (x) represents the voltage response signal function of the k-th time period, I k (x) represents the current response signal function of the k-th time period, U k,j represents the j-th voltage response signal value of the k-th time period, I k,j represents the j-th current response signal value of the k-th time period, T k,j represents the j-th threshold drift amount of the k-th time period, J represents the total number of data points within the k-th time period, and Q is the total change range of the signal within the time period.

[0046] Preferably, the steps for obtaining the damage type mapping table are as follows:

[0047] Obtain standard threshold data, arrange the combined feature values in chronological order to form combined feature values and standard threshold data;

[0048] Based on the combined feature values and standard threshold data, calculate the difference amount for each time period, match the calculated difference amounts along the time axis, and arrange all the difference amounts in the chronological order of the data points to obtain the difference amount between the combined value and the standard threshold;

[0049] Based on the difference amount between the combined value and the standard threshold, define the damage type discrimination rules, classify the damage types corresponding to all data points according to the preset classification threshold, and map the data of different damage types to the corresponding categories to establish a damage type mapping table.

[0050] Compared with the prior art, the advantages and positive effects of the present invention are as follows:

[0051] In the present invention, an input pulse sequence is input and the voltage response signal and current response signal under each pulse sequence are recorded, so that the data acquisition covers different working states and reduces the possible feature loss caused by a single test point. By using the wavelet decomposition method, the voltage response signal and current response signal are divided into different frequency intervals, which helps to separate high-frequency and low-frequency features, prevents local abnormal signals from being masked by global features, and enhances the analysis ability of short-time fault features. Calculate the instantaneous frequency and phase value and construct a frequency-phase vector, making the dynamic evolution process of the signal easier to analyze. Combined with time series reconstruction, the instantaneous features can form a continuous change trajectory, improving the traceability of fault features. Use the Hilbert transform to obtain the fault feature spectrum, making the change of signal energy distribution more intuitive and enhancing the ability to distinguish damage types. By performing time-axis segmentation processing on the amplitude spectrum and phase spectrum and calculating the linear fitting parameters of each segment, the local trend change of the signal is clearly visible, improving the analysis ability of complex fault modes. Combine the voltage-current ratio to establish a feature drift curve, enabling the fault features to form a trend in the time dimension and avoiding the interference of abnormal points at a single moment on the discrimination result. Adopt a numerical combination method to fuse the threshold drift amount, voltage response signal and current response signal, making the fault features more comprehensive and reducing the misjudgment caused by a single feature. Calculate the difference amount between the combined value and the standard threshold, providing a clearer quantitative basis for damage classification, and construct a damage type mapping table, making different damage modes form a traceable standard classification, improving the automation degree of damage diagnosis and reducing the need for manual intervention. Description of the Drawings

[0052] Figure 1 It is a step schematic diagram of the present invention. Detailed Embodiment

[0053] To make the objectives, technical solutions and advantages of the present invention more clear and understandable, the present 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 only used to explain the present invention and are not used to limit the present invention.

[0054] Please refer to Figure 1 , the present invention provides a technical solution, a diagnosis system for damaged varistors in combination with pulse testing, including the following modules:

[0055] An original signal acquisition module, which inputs a pulse sequence to the varistor, respectively records the voltage response signal and current response signal under each pulse sequence, and generates a set of response signals; based on the set of response signals, performs wavelet decomposition on the voltage response signal and current response signal to obtain a time-frequency feature matrix;

[0056] A signal wavelet decomposition module, based on the time-frequency feature matrix, calculates the instantaneous frequency and phase value for the high-frequency coefficient group and low-frequency coefficient group respectively to generate a frequency-phase vector; based on the frequency-phase vector, reconstructs the instantaneous frequency and phase value into a frequency sequence according to the time series, and performs Hilbert transform on the frequency sequence to generate a fault feature spectrum;

[0057] A fault feature spectrum generation module, based on the fault feature spectrum, segments the amplitude spectrum and phase spectrum on the time axis, calculates the linear fitting parameters for each segment of the amplitude spectrum and phase spectrum to generate a set of time-domain fitting parameters; based on the set of time-domain fitting parameters, combines the linear fitting parameters with the voltage-current ratio within the current time period to establish a characteristic drift curve;

[0058] A fault diagnosis and mapping module, based on the characteristic drift curve, numerically combines the threshold drift amount with the voltage response signal and current response signal in the current time period to generate a combined characteristic value, based on the combined characteristic value, calculates the difference between the combined value and the standard threshold, and discriminates and classifies the damaged type of the varistor according to the difference, and establishes a damaged type mapping table.

[0059] The steps for obtaining the set of response signals are as follows:

[0060] Connect the varistor to the pulse generator, start the emission of the pulse sequence, and record the voltage response signal and current response signal caused by each pulse sequence to obtain preliminary response data;

[0061] Based on the preliminary response data, classify and organize the voltage response signal and current response signal to obtain comprehensive response data;

[0062] Based on the comprehensive response data, integrate all the voltage response signals and current response signals to form a set of response signals.

[0063] Specifically, referring to the usage specifications of the varistor device and combining with the technical data of the pulse generator, first confirm the rated tolerance range of the connection cable and the rated voltage and current limits of the varistor device. For example, control the voltage in the range of 0V to 24V and the current in the range of 0A to 5A. After observing that the connection terminals comply with the equipment specifications, set the actual number of pulses to 20 and the emission interval of each pulse to 1 second. In this way, the fixed time interval facilitates comparing the response status at each emission moment one by one. Then, before starting the pulse emission, check the corresponding user manuals of the varistor device and the pulse generator to clarify that the safety protection values described therein are obtained based on multiple actual measurements and industry experience. For example, the pulse voltage threshold is generally taken as 80% of the calibrated range of the equipment as the initial setting value. If the voltage measured at any time during the emission exceeds this value, it is determined that the pulse intensity is too high, and the detection needs to be paused and the output position needs to be recalibrated. At the same time, the voltage response signal and current response signal caused by each pulse are recorded in real time at the acquisition end. The specific method is to read the voltage and current curves generated by each pulse in segments at a sampling frequency of 10kHz and compare them with the previously set intervals. For example, judge whether there is an oscillation overrun situation in the range of 0V to 24V, and judge whether there is a sudden increase in current in the range of 0A to 5A. If it is found that the peak value or average value of a single pulse collected is abnormal, it is prompted in a marked manner for further investigation. Finally, the signals generated by each pulse are saved in a temporary database, and the preliminary response data is obtained after summarization.

[0064] Compare with the preliminary response data, and confirm how the rated ranges of each voltage response signal and current response signal are established through multiple experiments and industry experience when consulting the relevant equipment data. For example, compare the voltage data with the range of 0V to 24V one by one, and compare the current data with the range of 0A to 5A one by one. If it is found that any record is in the suspicious range established by prior experience or exceeds the nominal upper limit, mark this record as an object that needs to be observed closely. Subsequently, sort and organize all records by voltage and current. When classifying the voltage signals, it is mainly based on the interval division, and determine the high-frequency occurrence interval or low-frequency interval according to the number of times the record appears. The same operation is performed for the current signals, but at the same time, evaluate whether there is an overshoot phenomenon in the current transient by combining the maximum current deviation value. These specific numerical ranges can be, for example, referred to the actually measured average value plus or minus the corresponding error correction amount. All sorting and grouping processes are carried out using a fixed method to avoid missing potential anomalies and maintaining the comparability of the overall data. After all classification results are completed, summarize the classified voltage response signals and current response signals one by one and record them correspondingly to obtain the comprehensive response data.

