A fault diagnosis method for shore power switching devices

Abstract

This invention belongs to the field of shore power transfer fault analysis technology. It discloses a fault diagnosis method for shore power transfer devices, comprising: generating monitoring events based on changes in opening and closing of circuit breakers and step changes in load, and extracting steady-state windows before and after the events; obtaining an equivalent contact impedance sequence by analyzing the linear consistency of voltage difference with the root mean square of current; calculating the exponentially weighted moving average and cumulative offset on the equivalent contact impedance sequence to identify drift and step change regions and generate a candidate set of impedance anomalies; calculating the high-frequency energy ratio and peak count of the current waveform within the transient window corresponding to the monitoring event, and constructing a candidate set of arcing anomalies by analyzing the fluctuation changes in the high-frequency energy ratio and the clustering changes in the peak count; and obtaining the fault phase and fault location by taking the intersection of the candidate set of impedance anomalies and the candidate set of arcing anomalies. This invention solves the problem of difficulty in early detection of contact deterioration during operation.

Description

Technical Field

[0001] This invention relates to the field of power transfer fault analysis technology, and more specifically, to a fault diagnosis method for shore power transfer devices. Background Technology

[0002] Shore power switching devices are used to electrically connect and switch shore power supplies at docks with the ship's power receiving system. They typically integrate functions such as switching control, phase sequence and interlock confirmation, metering, and protection, allowing ships to be powered by shore while berthed to reduce fuel consumption and emissions. These devices often operate in environments involving salt spray, humidity, vibration, and frequent plugging and unplugging operations. Maintaining connection reliability and switching safety is crucial under different ship types and load conditions. Therefore, online monitoring and fault diagnosis of their operational status is of great significance.

[0003] Current operation and maintenance methods largely rely on periodic inspections, temperature rise observations, or confirmation of contact and insulation status during shutdown windows, combined with experience-based judgment based on protection actions, alarm records, or electrical parameter fluctuations. In practical applications, the slow rise in contact impedance caused by contact oxidation, terminal loosening, etc., is often not easily apparent during steady-state operation, but is more likely to be exposed during transient processes such as opening and closing or load step changes, making it difficult to quantify and trace early degradation in a timely manner. At the same time, if online monitoring is judged only from the single perspective of steady-state electrical parameters or transient disturbances, the diagnostic results are prone to being unclear when factors such as load disturbances, switching transients, and sampling noise are present, making it difficult to further converge to specific phases and maintainable parts, thereby affecting the pertinence of maintenance strategies and the efficiency of fault handling. Summary of the Invention

[0004] To overcome the aforementioned deficiencies of the prior art and to achieve the above objectives, the present invention provides the following technical solution: a fault diagnosis method for a shore power switching device, comprising:

[0005] Read the opening and closing status and three-phase voltage and current sequence from the shore power transfer device, generate monitoring events based on the changes in opening and closing and the step changes in load, and capture the steady-state window before and after the event.

[0006] Based on the steady-state window, the root mean square voltage and root mean square current of each phase are calculated. By analyzing the linear consistency of the voltage difference with the root mean square current, the equivalent contact impedance sequence is obtained.

[0007] Calculate the exponentially weighted moving average and cumulative offset on the equivalent contact impedance sequence to identify drift and step change regions and generate a candidate set of impedance anomalies.

[0008] Within the transient window corresponding to the monitoring event, the high-frequency energy ratio and spike count of the current waveform are calculated. By analyzing the fluctuation changes of the high-frequency energy ratio and the clustering changes of the spike count, a candidate set of arcing anomalies is constructed.

[0009] The intersection of the impedance anomaly candidate set and the arcing anomaly candidate set is used to obtain the fault phase and the fault location.

[0010] Preferably, the method for generating monitoring events and capturing steady-state windows before and after the events includes:

[0011] Monitor the connection status of the circuit breaker auxiliary contacts and plug interlock switch. When a change in the connection status is detected, record the connection status and timestamp, and generate a monitoring event.

[0012] Generate event sequence numbers for monitored events, and align the three-phase voltage and current sequences according to the event sequence numbers to obtain the event-aligned sequence;

[0013] Within the event alignment sequence, calculate the increment and duration of the current amplitude within the preset load monitoring period. If the increment exceeds the preset transition threshold and the timing duration reaches the preset minimum duration, determine that a load change event has occurred, record the timestamp, and update the monitoring event.

[0014] Based on the start and end boundaries of the monitored events, the intervals where the rate of change of current is lower than the stable threshold are identified and recorded as the steady-state intervals.

[0015] Obtain the median timestamp of the steady-state interval and record it as the representative time of the steady-state interval. If the representative time is before the generation time of the monitored event, it is recorded as the pre-steady-state window of the corresponding monitored event; if the representative time is after the generation time of the monitored event, it is recorded as the post-steady-state window of the corresponding monitored event.

[0016] Both the pre-steady-state window and the post-steady-state window are steady-state windows. The statistics of the steady-state window are obtained and written into the monitoring event.

[0017] Preferably, the method for obtaining the equivalent contact impedance sequence includes:

[0018] Within each steady-state window, calculate the root mean square of the incoming line voltage, the root mean square of the outgoing line voltage, and the root mean square of the current for different phases. Calculate the difference between the root mean square of the incoming line voltage and the root mean square of the outgoing line voltage to obtain the voltage difference.