[0065] After obtaining the comprehensive response data, the voltage response signals and current response signals under each category are rearranged and integrated. The specific steps are to first filter out all records that exceed the empirical setting range of the equipment and list them separately. For example, if a voltage record is higher than 24V or lower than 0V, it is placed in the exception list separately. If the current record exceeds 5A or is lower than 0A, it is also placed in the exception list. Then, the remaining records in the range of 0V to 24V and 0A to 5A are paired in time sequence to ensure that the voltage and current collected at the same time can accurately correspond to each other and form a comparable integrated data set. Then, the voltage data and current data at the corresponding time points are jointly verified, for example, the ratio of each record is calculated and the interval is defined according to the empirical threshold. If the ratio exceeds the previous statistical mean plus or minus a certain deviation, it is marked as a possible abnormal point. After completing this batch processing, all the data in the normal interval and the exception list are placed in the same table for backup, and finally a response signal set is formed.

[0066] The steps to obtain the time-frequency feature matrix are:

[0067] Based on the response signal set, the voltage response signal and the current response signal are respectively decomposed by wavelet, the wavelet coefficients of each signal at different scales are extracted, and the wavelet decomposition coefficients are classified according to the preset frequency interval to obtain the high-frequency coefficient group and the low-frequency coefficient group;

[0068] According to the high-frequency coefficient group and the low-frequency coefficient group, the energy distribution value of each group is calculated. The calculation formula is:

[0069]

[0070] Among them, E k represents the energy distribution value of the kth group, C k,m represents the absolute value of the mth wavelet decomposition coefficient in the kth group, D k,m represents the deviation term of the mth coefficient in the kth group, QM is the number of wavelet coefficients;

[0071] Based on the energy distribution values, a time-frequency feature matrix is constructed, and the energy distribution values of each group are filled in the time-frequency feature matrix.

[0072] Specifically, based on the previously obtained set of response signals, the corresponding characteristics of the voltage response signal and the current response signal are compared with reference to the recorded pulse test conditions. The amplitude changes of each signal within the frequency range of 0 Hz to 100 kHz are read. Then, in combination with the segmented sequence after frequency-domain sampling, the wavelet function type of each signal at different scales is determined. Then, with reference to the frequency interval division, for example, frequencies below 20 Hz are defined as the low-frequency band, 20 Hz to 500 Hz are defined as the sub-low-frequency band, 500 Hz to 5 kHz are defined as the middle-frequency band, and 5 kHz to 100 kHz are defined as the high-frequency band. The wavelet coefficients corresponding to each signal are extracted segment by segment according to this interval, and the occurrence time and the absolute value of the amplitude are recorded. When the amplitude value that appears in any paragraph exceeds the empirical threshold statistically obtained in the previous experiments, this paragraph is marked as an abnormal segment. This empirical threshold is obtained through multiple pulse tests and on-site monitoring. The specific method is based on the pulse response record. After summarizing the average upper limit of the peak amplitudes of each paragraph in nearly ten groups of experimental data and adding a safety margin, this value is defined as the comparison standard. Then, a secondary comparison is performed on the marked abnormal segments to confirm whether the degree of amplitude deviation further increases. If the value still exceeds the set range of this empirical threshold, this paragraph is recorded to distinguish potential failure factors. Finally, after the segmentation is completed, from all the decomposition results obtained from the above operations, all the wavelet coefficient sets with frequency distributions falling in the high-frequency band and all the wavelet coefficient sets in the low-frequency band are concentrated and extracted, respectively forming a high-frequency coefficient group and a low-frequency coefficient group.

[0073] The advantage of the formula lies in combining the with the comprehensive characteristic quantity, and introducing in the denominator for proportional measurement, so that the energy distribution value takes into account both the absolute amplitude and the deviation, avoiding the local imbalance problem caused by simply using the square or absolute value.

[0074] C k,m The steps for obtaining the C k,m parameter are as follows. C k,m refers to the absolute value of the m-th wavelet decomposition coefficient in the k-th group, and this coefficient comes from the data component at the corresponding hierarchical scale after wavelet decomposition of the response signal. First, the wavelet scale is determined according to the number of decomposition layers, and then the coefficients at each time point at this scale are read in sequence, and their absolute values are taken to obtain C k,m . If in actual measurement, the original value of the coefficient at a certain moment is -2.3, then C k,m, for example, when a certain pulse signal is measured on-site at 500 Hz stratification, it contains 8 coefficient values. After taking the absolute value in sequence, it constitutes the specific sequence of C k,m of

[0075] D k,m The steps for obtaining the D parameter are as follows. D k,m represents the deviation term of the m-th wavelet coefficient in the k-th group. First, after calculating C k,m , subtract the average coefficient value at this stratification scale or the reference value of the previous cumulative coefficient from the actual observed value to obtain the deviation. By continuously accumulating the average coefficient within the same group and updating it during the pulse experiment, D k,m changes accordingly. The specific method is based on the mean MM k of the waveform record in the previous round within the same group. Then calculate the difference between the originally measured coefficient value and MM k . For example, if the originally measured coefficient value at a certain moment is 2.5 and the mean of the previous round is 2.2, then D k,m is 0.3. Repeating this process for all moments can form the entire group of D k,m sequence. If someone wants to obtain this deviation term under other conditions, it can also be achieved through the corresponding same pulse test and the same stratification scale. Then subtract the mean obtained from historical statistics within this group from the measured coefficient value to obtain the value of D k,m .

[0076] The steps for obtaining the QM parameter are as follows. QM represents the total number of wavelet coefficients in the k-th group, which is usually determined by the decomposition level and the number of time-series sampling points. For example, during pulse measurement, each segment of the signal is set with 400 sampling points. Through 5-layer wavelet decomposition, a certain number of coefficients are produced at each layer. After calculation, if 12 coefficients are merged at this level in the k-th group, QM can be accumulated to 12. This cumulative process may involve the merging of multiple sub-bands to include all the coefficient numbers in the same frequency band. In addition, when there are overlapping frequency bands in some layers, their duplicates also need to be removed. The finally calculated number of effective wavelet coefficients is QM. In specific operations, the available coefficient points of each layer can be pre-recorded and added, or the total number of coefficients included in this group can be directly counted after the transformation. These methods can all determine the value of QM.

[0077] The steps to obtain the k parameter are as follows. k represents the group index, which is used to distinguish different frequency bands or decomposition levels. If five-layer wavelet decomposition is performed, the optional range of k is usually from 1 to 5. However, in applications that require secondary subdivision of the high-frequency band, the range of k may also be expanded to 10 or more. In this case, the value of k needs to be determined according to the self-frequency division strategy. A common practice in on-site measurement is to first divide the frequency into several intervals, and then perform independent wavelet decomposition on each interval. Each interval is recorded as a group, and k corresponds to the numbers of these intervals. For example, the range from 0 Hz to 500 Hz is set as the first group, and the range from 500 Hz to 5 kHz is set as the second group, and so on.

[0078] The steps to obtain the m parameter are as follows. m is the coefficient serial number within each group, and its specific value depends on how many effective coefficients actually appear in the k-th group after decomposition. Usually, it starts from 1 and increases incrementally one by one to QM. For example, m = 1 means the first wavelet coefficient in this group, and m = 2 means the second wavelet coefficient in this group, and so on. This parameter can be directly obtained from time series alignment or index identification. Whenever it is confirmed which coefficients are included in the k-th group in the grouping result, numbering starts from 1 and continues until QM is reached.

[0079] Substitute the parameters into the formula for calculation to obtain 9.24. This result indicates that when E2 is close to 9.24, it means that the energy distribution within the second group of frequency intervals is relatively concentrated under the current pulse test environment. If E2 further increases to more than 30, it shows that both the sum of the absolute coefficients cubed and the cumulative deviation of this group increase abnormally. The E of other groups can also be calculated in the same way. k Thereby, further judge the overall energy distribution situation.

[0080] Based on the energy distribution values obtained above, first create a blank structure in the pre-determined time-frequency feature matrix area to accommodate the distribution records of each group. Subsequently, traverse the calculation results of each group one by one, and map the corresponding E k values to the row and column positions of the time-frequency feature matrix according to the group number index and time sequence, and conduct comparison and analysis. To ensure the alignment of the rows and columns in the matrix with the actual segmentation order, it is necessary to first assign indexes corresponding to different time periods in the row direction of the matrix. For example, the segmentation number between 1 second and 2 seconds is recorded as the first row, and the segmentation number between 2 seconds and 3 seconds is recorded as the second row, and so on. In the column direction, it continues to expand in the order of the first group, the second group, and so on. If the E k value calculated for any group within a certain period of time exceeds the upper limit of the range statistically obtained in the previous pulse test. For example, if the upper limit of this range is defined as 50, then a special mark is made at the corresponding position in this matrix during filling, and these marks are recorded in a separate warning list during the subsequent summary step. The upper limit of this range is mainly determined by statistically calculating the maximum E after multiple measurements. kCombined with a certain degree of fluctuation margin to determine an appropriate distribution boundary. For example, select the highest value of each group of E in 10 consecutive trials and then add a 10% compensation amount. Finally, set this value as the safety upper limit of the energy distribution. When it is confirmed that all rows and columns in the matrix have completed writing the corresponding E k values, it can be regarded that the time-frequency feature matrix has been constructed, and then the energy distribution values of each group are correctly filled into the corresponding positions. k After writing the

[0081] Steps for obtaining the frequency-phase vector are as follows:

[0082] Based on the time-frequency feature matrix, extract the components of the high-frequency coefficient group and the low-frequency coefficient group respectively, calculate the instantaneous amplitude change rate for each component, and obtain the instantaneous change feature by combining time information to get the instantaneous change feature data;

[0083] Based on the instantaneous change feature data, calculate the instantaneous frequency and phase values, organize the time series information of each component, and associate the instantaneous frequency and phase change conditions of each component to obtain the instantaneous frequency data and phase data;

[0084] Based on the instantaneous frequency data and phase data, match and classify the data according to the corresponding frequency intervals, and combine the corresponding instantaneous frequency and phase values into a frequency-phase vector.

[0085] Specifically, based on the time-frequency feature matrix, the components of the high-frequency coefficient group and the low-frequency coefficient group are extracted respectively. The instantaneous amplitude change rate is calculated for each component, and the instantaneous change feature is obtained by combining the time information. First, it is necessary to compare the values of each time segment in the high-frequency coefficient group and the low-frequency coefficient group point by point, record the amplitude difference of each component at adjacent time points, and then use the ratio of the difference to the current amplitude to determine the instantaneous amplitude change rate. For example, if the amplitude of a certain component is 3.2 at time t1 and 3.5 at time t2, the instantaneous increment of this component can be calculated as 3.5 minus 3.2 equals 0.3, and then 0.3 is divided by 3.2 to get approximately 0.0937. If this instantaneous amplitude change rate exceeds the empirical threshold obtained from the previous experimental statistics, for example, taking 0.1 as the judgment standard and determined by centralized comparison of the results of multiple rounds of detection, then this component will be marked as a rate deviation interval. Subsequently, after calculating the instantaneous amplitude change rate for each component, it is also necessary to determine the specific change trend of this component within the adjacent time period according to the corresponding time label. These values can be accumulated hour by hour and compared to see if the total exceeds the established upper limit. For example, if it is found through continuous observation that the cumulative change rate has reached 2.5 within a certain period of time, while the upper limit of the stable interval obtained from repeated experiments before usually does not exceed 1.5, then further investigation of this component is required. In this way, the instantaneous amplitude change conditions of each component can be effectively collected and recorded and distinguished one by one in combination with the time information. Finally, the obtained time series change data is sorted to obtain the instantaneous change feature data.

[0086] Based on the instantaneous change characteristic data, calculate the instantaneous frequency and phase values, organize the time series information of each component, and correlate the instantaneous frequency and phase change conditions of each component. It is necessary to split the instantaneous change characteristic data into independent sequences according to component attribution, and then read the amplitude and time position at each moment one by one. Determine the interval between adjacent peaks or adjacent zero-crossing points through measurable means such as zero-crossing or differentiation. Divide this interval by the corresponding time length to obtain the instantaneous frequency. For example, when the time interval between two peaks of a certain component is 0.002 seconds, the instantaneous frequency can be around 500 Hz, and record the specific value for comparison with similar components. If this instantaneous frequency exceeds the upper limit of the frequency range obtained in the previous on-site measurement, for example, according to the peak distribution of the component collected from ten pulse tests, most of them fall between 100 Hz and 3 kHz. If the current component appears above 5000 Hz, it is necessary to further analyze the reason. Infer whether abnormal oscillation occurs by statistically analyzing the frequency trend of this component on the entire time axis and combining the amplitude and time changes. As for the calculation of the phase value, it can be realized by using the relative position relationship between adjacent points. After determining the time series corresponding to each peak, compare the reference point of this component with the reference phase. If the difference reaches 180 degrees, it means that there is a large misalignment between this component and the reference period in the current interval. After arranging these information continuously, the instantaneous phase change data of each component can be formed. Finally, by making a one-to-one correspondence between the instantaneous frequency data and the phase data according to the sequence index and merging them in the same record, the instantaneous frequency data and the phase data are obtained.

[0087] Based on the instantaneous frequency data and the phase data, match and classify the data according to the corresponding frequency intervals. It is necessary to first confirm the specific division criteria of the frequency. For example, record 0 Hz to 500 Hz as the low-frequency band, 500 Hz to 5 kHz as the medium-frequency band, and 5 kHz to 100 kHz as the high-frequency band. The actual division can refer to the observation results of previous multiple pulse tests and the rated parameters of the equipment. Then, when matching the phase data for each instantaneous frequency data, check the time index and amplitude correlation degree of the two. If the instantaneous frequency at the same moment is 150 Hz and the phase is 45 degrees, then classify this pair of data into the low-frequency band list and continue to match the subsequent moments. If the instantaneous frequency of the same component rises to 3 kHz and the phase changes to 120 degrees at the next moment, then classify it into the medium-frequency band record. After traversing all moments and completing the matching, jointly arrange the corresponding instantaneous frequency and phase values in each record to form a binary vector, and then mark the time stamp in the segmented index table and sort them sequentially. During this period, it is also necessary to check whether there is duplicate or missing time period information. If there is a missing value, the average frequency and phase of adjacent moments can be used to interpolate this record. After re-checking and confirming that there are no obvious errors, combine all the matched instantaneous frequency and phase values into a frequency-phase vector.