[0019] Calculate the ratio of voltage difference to root mean square current to obtain the candidate impedance points for the corresponding phase.

[0020] Robust linear fitting is performed on candidate impedance points of the same phase under multiple continuous monitoring events to obtain the equivalent contact impedance sequence and the fitting residual sequence.

[0021] Preferably, the method for identifying drift and step change regions and generating a candidate set of impedance anomalies includes:

[0022] Read the equivalent contact impedance values ​​sorted by time on the monitoring event sequence of the same phase to generate an impedance input sequence;

[0023] The exponentially weighted moving average is updated on the impedance input sequence according to the time interval of the monitoring events to obtain the impedance smoothing sequence and retain the smoothed residual for each monitoring event;

[0024] Smoothed residuals of the same phase are spliced ​​together according to the time sequence of the monitored events to generate a residual window;

[0025] Identify the median in the residual window, calculate the absolute deviation of the median in the residual window, denoted as the robust dispersion, and calculate the noise threshold based on the robust dispersion.

[0026] By analyzing the consistency between the continuous positive slope of the impedance smoothing sequence and the sustained growth of the cumulative offset, drift regions can be identified.

[0027] By analyzing the synchronicity between the fitted changes in impedance values ​​and the crossing of cumulative offset between monitored events, step abrupt change regions of cumulative offset can be identified.

[0028] By integrating drift regions and step change regions, a candidate set of impedance anomalies is generated.

[0029] Preferably, the method for identifying drift regions includes:

[0030] Calculate the difference between the smoothed residual and the noise threshold to obtain the effective offset. If the effective offset is not greater than 0, then the effective offset is set to 0.

[0031] The effective offset of the residual window is accumulated according to the event sequence of the monitored events. The accumulated offset is obtained and updated. When the accumulated offset does not increase under the monitoring event corresponding to the preset fallback confirmation length, it is determined that the offset has disappeared and the accumulated offset is set to 0.

[0032] Calculate the smoothing slope between the equivalent contact impedances of adjacent monitoring events on the impedance smoothing sequence, and identify the smoothing slope region where the smoothing slope is continuously positive.

[0033] Mark several consecutive monitoring events corresponding to the smooth slope region as target events, and concatenate the cumulative offsets in the residual window according to the time order of the target events to obtain the cumulative offset sequence.

[0034] Determine if the cumulative offset sequence is increasing. If the cumulative offset is increasing, then generate a drift region based on consecutive target events.

[0035] Preferably, the method for identifying step abrupt change regions of cumulative offset includes:

[0036] For the same phase impedance input sequence, calculate the difference between the equivalent contact impedance value of each monitoring event and the exponentially weighted moving average value under the adjacent previous monitoring event, and record the result as the single-event surge amplitude of the previous monitoring event;

[0037] The robustness dispersion and noise threshold are fused to obtain the surge threshold. When the surge amplitude of a single event is greater than the surge threshold, the corresponding monitoring event is marked as a trigger event.

[0038] After the event is triggered and within the monitoring event range of the preset fallback confirmation length, retrieve the cumulative offset that is greater than the noise threshold and mark it as a cross offset;

[0039] Mark the monitoring event corresponding to the last valid offset that has been generated across offsets as the triggering event;

[0040] A step mutation region is generated based on the triggering event, the inducing event, and the monitoring event that is between the triggering event and the inducing event in terms of timestamp.

[0041] Preferably, the method for constructing the candidate set of arcing anomalies includes:

[0042] Within the transient window corresponding to the monitoring event, the original waveforms of the current in each phase are bandpass filtered to obtain a high-frequency component sequence;

[0043] Zero-crossing detection is performed on the original waveforms of the current in each phase to obtain zero-crossing sampling point pairs, and the zero-crossing time is calculated by interpolation;

[0044] The ratio of high-frequency energy to full-frequency energy is calculated based on the high-frequency component sequence to generate the high-frequency energy percentage.

[0045] By analyzing the number of spike amplitudes and the cluster length of consecutive spikes in a short time segment near the zero crossover moment, spike count and clustering characteristics are generated.

[0046] The arcing index of the monitored event is obtained by weighted fusion of high-frequency energy ratio, peak count and aggregation characteristics. Based on the abnormal rise of the arcing index, an arcing anomaly candidate set is constructed.

[0047] Preferably, the method for generating peak counts and clustering characteristics includes:

[0048] For each phase current waveform, extract the zero-crossing time sequence within the transient window, and extract symmetrical short time segments based on the zero-crossing times;

[0049] The local maxima of the high-frequency component sequence within the zero-crossing segment are statistically analyzed to obtain the peak amplitude. The peak amplitudes are then spliced ​​together in chronological order to obtain the peak amplitude sequence.

[0050] The maximum value of the peak amplitude under recent no-alarm monitoring events is obtained as the amplitude threshold. Peak amplitudes exceeding the amplitude threshold are marked as over-threshold peaks. The number of over-threshold peaks is counted and recorded as the peak count.

[0051] All peaks exceeding the threshold are sorted by time, peak clusters formed by several adjacent peaks are identified, the number of peaks in each peak cluster is calculated as the cluster length, and the maximum cluster length is identified.

[0052] Determine whether there is a threshold peak within a short time segment; if so, mark the threshold peak as the target peak.

[0053] Determine whether different target peaks belong to the same peak cluster. If different target peaks belong to the same peak cluster, then determine that the short segment is contaminated by the peak cluster and mark it as a contaminated segment.

[0054] The proportion of the number of contaminated fragments to the total number of short-time fragments in the transient window is denoted as the cluster duty cycle. This is then combined with the maximum cluster length to obtain the clustering characteristic.

[0055] Preferably, the method for constructing the candidate set of arcing anomalies includes:

[0056] Obtain the arcing index for each monitoring event. When the increase in the arcing index of adjacent monitoring events exceeds the surge threshold, both adjacent monitoring events are marked as arcing abnormal events.

[0057] Obtain the arc index of historical monitoring events, calculate the arc index histogram, sort the arc index intervals according to the median of the arc index intervals, and obtain a list of arc index intervals;

[0058] Mark the first three arc index intervals in the arc index interval list as the fusion interval, normalize the distribution probability of the fusion interval, and obtain the interval fusion weight.

[0059] The anomaly detection threshold is obtained by weighting the median of the fusion interval based on the interval fusion weight.

[0060] When the arcing index of a monitored event exceeds the anomaly determination threshold, it is marked as an arcing abnormal event, and an arcing abnormal candidate set is constructed based on the abnormal events.

[0061] Preferably, the method for obtaining the fault phase and fault location includes:

[0062] Perform an intersection operation on the candidate set of impedance anomalies and the candidate set of arcing anomalies, and record the result as a fault mapping event;

[0063] Obtain the wiring topology description data of the fault mapping event, map it to maintainable component nodes, and output the fault location and fault type.

[0064] The technical effects and advantages of the fault diagnosis method for shore power switching device of the present invention are as follows:

[0065] (1) By generating monitoring events from opening and closing and load step changes and extracting steady-state windows before and after the events, the root mean square of voltage and root mean square of current of each phase are calculated within the steady-state window to form voltage difference-current sample pairs. The equivalent contact impedance sequence is obtained by robust linear fitting. Then, the exponential weighted moving average and cumulative offset are calculated on the impedance sequence to identify drift and step change regions and form a candidate set of impedance anomalies. Thus, the small changes in the contact state are transformed into trends and abrupt changes that can be continuously tracked, solving the problem that contact deterioration is difficult to detect in advance during operation.

[0066] (2) By bandpass filtering the current waveform within the transient window corresponding to the monitoring event, a high-frequency component sequence is obtained. The peak count and aggregation characteristics of the short-time segment near the zero crossing are statistically combined with the high-frequency energy ratio to obtain the arcing index and construct an arcing anomaly candidate set. Then, the intersection of the impedance anomaly candidate set and the arcing anomaly candidate set is taken and combined with the wiring topology to map the fault phase and fault location of the maintainable component node output. Thus, the diagnosis results are converged with the consistency of the dual evidence of "steady-state impedance change evidence + transient arcing precursor evidence", which solves the problem that it is difficult to accurately point to the phase and location and the interpretability of the location based solely on conventional lateral anomalies. Attached Figure Description

[0067] Figure 1 This is a schematic flowchart of a fault diagnosis method for a shore power transfer device according to the present invention.

[0068] Figure 2 This is a schematic diagram of the method for identifying drift and step change regions and generating a candidate set of impedance anomalies in a fault diagnosis method for a shore power transfer device according to the present invention.

[0069] Figure 3 This is a flowchart illustrating the method for constructing a candidate set of arcing anomalies in a fault diagnosis method for a shore power transfer device according to the present invention. Detailed Implementation

[0070] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0071] This application provides a fault diagnosis method for shore power transfer devices. Based on monitoring events generated by circuit breaker opening / closing and load step changes, the method calculates the linear consistency of voltage difference-current root mean square within steady-state windows before and after the circuit breaker opening / closing to obtain an equivalent contact impedance sequence. It then uses exponential weighting and cumulative offset to identify drift and abrupt changes to form impedance anomaly candidates. Simultaneously, it extracts the proportion of high-frequency energy and peak aggregation within transient windows to construct arcing candidates. By taking the intersection of the two candidates and combining it with topological mapping, the method locates the faulty phase and location, solving the problems of difficulty in early detection of contact degradation during operation and difficulty in accurately identifying the phase and location.

[0072] Please see Figure 1 , Figure 2 and Figure 3 In this embodiment of the invention, a fault diagnosis method for a shore power transfer device is implemented in detail through the following steps:

[0073] Read the opening and closing status and three-phase voltage and current sequence from the shore power transfer device, generate monitoring events based on the changes in opening and closing and the step changes in load, and capture the steady-state window before and after the event.

[0074] Methods for generating monitoring events and capturing steady-state windows before and after the events include:

[0075] Monitor the connection status of the circuit breaker auxiliary contacts and plug interlock switch. When a change in the connection status is detected, record the connection status and timestamp, and generate a monitoring event.

[0076] The monitored events are stored using a fixed-field data structure. The fields include at least the event identifier, event type, phase identifier, channel identifier, start timestamp, end timestamp, and trigger source identifier. In this embodiment, connection status changes include, for example, the closing and opening of the circuit breaker auxiliary contacts, and the locking and unlocking of the plug interlocking switch. The trigger source identifier is determined by one of the opening or closing commands issued by the circuit breaker auxiliary contacts, the interlocking switch, or the controller.