[0088] The steps to obtain the fault characteristic spectrum are as follows:

[0089] Based on the frequency-phase vector, extract the instantaneous frequency and phase values corresponding to each time point, sort them in the order of the time series, and perform interpolation to complete the data. At the same time, smooth the frequency points with abnormal changes to obtain the reconstructed frequency series;

[0090] Perform Hilbert transform on the reconstructed frequency series, and calculate the instantaneous amplitude spectrum value after the transform. The calculation formula is:

[0091]

[0092] where A j is the instantaneous amplitude spectrum value corresponding to the j-th frequency point, P j is the original component value in the frequency series, is the component after applying Hilbert transform to P j ;

[0093] Based on the instantaneous amplitude spectrum value, construct a feature matrix including amplitude information, classify the amplitude values corresponding to each time point according to the frequency distribution, and arrange the classified data in time sequence to form a fault feature spectrum.

[0094] Specifically, based on the frequency-phase vector obtained previously, extract the instantaneous frequency and phase values corresponding to each time point and conduct a preliminary sorting. Subsequently, arrange all records one by one in the order of the time series and interpolate and complete the data of the missing moments by the average value of adjacent time slices. In the interpolation process, first determine the left and right boundary time points of the interpolation interval and analyze the existing instantaneous frequency and phase data at both ends, and then take the values of these two points for point-by-point calculation of linear or polynomial. For example, set several stepping points in the interval with an adjacent observation time difference of 0.01 seconds and interpolate and fill the instantaneous frequency and phase of these points one by one. If it is detected that the sudden increase amplitude of the frequency at certain moments is greater than the range of the previously established empirical threshold, for example, the threshold range is from 50 Hz to 5000 Hz and is obtained from the distribution analysis of the sampling data of multiple pulse experiments. When the frequency at any moment breaks through 5000 Hz, it is marked as an abnormal record. Calculate its specific sudden increase ratio by comparing the frequency increment of this record with the previous and subsequent moments. If the increment is greater than 50% of the original value and lasts for more than two samplings, it can be determined as a local abnormal change point. Then, perform a smoothing operation on the instantaneous frequency within this time period. The smoothing method can adopt the median processing of adjacent 3 or 5 time points to correct the significantly deviated values to the relatively close values around. During this period, if the phase value also jumps at the same moment point, synchronous inspection can be carried out and similar processing can be done, and mark the time point with an outlier amplitude in the record. After all the data of all moments are completed with interpolation and smoothing, merge the finally corrected frequency records into a continuous sequence in time order to obtain the reconstructed frequency series.

[0095] The benefit of the formula is through For the original component P j After performing the Hilbert transform, the real part of the original frequency component is combined with its corresponding imaginary part, so as to obtain a more intuitive instantaneous intensity quantity at the amplitude spectrum level.

[0096] P j The steps for obtaining the parameter are as follows:

[0097] This parameter is the original component value in the frequency sequence, representing the frequency observed at the j-th time index or the amplitude record related to this frequency. It is usually obtained by performing a sub-analysis on the segmented data during the pulse test. Each segment of data is attached with a specific timestamp, and these timestamps can be used to accurately locate the actual sampling moment corresponding to the j-th observation value. The frequency value or amplitude detected at this moment is marked as P j If higher accuracy is required, the original sampling can be used to eliminate random noise by means of averaging or multiple repeated tests, and finally a stable and reliable P j sequence is formed. Taking the following example, when j = 3, the original record at this observation moment can be read as 2.0.

[0098] The steps for obtaining the parameter are as follows:

[0099] This parameter refers to the component obtained by applying the Hilbert transform to P j It is realized by performing a fast transform operation on the continuous frequency sequence. It is required to first perform Fourier domain processing on the entire frequency sequence and introduce a phase shift, and then convert it back to the time domain. The result obtained is the component value corresponding to P j after the Hilbert transform. The numerical value also depends on the sampling accuracy and the frequency range division. If is to be obtained, the Fourier transform can be performed on the record of j = 3 during sampling and combined with the existing frequency band index for calculation. Finally, the imaginary part corresponding to this index is extracted as in the following example, the value of will be listed as 1.5.

[0100] The steps for obtaining the j parameter are as follows:

[0101] This parameter is the index number used to identify the position of each record point in the entire frequency sequence. j increases sequentially from 1 according to the time order. For example, in the interval of observing the pulse signal from 0 to 2 seconds, a total of 10 points are sampled, then the value range of j is from 1 to 10, and each j corresponds to an independent P j and observation value.

[0102] Substitute the parameters into the formula for calculation to obtain 2.5. This result indicates that when the original component P3 = 2.0 and the Hilbert transform component At this time, the instantaneous amplitude spectrum value A3 at the third frequency point is 2.5. If a significant increase or decrease in A is continuously detected at adjacent times j = 4 or j = 5, the corresponding change trend can be recorded in the subsequent feature matrix and summarized into the next analysis step. j Significantly increase or decrease, and the corresponding change trend can be recorded in the subsequent feature matrix and summarized into the next analysis step.

[0103] Based on the instantaneous amplitude spectrum value, combined with the amplitude value and frequency distribution information corresponding to each time point, it is necessary to first select a continuous time range during the pulse signal sampling process, read the instantaneous amplitude spectrum values at different times in this time period one by one, and compare them with a predetermined frequency classification list. For example, 0Hz to 500Hz is defined as the low-frequency region, 500Hz to 5kHz is defined as the medium-frequency region, and above 5kHz is defined as the high-frequency region. Then, the instantaneous amplitude spectrum values are classified into corresponding data segments according to the frequency range where each time point is located, and the corresponding frequencies and amplitudes are matched one by one. If it is found that the amplitude at a certain moment exceeds the upper limit of the range statistically obtained through cumulative observations in advance, such as setting the upper limit to 110 by adding a 10% buffer when the amplitude peak value obtained after multiple rounds of detection is 100, if a record with an amplitude of 120 appears, it is marked as crossing the boundary value, and the marking information is saved in the subsequent record summary for comparison. Subsequently, all amplitude spectrum values are arranged hour by hour or sample by sample according to the time series, so that the corresponding amplitude and frequency combinations can be found for each time node, and the matched records are generated into a row or a column in the matrix, and the complete time series distribution structure is expanded column by column or row by row. This structure can be located row by column using the index j mentioned above. Finally, each record is summarized to form a feature matrix containing amplitude information. Subsequently, traversing different time periods and frequency combinations in the matrix can export the distribution data associated with the fault to form a fault feature spectrum.

[0104] The steps for obtaining the time-domain fitting parameter set are as follows:

[0105] Based on the fault feature spectrum, divide it into multiple time periods according to the time axis, extract the amplitude data and phase data in each time period to form a segmented amplitude spectrum and phase spectrum;

[0106] According to the segmented amplitude spectrum and phase spectrum, calculate the linear fitting parameters for each time period. The calculation formula is:

[0107]

[0108] Among them, L k represents the linear fitting parameter of the kth segment, p k,m represents the mth amplitude data point in the kth segment, q k,m represents the mth phase data point in the kth segment;

[0109] Based on the linear fitting parameters, arrange the fitting parameters of all time periods in chronological order and construct a complete set of time-domain fitting parameters. Associate the fitting parameters of each time period according to the distribution relationship between the amplitude spectrum and the phase spectrum to generate the time-domain fitting parameter set.