[0077] Generate event sequence numbers for monitored events, and align the three-phase voltage and current sequences according to the event sequence numbers to obtain the event-aligned sequence;

[0078] Among them, the event sequence number is a monotonically increasing count generated by the metering module or controller of the transfer device for data acquisition.

[0079] Within the event alignment sequence, calculate the increment and duration of the current amplitude within the preset load monitoring period. If the increment exceeds the preset transition threshold and the timing duration reaches the preset minimum duration, determine that a load change event has occurred, record the timestamp, and update the monitoring event.

[0080] In this embodiment, the preset load monitoring period is an integer multiple of several consecutive sampling periods; the preset transition threshold is obtained by multiplying the current measurement error upper limit by 10 times and the rated current to get the amplified error amount, and comparing the upper limit of steady-state jitter during the debugging phase with the larger value as the preset transition threshold; wherein, the upper limit of current measurement error is the historical maximum relative error; the upper limit of steady-state jitter during the debugging phase is the maximum value of the current difference between two adjacent points calculated within 5 minutes when the load remains unchanged during factory debugging; the preset minimum duration is obtained by acquiring 5 times the sampling period and 2 times the root mean square current statistics. The window length and the corresponding contact bounce time of the device are compared, and the longest data duration is taken as the corresponding preset minimum duration. Among them, the contact bounce time refers to the time from the first contact contact (or separation) to the final stable contact state of the contactor or circuit breaker when it is just closed or opened. The contacts do not immediately "stablely stick together" or "completely separate". Instead, due to mechanical impact and elastic rebound, there are several very short repeated contact-separation. The upper limit value can be taken from the equipment manufacturer's data: such as the "bounce time" item commonly found in the technical parameters of circuit breakers, contactors and relays.

[0081] Based on the start and end boundaries of the monitored events, the intervals where the rate of change of current is lower than the stability threshold are identified and recorded as the steady-state intervals; where the stability threshold is the median of the rate of change of current during the historical steady-state periods.

[0082] Obtain the median timestamp of the steady-state interval and record it as the representative time of the steady-state interval. If the representative time is before the generation time of the monitored event, it is recorded as the pre-steady-state window of the corresponding monitored event; if the representative time is after the generation time of the monitored event, it is recorded as the post-steady-state window of the corresponding monitored event.

[0083] Both the pre-steady-state window and the post-steady-state window are steady-state windows. The statistics of the steady-state window are obtained and written into the monitoring event.

[0084] Among them, the window statistics include at least the root mean square value of the three-phase voltage, the root mean square value of the three-phase current, and the corresponding variance;

[0085] The root mean square of voltage and root mean square of current for each phase are calculated based on the steady-state window. By analyzing the linear consistency of the voltage difference with the root mean square of current, the equivalent contact impedance sequence is obtained.

[0086] Methods for obtaining the equivalent contact impedance sequence include:

[0087] Within each steady-state window, calculate the root mean square of the incoming line voltage, the root mean square of the outgoing line voltage, and the root mean square of the current for different phases. Calculate the difference between the root mean square of the incoming line voltage and the root mean square of the outgoing line voltage to obtain the voltage difference.

[0088] Calculate the ratio of voltage difference to root mean square current to obtain the candidate impedance points for the corresponding phase.

[0089] When only a single-sided voltage is available, the root mean square voltage offset relative to the baseline and the root mean square current under the same monitoring event are used together to form a candidate impedance point; the baseline is updated on a rolling basis by the health interval samples after maintenance confirmation.

[0090] Robust linear fitting is performed on candidate impedance points of the same phase under multiple consecutive monitoring events to obtain the equivalent contact impedance sequence and the fitting residual sequence.

[0091] In this method, robust linear fitting employs an iterative weighted least squares approach. Initial parameters are obtained from standardized samples using ordinary least squares. In each iteration, the predicted voltage difference and residual for each sample are calculated first. Then, the weights are updated according to the rule that "if the absolute value of the residual is less than the inflection threshold, the weight is set to one; if the absolute value of the residual is greater than the inflection threshold, the weight is reduced by the ratio of the inflection threshold to the absolute value of the residual." The linear parameters are then recalculated under the new weights. The inflection threshold can be obtained by multiplying the robust scale of the residual from the previous iteration by a preset factor. The iteration termination condition can be that the relative change in parameters is lower than the change threshold or the number of iterations reaches the upper limit, thereby suppressing the influence of outliers on the slope without introducing additional model libraries. In this embodiment, the preset factor is 2.5, the change threshold is 0.5%, and the upper limit of the number of iterations is 15.

[0092] Calculate the exponentially weighted moving average and cumulative offset on the equivalent contact impedance sequence to identify drift and step change regions and generate a candidate set of impedance anomalies.

[0093] Methods for identifying drift and step change regions and generating a candidate set of impedance anomalies include:

[0094] Read the equivalent contact impedance values ​​sorted by time on the monitoring event sequence of the same phase to generate an impedance input sequence;

[0095] The exponentially weighted moving average is updated on the impedance input sequence according to the time interval of the monitoring events to obtain the impedance smoothing sequence and retain the smoothed residual for each monitoring event;

[0096] In this method, the exponentially weighted moving average uses adaptively updated coefficients over time intervals. Let the time interval for the nth event be Δt, and the smoothing coefficient be... , The smoothed time constant, calculated in minutes or number of events, is updated smoothly. Smooth residuals In the formula, This is the estimated equivalent contact impedance value corresponding to the nth monitoring event; This is the smoothed value of the exponentially weighted moving average at time n; The smoothed residual corresponding to the nth monitoring event;

[0097] Smoothed residuals of the same phase are spliced ​​together according to the time sequence of the monitored events to generate a residual window;

[0098] In this embodiment, a residual window is constructed by selecting between 20 and 60 to cover multiple monitoring events corresponding to closing and load step events.