[0110] Specifically, based on the fault characteristic spectrum, first split the obtained amplitude data and phase data in the order of the time axis. Determine the specific number of time periods according to the number of records included at each moment in the previously obtained fault characteristic spectrum. Each period can be set between 1 second and 2 seconds or longer. When actually dividing, it is necessary to combine the on-site pulse sampling rate and the continuous observation range of the equipment to judge whether sufficient amplitude data and phase data can be included in this time period. After each period is determined, group and register the amplitude values inside it. Aggregate all the amplitude values belonging to the same time period together, and at the same time read the corresponding phase values to form one-to-one records. Then, compare the amplitude and phase information in the same paragraph with the pre-set valid range one by one. For example, compare the amplitude with the range from 0 to 100, and the phase with the range from 180 degrees to 180 degrees. Mark all the parts that exceed the equipment statistical upper limit as abnormal records and mark them in the abnormal list of this time period. If the number of abnormal records exceeds a certain empirical threshold, such as 10, which is obtained by statistically averaging the average abnormal value of each segment in multiple industrial field tests and adding two standard deviations, it is prompted that this segment needs to be further checked. Otherwise, mark this segment as a normal segment that can enter the fitting calculation process. Process the amplitude data and phase data of all time periods according to this idea, and after completing the segmentation, summarize the amplitude spectrum and phase spectrum within each time period into the amplitude spectrum and phase spectrum after segmentation.

[0111] The advantage of the formula is that it combines the amplitude data and phase data in ways such as multiplication and the cube of the difference, and at the same time incorporates the absolute value and square root operations, which can take into account multiple difference measures between the amplitude and the phase in the same expression, thus providing richer references for subsequent time-domain analysis.

[0112] p k,m The steps for obtaining the parameter are as follows:

[0113] This parameter represents the m-th amplitude data point in the k-th segment, which mostly appears in the post-processing stage of pulse measurement. Its value is usually between 0 and 200, depending on the quantization method of the waveform amplitude by the measurement equipment and is affected by the on-site environment or test configuration. By collecting the maximum amplitude or effective value of each pulse during the operation of the equipment and recording the corresponding timestamp, the amplitude values falling within the k-th time range are uniformly marked as p k,m , m increases from 1 to the total number of records in this paragraph. For example, through multiple on-site measurements, it can be found that the amplitude peaks of some pulses are 150 or higher. Whenever it is confirmed that this amplitude peak belongs to the k-th segment, write it into pk,m , we can get the comparable p k,m Sequence. Take the following example as an example. When k = 3, m takes p from 1 to 4 respectively. 3,1 =80, p 3,2 =120, p 3,3 =95, p 3,4 =110.

[0114] q k,m The steps to obtain the parameters are:

[0115] This parameter refers to the mth phase data point in the kth segment, and the range is usually between 180 degrees and 180 degrees. For ease of calculation, the degree can be converted to radians or the original angle unit can be maintained, but in subsequent calculations, it must be ensured that it is consistent with p k,m There will be no unit conflict in the calculation relationship. In field measurements, the phase difference of the pulse waveform relative to a certain reference zero point or relative to the moment of maximum amplitude is often given. If it is confirmed that the phase difference occurs within the kth segment of the recording sequence, it is marked as q k,m , one pulse may correspond to multiple phase detection points, all of which can be collected to form q k,m To obtain a complete sequence of , the time domain alignment must be completed in the previous step to ensure that the amplitude peak and phase information of the same pulse can be paired under the same index. Taking the following example, q 3,1 =30, q 3,2 =45,q 3,3 =60,q 3,4 =10.

[0116] The steps to obtain the k parameter are:

[0117] This parameter is used to indicate the sequence number of the time period, which is usually determined by the time axis division set in advance by the researcher. For example, if each segment is set to 2 seconds and the entire observation time is 10 seconds, then k is between 1 and 5, k=1 means the first segment is 0 to 2 seconds, k=2 means the second segment is 2 to 4 seconds, etc. In practice, it can also be divided according to other lengths. As long as the same sampling strategy is maintained in each segment and the start and end boundaries of the amplitude and phase data can be distinguished, the k value can be obtained.

[0118] The steps to obtain the m parameter are:

[0119] This parameter is used to distinguish the sequential number of each data point in the kth segment. The value starts from 1 and increases in sequence to the total number of amplitude or phase data in the segment. If a total of 50 records are collected in the same segment, then m ranges from 1 to 50. Different m corresponds to p at different observation times. k,m With q k,m, in other environments, if the sampling resolution is kept the same, a comparable sequence of m values can be obtained. m plays an indexing role in subsequent operations, bundling each amplitude and phase data with its corresponding time period k, facilitating aggregation into the same formula for overall calculation.

[0120] Substitute the parameters into the formula for calculation to obtain L3 = 0.51. This result indicates that when the amplitude and phase data within the third segment yield L3 = 0.51 according to the above operation process, this value can be regarded as a linear fitting parameter measuring the coupling degree between the amplitude and phase of this time period. If the L k values increase segment by segment beyond empirical upper limit ranges such as 1.0 or 2.0, potential anomalies can be identified and further investigated in subsequent steps as needed.

[0121] When arranging the fitting parameters of all time periods in chronological order based on the linear fitting parameter, after calculating L for each segment k it needs to be coherently sorted in the order of k. For example, k = 1 corresponds to the data within the first 2 seconds, k = 2 corresponds to the data within 2 to 4 seconds, and so on. All L k values are sorted on the time axis and checked for data missing. If the L k of any paragraph is found to be empty, it is necessary to recheck whether the amplitude or phase information of this paragraph is complete. If it is confirmed to be missing, the average value of the surrounding paragraphs can be used to fill it within the same time period. In this way, the mapping between the time period and L k is ensured to be continuous. Then, an association operation is performed in combination with the previous amplitude spectrum and phase spectrum distribution relationship. The specific method is to maintain an index record of amplitude and phase for each k segment, and make its index correspond one by one with L k If it is found that the average value of the corresponding amplitude spectrum reaches 60 but some records in the phase spectrum deviate significantly when k = 1, record the L1 value of this segment for comparison with other segments, and then observe the continuous changes of multiple segments at the time domain level. Each comparison will link the L k value with the amplitude and phase of the same paragraph, mark the time points and integrate these values into a unified sequence in order. For the specific inspection inside the device, other parameters can also be stored together. Finally, after arranging the fitting parameters of all time periods, a complete time domain fitting parameter set is jointly formed.

[0122] The steps to obtain the characteristic drift curve are as follows:

[0123] Based on the time domain fitting parameter set, extract the linear fitting parameters of each time period, obtain the voltage - current ratio of the corresponding time period, and at the same time use the sliding window method for smoothing processing to form time period matching data;

[0124] According to the time period matching data, calculate the threshold drift amount of each time period. The calculation formula is:

[0125]

[0126] Among them, T k represents the threshold drift amount in the k-th time period, and P k,m represents the m-th linear fitting parameter value in the k-th time period, and R k,m represents the m-th voltage-current ratio in the k-th time period, and M represents the total number of data points in the k-th time period;

[0127] Based on the threshold drift amount, arrange the drift data of all time periods in chronological order, and construct a characteristic drift curve including the drift characteristics of all time periods.

[0128] Specifically, based on the time-domain fitting parameter set, extract the linear fitting parameters of each time period and obtain the voltage-current ratio corresponding to the time period. First, it is necessary to find the start and end positions of each time period one by one in the previously counted time-domain fitting parameter set, and then find the records of the linear fitting parameter data points and the voltage-current ratio within each period according to these position ranges, and perform smoothing processing on the data of the same period in a sliding window manner. The specific method is to successively read several adjacent data points according to the determined window size, and use their mean or median as the smoothing result of the current window. For example, if the window size is set to 5 data points, then calculate their average value after reading these 5 data points, and then write the average value back to the corresponding index in the current paragraph, and then move forward 1 data point and repeat the operation. If a certain period has 50 data points, 46 smoothing results can be obtained. Once it is found during the smoothing process that the values of the linear fitting parameters or the voltage-current ratio of some data points are far beyond the pre-set allowable range, for example, in multiple pulse experiments, the peak value of the ratio is statistically 2.5 and a safety margin of 0.3 is added to get 2.8 as an upper threshold. When it is detected that the voltage-current ratio of a certain record reaches 3.2, a mark is made for this data point after smoothing. When returning to the final summary, the marked points can also be excluded from the subsequent calculation or saved separately. These specific processing procedures depend on the on-site requirements. After all the smoothing operations are completed, the entire record is rearranged in chronological order and the abnormal points that have been marked are explained, and these smoothed data are summarized into time period matching data.

[0129] The advantage of the formula is that it takes into account both the squared ratio of the deviation between the linear fitting parameter value and the voltage-current ratio and incorporates multi-level indicators such as the absolute product sum and the fourth power difference, thus integrating the comprehensive measurement of the degree of difference and the coupling strength in one formula.

[0130] P k,m The steps for obtaining the parameter are as follows:

[0131] This parameter represents the value of the m-th linear fitting parameter within the k-th time period. Its range can be obtained from the previous statistics. For example, the typical values of the linear fitting parameter in previous multiple tests show that it falls within the range of 0 to 2. In some specific extreme cases, it can reach 3 to 4. To accurately collect P k,m All fitting results within the k-th period need to be recorded one by one in chronological order, and each recorded value is taken as P k,m , where m increments from 1 to M. If the number of data points in the k-th period is 50, then m = 1 to 50, and a comparable sequence of P k,m values can be obtained. The following calculation examples will give specific values to illustrate how to substitute them into the formula.

[0132] R k,m The steps to obtain the parameter are as follows:

[0133] This parameter refers to the ratio of voltage to current of the m-th within the k-th time period. During on-site measurement, the voltage value and current value are often collected separately and a ratio is formed at the same time stamp. If the voltage is 48V and the current is 2A, the ratio is 24. To maintain consistency, such ratios need to be made into an (R k,1 ,R k,2 ,…,R k,M ) sequence within the same period k. The range usually varies with different device types. For some devices, the highest ratio within the working range can reach 50 or 60. When collecting, the actual values obtained during the previous operation can be referred to and their upper and lower limits can be statistically determined, and then all of them are included in the record of this paragraph. After that, matching R k,m can be obtained whether in the same device environment or similar working conditions.

[0134] The steps to obtain the M parameter are as follows:

[0135] This parameter is used to count the total number of data points within the k-th period. It is an integer related to the time period length and sampling frequency. For example, if a 10-second observation is divided into 5 time periods, then each period is 2 seconds. If 25 records are obtained per second, then each period has approximately 50 records, so M = 50. When slightly adjusting the sampling frequency or dealing with short-term fault periods, M may also fluctuate slightly. Just count the actual number of valid records retained after the segmentation to determine M, and then use this integer to calculate the range of each summation and product expression in the formula item by item.

[0136] Substituting the parameters into the formula for calculation gives 0.928. This result indicates that when the linear fitting parameter and the voltage-current ratio within the example paragraph are as listed above, the threshold drift amount is about 0.928. If the same paragraph is detected multiple times and it is found that T3 continuously approaches 2 or above 3, it can be recorded as a potential abnormal area related to large-scale drift. Subsequently, T k in different k segments can be marked and compared on the time axis and further in-depth observation can be carried out.

[0137] Based on the threshold drift amount, in the order of time period k starting from 1 to the last paragraph, read the previously calculated T one by one k , form a continuous drift data sequence, then splice this sequence in the order of k and arrange it within the same chart or the same data structure. If T1 is approximately equal to 0.5 when k = 1, T2 is approximately equal to 0.7 when k = 2, and T3 is equal to the previously calculated 0.928 when k = 3 and increases gradually in subsequent paragraphs, it indicates that this index shows a gradually rising tendency along the time axis. It is necessary to view the corresponding peak height in combination with the linear fitting parameters and the voltage-current ratio obtained previously. If the drift characteristics of some paragraphs increase excessively, record it during summarization and incorporate it into the subsequent judgment of the fault situation. Finally, after summarizing the drift amounts of all paragraphs through this method, a complete characteristic drift curve can be obtained.

[0138] The steps for obtaining the combined characteristic value are as follows:

[0139] According to the characteristic drift curve, calculate the combined characteristic value, and the calculation formula is:

[0140]

[0141] Among them, TC k represents the combined characteristic value of the k-th time period, U k (x) represents the voltage response signal function of the k-th time period, I k (x) represents the current response signal function of the k-th time period, U k,j represents the j-th voltage response signal value of the k-th time period, I k,j represents the j-th current response signal value of the k-th time period, T k,j represents the j-th threshold drift amount of the k-th time period, J represents the total number of data points within the k-th time period, and Q is the total change range of the signal within the time period.

[0142] Specifically, the advantage of the formula lies in integrating the signal function form and the discrete data form in the same expression. By performing integration and difference operations on the voltage response signal function U k (x) and the current response signal function I k (x), and performing multiple measurements such as multiplication and sum of squares on the discrete voltage and current data U k,j , I k,j and the threshold drift amount T k,j , so as to obtain an overall description of the multi-layer relationship between voltage and current within the same time period in a comprehensive formula.

[0143] The steps for obtaining the parameter of U k (x) are as follows:

[0144] This parameter represents the voltage response signal function in the k-th time period. Usually in a continuous measurement system, it is a record of how the voltage signal changes over time during this period. By using established sensors and a sampling card to track the voltage curve in real-time and through interpolation or curve fitting of the original discrete data, a functional form is obtained. The independent variable x of the function can start from the zero point of this period of time. As time progresses, U k (x) can output the corresponding voltage value. To ensure accuracy, usually, multiple periodic waveforms are first averaged or filtered to eliminate noise and instantaneous fluctuations, and then a reliable fitting method is used to construct U k (x) within the range from the start to the end of the time period. In industrial equipment, the voltage range may be between 0V and 24V, or it may be higher. In practical applications, it is necessary to combine the current device rating and the sampling upper limit. If the data curve obtained when k = 3 is relatively stable, interpolation between adjacent time points can be used to form a continuous function. For example, 100 data points are selected within a time period of 0.01 seconds, and a smooth voltage curve is obtained through fitting, which can represent the overall voltage function of the k-th segment.

[0145] I k (x) The steps to obtain the parameter are as follows:

[0146] This parameter refers to the current response signal function corresponding to U k (x), which represents the current value changing over time within the k-th time period. Similar to the voltage acquisition process, current sampling needs to be carried out at the same frequency when the on-site equipment is running, and then the obtained discrete points are denoised or curve-fitted to obtain I k (x). In industrial equipment, the current range may be between 0A and 5A, or it may vary within a wider range. The current is finely sampled through an appropriate power meter or sensor, and each sampling point is marked with the same time reference and then aligned with the start point of the k-th segment of time, and the complete curve of I k (x) can be obtained. If there are sudden change sections in this current signal, they can be checked one by one in the discrete data and excluded or down-weighted during fitting to ensure that the finally formed I k (x) function has sufficient representativeness. For example, when the instantaneous current measured at a certain moment reaches 10A while the system design allows a maximum of 5A, it is necessary to combine the actual on-site situation to judge whether it is a sensor failure or a transient jump. After establishing the corrected current curve, it can be mapped to I k (x) for subsequent calculations.