[0099] Identify the median in the residual window, calculate the absolute deviation of the median in the residual window, denoted as the robust dispersion, and calculate the noise threshold based on the robust dispersion.

[0100] Where med is the median of the residual window, and the robustness of dispersion and noise threshold are calculated using the following formulas:

[0101] ; ;

[0102] In the formula, MAD represents the robustness of dispersion; is the noise threshold; median() is the median operator, used to sort a set of numbers and take the middle value (for odd-numbered samples, take the middle value; for even-numbered samples, the average of the two middle values ​​is usually taken); q is the amplification factor, used to cover the upper edge of normal fluctuations; in this embodiment, the amplification factor q is preferably 3;

[0103] By analyzing the consistency between the continuous positive slope of the impedance smoothing sequence and the sustained growth of the cumulative offset, drift regions can be identified.

[0104] Methods for identifying drift regions include:

[0105] Calculate the difference between the smoothed residual and the noise threshold to obtain the effective offset. If the effective offset is not greater than 0, then the effective offset is set to 0.

[0106] Among them, it can be expressed by formula Calculate the effective offset When smoothing residuals Not greater than the noise threshold When the noise level is not exceeded, the effective offset is set. The value is 0, thus preventing the subsequent cumulative amount from increasing due to normal fluctuations;

[0107] The effective offset of the residual window is accumulated according to the event sequence of the monitored events. The accumulated offset is obtained and updated. When the accumulated offset does not increase under the monitoring event corresponding to the preset fallback confirmation length, it is determined that the offset has disappeared and the accumulated offset is set to 0.

[0108] In this embodiment, the preset fallback confirmation length is 5, that is, if 5 consecutive monitoring events meet the condition that the cumulative offset has not increased, that is, the effective offset of 5 consecutive monitoring events is 0, and the cumulative offset is set to 0.

[0109] Calculate the smoothing slope between the equivalent contact impedances of adjacent monitoring events on the impedance smoothing sequence, and identify the smoothing slope region where the smoothing slope is continuously positive.

[0110] Mark several consecutive monitoring events corresponding to the smooth slope region as target events, and concatenate the cumulative offsets in the residual window according to the time order of the target events to obtain the cumulative offset sequence.

[0111] Determine if the cumulative offset sequence is increasing. If the cumulative offset is increasing, generate a drift region based on consecutive target events.

[0112] By analyzing the synchronicity between the fitted changes in impedance values ​​and the crossing of cumulative offset between monitored events, step abrupt change regions of cumulative offset can be identified.

[0113] Methods for identifying step abrupt change regions in cumulative offset include:

[0114] For the same phase impedance input sequence, calculate the difference between the equivalent contact impedance value of each monitoring event and the exponentially weighted moving average value under the adjacent previous monitoring event, and record the result as the single-event surge amplitude of the previous monitoring event;

[0115] The robustness dispersion and noise threshold are fused to obtain the surge threshold. When the surge amplitude of a single event is greater than the surge threshold, the corresponding monitoring event is marked as a trigger event.

[0116] After the event is triggered and within the monitoring event range of the preset fallback confirmation length, retrieve the cumulative offset that is greater than the noise threshold and mark it as a cross offset;

[0117] Mark the monitoring event corresponding to the last valid offset that has been generated across offsets as the triggering event;

[0118] A step mutation region is generated based on the triggering event, the inducing event, and the monitoring event that is between the triggering event and the inducing event in terms of timestamp;

[0119] By integrating drift regions and step change regions, a candidate set of impedance anomalies is generated;

[0120] To address the problem that existing technologies for shore power transfer device operation and maintenance rely heavily on periodic inspections, thermal imaging checks, or difficulties in confirming insulation or contact status during shutdown windows, this invention uses monitoring events and steady-state windows to treat each opening / closing and load step as a natural excitation source. First, within the steady-state window, the root mean square of voltage and current for each phase is calculated to establish a voltage difference-current relationship. Then, a robust linear fitting is used to obtain an equivalent contact impedance sequence. Subsequently, exponentially weighted moving averages and cumulative offsets are used on the impedance sequence to identify drift and step change regions and generate a candidate set of impedance anomalies. This approach transforms invisible, minute impedance changes into a continuously trackable sequence feature without altering hardware or introducing additional testing. Furthermore, by associating anomalies with specific operational events through precipitating events and abrupt change regions, it solves the problem of providing early warnings and trend judgments that are difficult to achieve with traditional on-site measurements.

[0121] Within the transient window corresponding to the monitoring event, the high-frequency energy ratio and spike count of the current waveform are calculated. By analyzing the fluctuation changes of the high-frequency energy ratio and the clustering changes of the spike count, a candidate set of arcing anomalies is constructed.