[0147] U k,j The steps to obtain the parameter are as follows:

[0148] This parameter represents the j-th voltage response signal value within the k-th time period, which is for U k(x) The result after discretization usually directly comes from the voltage sampling of on-site measurement. In the previously distinguished time slice k, for each acquired voltage data record, an index j is assigned and arranged according to the sampling time sequence. For example, if a total of 200 sampling points are collected in the k = 1 segment, then j ranges from 1 to 200, and the corresponding voltage value U k,1 , U k,2 , …, U k,200 is obtained. The numerical value may be between 0V and 24V or within the rated range of other devices. In order to better combine with integral and difference operations, some analysis methods will normalize these discrete voltage values again, and some people will directly use the original values, which depends on the accuracy requirements of on-site applications. If other observers measure with the same time segmentation strategy and sampling frequency, they can also obtain U k,j sequence under the same index j. The following example part will show how to bring U k,j into the operation.

[0149] I k,j The steps to obtain the parameter are as follows:

[0150] This parameter corresponds to U k,j and represents the current response signal value at the j-th sampling moment in the k-th time period. The acquisition method is the same as the discrete sampling before the construction of the current function I k (x). In the range of k segments, the current value at each moment is recorded, and the j index is used for sequential marking. If 250 current points are collected in this paragraph, then j ranges from 1 to 250. Each record corresponds uniquely to its sampling moment and is consistent with the time reference of voltage sampling to ensure that it can be subtracted or squared with U k,j during the operation. If there are spikes or zero drifts in the current, it needs to be identified during the sampling process and reasonably corrected or confirmed before forming the I k,j sequence, and then the matching I k,j sequence can be obtained.

[0151] T k,j The steps to obtain the parameter are as follows:

[0152] This parameter represents the threshold drift amount at the j-th in the k-th time period, usually the calculation result in the previous formula or discrimination process. During the pulse measurement or the sampling period of industrial equipment, by recording the aforementioned linear fitting parameters and ratio information, and then substituting them into the previously introduced threshold drift formula to calculate T k , and then combining the sampling index j to split the local offset values of each data point in the same time period, the sequence of T k,j can be formed. For example, for an observation of 2 seconds in length, if the local drift change is calculated every 0.02 seconds, 100 T k,jRecords, with j ranging from 1 to 100, each reflecting the threshold drift at that time point.

[0153] The steps to obtain the J parameter are as follows:

[0154] This parameter is the total number of data points within the kth time period and is used to and perform upper and lower index limits on operations such as. Its specific size is directly related to the sampling frequency and the length of the time period. For example, if 1000 samples are taken per second and the length of the kth segment is 0.5 seconds, then J is equal to 500. The sampling rate or segment size will be set on-site according to the required resolution and device bandwidth. All subsequent operations on the same segments can use the same J value as the upper limit index. As long as the actual number of remaining data points is checked before recording, J can be confirmed.

[0155] The steps to obtain the Q parameter are as follows:

[0156] This parameter represents the total change range of the signal within the time period and is used to perform integral bounds between U k (x) and I k (x). Understanding from 0 to Q can be regarded as a complete coverage of this section of the voltage and current functions in terms of time or sample magnitude. If this section lasts for 0.01 seconds and it is desired to smooth x from 0 to 0.01 after interpolation, then Q is taken as 0.01; if x represents the total number of sampling points, Q can also be taken as its maximum value according to the number of sampling points. The key is to ensure

[0157] that it can correctly reflect the voltage-current difference within this section.

[0158] Substituting the parameters into the formula gives 0.171. This result indicates that when the voltage and current functions within the second time period are in the interval of x = 0 to 0.02 seconds, and the corresponding discrete data and threshold drift amounts are as above, the combined eigenvalue is around 0.171. If the same process is applied to other time periods and it is found that TC k is above 1.0, then the interval can be marked as an interval with significant voltage-current differences and can also be further used for difference comparison with standard threshold data and subsequent classification discrimination.

[0159] The steps to obtain the damage type mapping table are as follows:

[0160] Obtain the standard threshold data, arrange the combined eigenvalues in chronological order to form the combined eigenvalues and standard threshold data;

[0161] Based on the combined eigenvalues and standard threshold data, calculate the difference amount for each time period, match the calculated difference amounts along the time axis, and arrange all the difference amounts in the chronological order of the data points to obtain the difference amount between the combined value and the standard threshold;

[0162] Define the damage type discrimination rules based on the difference between the combined value and the standard threshold. Classify the damage types corresponding to all data points according to the preset classification threshold, and map the data of different damage types to the corresponding categories to establish a damage type mapping table.

[0163] Specifically, obtain the standard threshold data, arrange the combined feature values in chronological order. It is necessary to first accumulate the corresponding standard ranges in multiple device detections or pulse tests, and obtain a sequence of one or more reference values by referring to the device user manual or existing measured statistical results. After confirming that the combined feature values of each observation paragraph have been calculated, pair them with the standard threshold data in chronological order, and check whether the combined feature values of some paragraphs exceed the upper limit range formed by multiple previous measurements during the comparison process. For example, if the highest combined feature value recorded in 10 tests is 0.85 and 0.15 is added as a buffer to get 1.0, if the actual measured combined feature value exceeds 1.0, it will be included in the interval that needs to be key checked. Subsequently, arrange the combined feature values of all paragraphs and the corresponding standard thresholds in the same sequence and mark the deviation of each paragraph. The deviation is the difference between the actual combined feature value of this paragraph and the standard threshold. If the deviation is greater than the alarm value set based on previous experience, such as 0.2, then mark this paragraph as potentially abnormal. After collecting the deviation data corresponding to all paragraphs, the corresponding difference amounts are arranged in sequence and matched with the specific time of each paragraph. Finally, summarize the overall difference between the combined feature value and the standard threshold to observe its change situation in chronological order.

[0164] Define the damage type discrimination rules based on the difference between the combined value and the standard threshold. Usually, list some classification thresholds in combination with industry standards or typical failure modes statistically obtained in multiple on-site tests. For example, classify the difference amount greater than 0.5 as the high-risk area, record the difference amount in the range of 0.2 to 0.5 as the medium-risk area, and regard it as the normal range below 0.2. Subsequently, judge the difference amount corresponding to each data point to see which interval it falls into. If the difference amount of a certain paragraph reaches the high-risk area standard, directly mark this paragraph as the serious damage type. If it is only in the medium-risk range, record it as the general damage type. After completing the judgment of all paragraphs, map the results of each paragraph to the pre-defined category labels. For example, label A represents minor damage, and label B represents serious damage. Then organize these mapping relationships into a damage type mapping table. If in the actual production line or experimental link, users need to conduct subsequent statistics or early warnings on different types of damage information, they can quickly find the corresponding records according to this mapping table and retrieve the detailed data of the abnormal paragraphs in combination with the time axis, so as to form a judgment result on the overall operation status.

[0165] The above are only the preferred embodiments of the present invention, and do not limit the present invention in other forms. Any person skilled in the art may use the technical content disclosed above to make changes or modifications into equivalent embodiments with equivalent changes and apply them to other fields. However, as long as it does not depart from the technical solution content of the present invention, any simple modification, equivalent change and modification made to the above embodiments based on the technical essence of the present invention still fall within the protection scope of the technical solution of the present invention.

Claims

1. A diagnosis system for damage of a voltage-sensitive device combined with pulse testing, characterized in that It includes the following modules: The original signal acquisition module inputs a pulse sequence to the piezoresistive device, respectively records the voltage response signal and current response signal under each pulse sequence, and generates a response signal set; Based on the response signal set, wavelet decomposition is performed on the voltage response signal and current response signal to obtain a time-frequency feature matrix; The signal wavelet decomposition module calculates the instantaneous frequency and phase value for the high-frequency coefficient group and low-frequency coefficient group respectively based on the time-frequency feature matrix, and generates a frequency-phase vector; based on the frequency-phase vector, the instantaneous frequency and phase value are reconstructed into a frequency sequence according to the time series, and Hilbert transform is performed on the frequency sequence to generate a fault feature spectrum; The fault feature spectrum generation module segments the amplitude spectrum and phase spectrum on the time axis based on the fault feature spectrum, calculates the linear fitting parameters for each segment of the amplitude spectrum and phase spectrum, and generates a time-domain fitting parameter set; based on the time-domain fitting parameter set, combines the linear fitting parameters with the voltage-current ratio within the current time period to establish a characteristic drift curve; The fault diagnosis and mapping module numerically combines the threshold drift amount with the voltage response signal and current response signal in the current time period based on the characteristic drift curve to generate a combined characteristic value, calculates the difference between the combined value and the standard threshold based on the combined characteristic value, discriminates and classifies the damage type of the piezoresistive device according to the difference, and establishes a damage type mapping table.

2. The diagnosis system for damage of the voltage-sensitive device combined with pulse testing according to claim 1, wherein The acquisition steps of the response signal set are as follows: Connect the piezoresistive device to the pulse generator, start the emission of the pulse sequence, and record the voltage response signal and current response signal caused by each pulse sequence to obtain preliminary response data; Based on the preliminary response data, classify and organize the voltage response signal and current response signal to obtain comprehensive response data; Based on the comprehensive response data, integrate all the voltage response signals and current response signals to form a response signal set.

3. The diagnosis system for damage of a voltage-sensitive device combined with a pulse test according to claim 1, wherein The acquisition steps of the time-frequency feature matrix are as follows: Based on the response signal set, wavelet decomposition is respectively performed on the voltage response signal and current response signal, the wavelet coefficients of each signal at different scales are extracted, and the wavelet decomposition coefficients are classified according to a preset frequency interval to obtain a high-frequency coefficient group and a low-frequency coefficient group; According to the high-frequency coefficient group and low-frequency coefficient group, calculate the energy distribution value of each group, and the calculation formula is: Among them, E k represents the energy distribution value of the k-th group, and C k,m represents the absolute value of the m-th wavelet decomposition coefficient in the k-th group, and D k,m represents the deviation term of the m-th coefficient in the k-th group, and QM is the number of wavelet coefficients; Based on the energy distribution value, construct a time-frequency feature matrix, and fill the energy distribution value of each group in the time-frequency feature matrix.

4. The diagnosis system for damage of the voltage-sensitive device combined with pulse testing according to claim 1, wherein The acquisition steps of the frequency-phase vector are as follows: Based on the time-frequency feature matrix, extract the components of the high-frequency coefficient group and low-frequency coefficient group respectively, calculate the instantaneous amplitude change rate for each component, and obtain the instantaneous change characteristic data by combining the time information; Based on the instantaneous change characteristic data, calculate the instantaneous frequency and phase value, organize the time series information of each component, and associate the instantaneous frequency and phase change conditions of each component to obtain the instantaneous frequency data and phase data; Based on the instantaneous frequency data and phase data, match and classify the data according to the corresponding frequency interval, and combine the corresponding instantaneous frequency and phase values into a frequency-phase vector.

5. The diagnosis system for damage of a voltage-sensitive device combined with pulse testing according to claim 1, characterized in that The acquisition steps of the fault feature spectrum are as follows: Based on the frequency-phase vector, extract the instantaneous frequency and phase value corresponding to each time point, sort them in the order of time series, and perform interpolation and completion. At the same time, smooth the frequency points with abnormal changes to obtain the reconstructed frequency series; Perform Hilbert transform on the reconstructed frequency series, and calculate the instantaneous amplitude spectrum value after the transform. The calculation formula is: Among them, A j is the instantaneous amplitude spectrum value corresponding to the j-th frequency point, P j is the original component value in the frequency sequence, is the component after applying the Hilbert transform to P j ; Based on the instantaneous amplitude spectrum value, construct a feature matrix including amplitude information, classify the amplitude values corresponding to each time point according to the frequency distribution, and arrange the classified data in time series to form a fault feature spectrum.

6. The diagnosis system for damage of a voltage-sensitive device combined with a pulse test according to claim 1, wherein The steps for obtaining the time-domain fitting parameter set are as follows: Based on the fault feature spectrum, divide it into multiple time periods along the time axis, extract the amplitude data and phase data within each time period to form the segmented amplitude spectrum and phase spectrum; According to the segmented amplitude spectrum and phase spectrum, calculate the linear fitting parameters of each time period. The calculation formula is: where L k represents the linear fitting parameter of the k-th segment, p k,m represents the m-th amplitude data point within the k-th segment, q k,m represents the m-th phase data point within the k-th segment; Based on the linear fitting parameters, arrange the fitting parameters of all time periods in chronological order, and construct a complete time-domain fitting parameter set. Associate the fitting parameters of each time period according to the distribution relationship between the amplitude spectrum and the phase spectrum to generate a time-domain fitting parameter set.

7. The diagnosis system for damage of a voltage-sensitive device combined with pulse testing according to claim 1, characterized in that, The steps for obtaining the characteristic drift curve are as follows: Based on the time-domain fitting parameter set, extract the linear fitting parameters of each time period, obtain the voltage-current ratio corresponding to the corresponding time period, and at the same time use the sliding window method for smoothing processing to form time period matching data; According to the time period matching data, calculate the threshold drift amount of each time period. The calculation formula is: Among them, T k represents the threshold drift amount in the k-th time period, P k,m represents the m-th linear fitting parameter value in the k-th time period, R k,m represents the m-th voltage-current ratio in the k-th time period, and M represents the total number of data points in the k-th time period; Based on the threshold drift amount, arrange the drift data of all time periods in chronological order, and construct a characteristic drift curve including the drift characteristics of all time periods.

8. The diagnosis system for damage of the voltage-sensitive device combined with pulse testing according to claim 1, wherein The steps for obtaining the combined eigenvalue are as follows: According to the characteristic drift curve, calculate the combined eigenvalue. The calculation formula is: Among them, TC k represents the combined characteristic value in the k-th time period, U k (x) represents the voltage response signal function in the k-th time period, I k (x) represents the current response signal function in the k-th time period, U k,j represents the j-th voltage response signal value in the k-th time period, I k,j represents the j-th current response signal value in the k-th time period, T k,j represents the j-th threshold drift amount in the k-th time period, J represents the total number of data points in the k-th time period, and Q is the total change range of the signal within the time period.

9. The diagnosis system for damage of a voltage-sensitive device combined with pulse testing according to claim 1, wherein The steps for obtaining the damage type mapping table are as follows: Obtain the standard threshold data, arrange the combined eigenvalues in chronological order to form the combined eigenvalues and standard threshold data; Based on the combined eigenvalues and standard threshold data, calculate the difference amount of each time period, match the calculated difference amounts along the time axis, and arrange all the difference amounts in the time order of the data points to obtain the difference amount between the combined value and the standard threshold; Based on the difference amount between the combined value and the standard threshold, define the damage type discrimination rule, classify the damage types corresponding to all data points according to the preset classification threshold, and map the data of different damage types to the corresponding categories to establish a damage type mapping table.

Images