[0122] Methods for constructing a candidate set of arcing anomalies include:

[0123] Within the transient window corresponding to the monitoring event, the original waveforms of the current in each phase are bandpass filtered to obtain a high-frequency component sequence;

[0124] In this embodiment, the specific process of bandpass filtering the current waveforms of each phase includes:

[0125] Read the start and end sampling sequence number, sampling rate and phase channel mapping of the transient window from the monitoring event, extract the original current discrete waveform of each phase according to the start and end sampling sequence number and generate phase transient segments;

[0126] The monitoring event includes at least the start sampling sequence number, end sampling sequence number and sampling rate fields. The phase channel mapping is used to bind the three-phase current channels to phase A, phase B and phase C. The timestamp of each sampling point is retained synchronously during the interception.

[0127] The DC component of the phase transient segment is eliminated and the amplitude is normalized to obtain the DC-free phase current waveform, which is used as the input sequence for bandpass filtering.

[0128] Among them, DC component elimination is achieved by subtracting the average value of transient segments point by point, and amplitude normalization is achieved by scaling the waveform using the root mean square of the steady-state window current as the scale, so that the high-frequency energy under different load amplitudes is comparable.

[0129] The lower and upper cutoff frequencies of the bandpass filter are determined based on the sampling rate and the power frequency, and the digital bandpass filter coefficients that satisfy the stability constraints are generated.

[0130] The lower cutoff frequency is used to avoid the fundamental frequency and low-order harmonics of the power frequency. It can be set to a larger value between a multiple of the power frequency and the minimum resolvable transient frequency. The upper cutoff frequency does not exceed a preset ratio of the Nyquist frequency to avoid aliasing. The digital bandpass filter can be a double second-order cascaded infinite impulse response structure or a finite impulse response structure. The coefficients are stored with double precision and quantization range limitation to ensure numerical stability during online implementation. For example, if the power frequency of the shore power system is 50 Hz, the goal is to extract high-frequency components such as "contact arcing or glitch" from the current waveform. Therefore, it is necessary to avoid the fundamental frequency and low-order harmonics (such as 100 Hz, 150 Hz, 250 Hz, etc.) while retaining the energy of typical switching transients and arcing pulses in the mid-to-high frequency range.

[0131] Example of determining the lower cutoff frequency: The lower cutoff frequency is calculated using the following formula. :

[0132] In the formula, is the power frequency of the shore power system; m is the power frequency multiple coefficient, which is 10 in this embodiment; The minimum resolvable transient frequency can be calculated by taking the reciprocal of the transient window length;

[0133] Example of determining the upper cutoff frequency: Obtain the current sampling rate and denot it as... Nyquist frequency Current sampling rate Half of the upsampling frequency satisfy ; The preset scaling ratio; in this embodiment, the preset scaling ratio is... It is 0.8;

[0134] The DC phase current waveform is input into a digital bandpass filter, and the corresponding high-frequency component sequence is output.

[0135] Zero-crossing detection is performed on the original waveforms of the current in each phase to obtain zero-crossing sampling point pairs, and the zero-crossing time is calculated by interpolation;

[0136] The specific methods for zero-crossing detection of the original waveforms of the currents in each phase include:

[0137] Within the phase-series sampling frames, the original current sample values ​​are zero-point reference corrected to generate a reference-free current sequence; the zero-point reference is the median of the sample values ​​within the transient window.

[0138] A symbol state machine is used to determine the symbol state on the reference current sequence. The symbol state of "positive state, negative state, and transition state" is generated according to the positive and negative thresholds. When the symbol state of adjacent sampling points flips from positive to negative or from negative to positive, a zero-crossing sampling point pair is output. The positive and negative thresholds are obtained by amplifying the noise threshold by the same factor. In this embodiment, the amplification factor is 3 times.

[0139] The "normal state, negative state, and transition state" are illustrated with examples, and the positive threshold is... The negative threshold is For each sampling point i, a state is assigned. If i is not less than the positive threshold, it is a normal state; if i is not greater than the negative threshold, it is a negative state; if i is between the positive and negative thresholds, it is a transitional state.

[0140] For each zero-crossing sampling point, interpolation is performed based on the sampling timestamp and the reference current value to obtain the zero-crossing time. The interpolation uses two-point linear interpolation, allocating the time interval between the previous and subsequent points according to the relative proportion of the current values ​​at the two points to zero to obtain the zero-crossing time. For example, assuming zero-crossing detection is performed on a phase current waveform, a pair of adjacent sampling points are found (the previous point is positive, and the subsequent point is negative). For the previous sampling point: sampling time t1 = 10.000 ms, reference current value i1 = 5.0 A; for the subsequent sampling point: sampling time t1 = 10.100 ms, reference current value i2 = -3.0 A. Since the zero-crossing occurs between these two points, and assuming that the current changes approximately linearly within this period, the zero-crossing time t0 can be allocated according to the "proportion closer to zero": Δt = t2 - t1 = 0.100 ms.

[0141] The proportion of time required to move from the previous sampling point to zero, i.e., the proportion of linear interpolation, is calculated using the following formula: ;

[0142] Distribute the time interval proportionally to the distance from the previous sampling point to zero, i.e., using the formula:

[0143] t0=t1+BL×Δt=10.000+0.625×0.100=10.0625ms;

[0144] The ratio of high-frequency energy to full-frequency energy is calculated based on the high-frequency component sequence to generate the high-frequency energy percentage.

[0145] Among them, the energy is calculated using the sum of squares of samples within the transient window; the full-frequency energy is calculated by the sum of squares of the original waveform after removing the DC component; and the high-frequency energy ratio is used to characterize the relative intensity of the high-frequency pulse component caused by poor contact.

[0146] By analyzing the number of spike amplitudes and the cluster length of consecutive spikes in a short time segment near the zero crossover moment, spike count and clustering characteristics are generated.

[0147] Methods for generating spike counts and clustering features include:

[0148] For each phase current waveform, extract the zero-crossing time sequence within the transient window, and extract symmetrical short time segments based on the zero-crossing times;

[0149] The short time segment is defined by several milliseconds before and after the zero crossing time, and is converted into a sample point interval according to the sampling rate. In this embodiment, the short time segment covers 4 milliseconds before and after the zero crossing time, which is convenient for focusing on the phase-sensitive interval where the contact reversal and the arc are more likely to occur.

[0150] The local maxima of the high-frequency component sequence within the zero-crossing segment are statistically analyzed to obtain the peak amplitude. The peak amplitudes are then spliced ​​together in chronological order to obtain the peak amplitude sequence.

[0151] The maximum value of the peak amplitude under recent no-alarm monitoring events is obtained as the amplitude threshold. Peak amplitudes exceeding the amplitude threshold are marked as over-threshold peaks. The number of over-threshold peaks is counted and recorded as the peak count.

[0152] All peaks exceeding the threshold are sorted by time, peak clusters formed by several adjacent peaks are identified, the number of peaks in each peak cluster is calculated as the cluster length, and the maximum cluster length is identified.

[0153] Wherein, when the interval between two adjacent overthreshold peaks does not exceed the clustering interval threshold, two adjacent overthreshold peaks are included in the same peak cluster. A peak cluster is generated based on a number of consecutive overthreshold peaks. The interval between any two adjacent overthreshold peaks in the peak cluster does not exceed the clustering interval threshold. In this embodiment, the clustering interval threshold is 1 millisecond.

[0154] Determine whether there is a threshold peak within a short time segment; if so, mark the threshold peak as the target peak.

[0155] Determine whether different target peaks belong to the same peak cluster. If different target peaks belong to the same peak cluster, then determine that the short segment is contaminated by the peak cluster and mark it as a contaminated segment.

[0156] The proportion of the number of contaminated fragments to the total number of short fragments in the transient window is denoted as the cluster duty cycle. This is then combined with the maximum cluster length to obtain the clustering characteristic.

[0157] The high-frequency energy ratio, peak count, and aggregation characteristics are weighted and fused to obtain the arcing index of the monitored event. Based on the abnormal rise of the arcing index, an arcing anomaly candidate set is constructed.

[0158] In this embodiment, the fusion weights for high-frequency energy ratio, spike count, maximum cluster length, and cluster duty cycle are set to 0.4, 0.25, 0.2, and 0.15, respectively.

[0159] Methods for constructing a candidate set of arcing anomalies include:

[0160] Obtain the arcing index for each monitoring event. When the increase in the arcing index of adjacent monitoring events exceeds the surge threshold, both adjacent monitoring events are marked as arcing abnormal events.

[0161] Obtain the arc index of historical monitoring events, calculate the arc index histogram, sort the arc index intervals according to the median of the arc index intervals, and obtain a list of arc index intervals;

[0162] Mark the first three arc index intervals in the arc index interval list as the fusion interval, normalize the distribution probability of the fusion interval, and obtain the interval fusion weight.

[0163] The anomaly detection threshold is obtained by weighting the median of the fusion interval based on the interval fusion weight.

[0164] When the arcing index of a monitored event exceeds the anomaly determination threshold, it is marked as an arcing anomaly event, and an arcing anomaly candidate set is constructed based on the anomaly events.

[0165] Among them, robust normalization uses the median and median absolute deviation to scale the features, and abnormal rise is determined by the arcing index exceeding the healthy baseline quantile threshold and remaining consistent across adjacent monitoring events. The candidate set entries must include at least the monitoring event identifier, phase identifier, and the peak value of the arcing index.

[0166] By performing an intersection operation on the candidate sets of impedance anomalies and arcing anomalies, the fault phase and fault location can be obtained;

[0167] Methods for obtaining the fault phase and fault location include:

[0168] Perform an intersection operation on the candidate set of impedance anomalies and the candidate set of arcing anomalies, and record the result as a fault mapping event;

[0169] Obtain the wiring topology description data of the fault mapping event, map it to maintainable component nodes, and output the fault location and fault phase;

[0170] The wiring topology description data uses nodes and connection relationships to express the electrical path of plug contacts, contactor main contacts, terminal blocks and cable joints. The channel segment corresponds to the topology edge or edge set. During mapping, the nearest maintainable component node is traced from the corresponding edge of the channel segment as the fault location. If there are multiple mapping candidates for the same phase, the component node with the largest product of intersection confidence and consistency score is selected as the final location result.

[0171] To address the issue that existing online monitoring technologies often fail to pinpoint the phase and location of anomalies when they originate from various sources, such as poor contact, load disturbances, transient switching, or sampling noise, conventional indicators may suggest abnormalities. This is addressed by introducing a transient window for arcing precursor analysis. The original current waveform is bandpass filtered to obtain a high-frequency component sequence. This sequence is then combined with statistical peak counts and clustering characteristics of short-time segments near the zero-crossing point. This weighted fusion yields an arcing index and constructs a candidate set for arcing anomalies. Finally, the intersection of the impedance anomaly candidate set and the arcing anomaly candidate set is used to determine the faulty phase and location, which is then mapped to the maintainable component node output on the wiring topology description data. This "intersection" mechanism uses two complementary types of evidence (steady-state impedance changes reflecting changes in contact surface energy loss, and transient high-frequency spikes reflecting micro-arcs and contact discontinuities) to jointly constrain fault judgment. This allows for the convergence of anomalies from "potential problems" to interpretable results specifying "which phase, which channel, and which maintainable component" without requiring manual inspection. This solves the shortcomings of conventional monitoring perspectives in terms of positioning accuracy, false alarm suppression, and maintenance direction.

[0172] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

[0173] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0174] All formulas in this manual are dimensionless and calculated numerically. The formulas are derived from software simulations based on a large amount of collected data to obtain the most recent real-world results. The preset parameters and thresholds in the formulas are set by those skilled in the art according to the actual situation.

[0175] Although embodiments of the invention have been shown and described, those skilled in the art will understand that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the claims and their equivalents.

Claims

1. A fault diagnosis method for a shore power transfer device, characterized in that, include: Read the opening and closing status and three-phase voltage and current sequence from the shore power transfer device, generate monitoring events based on the changes in opening and closing and the step changes in load, and capture the steady-state window before and after the event. The method for generating monitoring events and capturing steady-state windows before and after the events includes: Monitor the connection status of the circuit breaker auxiliary contacts and plug interlock switch. When a change in the connection status is detected, record the connection status and timestamp, and generate a monitoring event. Generate event sequence numbers for monitored events, and align the three-phase voltage and current sequences according to the event sequence numbers to obtain the event-aligned sequence; Within the event alignment sequence, calculate the increment and duration of the current amplitude within the preset load monitoring period. If the increment exceeds the preset transition threshold and the timing duration reaches the preset minimum duration, determine that a load change event has occurred, record the timestamp, and update the monitoring event. Based on the start and end boundaries of the monitored events, the intervals where the rate of change of current is lower than the stable threshold are identified and recorded as the steady-state intervals. Obtain the median timestamp of the steady-state interval and record it as the representative time of the steady-state interval. If the representative time is before the generation time of the monitored event, it is recorded as the pre-steady-state window of the corresponding monitored event; if the representative time is after the generation time of the monitored event, it is recorded as the post-steady-state window of the corresponding monitored event. Both the pre-steady-state window and the post-steady-state window are steady-state windows. The statistics of the steady-state window are obtained and written into the monitoring event. Based on the steady-state window, the root mean square voltage and root mean square current of each phase are calculated. By analyzing the linear consistency of the voltage difference with the root mean square current, the equivalent contact impedance sequence is obtained. The method for obtaining the equivalent contact impedance sequence includes: Within each steady-state window, calculate the root mean square of the incoming line voltage, the root mean square of the outgoing line voltage, and the root mean square of the current for different phases. Calculate the difference between the root mean square of the incoming line voltage and the root mean square of the outgoing line voltage to obtain the voltage difference. Calculate the ratio of voltage difference to root mean square current to obtain the candidate impedance points for the corresponding phase. Robust linear fitting is performed on candidate impedance points of the same phase under multiple consecutive monitoring events to obtain the equivalent contact impedance sequence and the fitting residual sequence. Calculate the exponentially weighted moving average and cumulative offset on the equivalent contact impedance sequence to identify drift and step change regions and generate a candidate set of impedance anomalies. The method for identifying drift and step change regions and generating a candidate set of impedance anomalies includes: Read the equivalent contact impedance values ​​sorted by time on the monitoring event sequence of the same phase to generate an impedance input sequence; The exponentially weighted moving average is updated on the impedance input sequence according to the time interval of the monitoring events to obtain the impedance smoothing sequence and retain the smoothed residual for each monitoring event; Smoothed residuals of the same phase are spliced ​​together according to the time sequence of monitored events to generate residual windows; Identify the median in the residual window, calculate the absolute deviation of the median in the residual window, denoted as the robust dispersion, and calculate the noise threshold based on the robust dispersion. By analyzing the consistency between the continuous positive slope of the impedance smoothing sequence and the sustained growth of the cumulative offset, drift regions can be identified. By analyzing the synchronicity between the fitted changes in impedance values ​​and the crossing of cumulative offset between monitored events, step abrupt change regions of cumulative offset can be identified. By integrating drift regions and step change regions, a candidate set of impedance anomalies is generated; Within the transient window corresponding to the monitoring event, the high-frequency energy ratio and spike count of the current waveform are calculated. By analyzing the fluctuation changes of the high-frequency energy ratio and the clustering changes of the spike count, a candidate set of arcing anomalies is constructed. The method for constructing the candidate set of arcing anomalies includes: Within the transient window corresponding to the monitoring event, the original waveforms of the current in each phase are bandpass filtered to obtain a high-frequency component sequence; Zero-crossing detection is performed on the original waveforms of the current in each phase to obtain zero-crossing sampling point pairs, and the zero-crossing time is calculated by interpolation; The ratio of high-frequency energy to full-frequency energy is calculated based on the high-frequency component sequence to generate the high-frequency energy percentage. By analyzing the number of spike amplitudes and the cluster length of consecutive spikes in a short time segment near the zero crossover moment, spike count and clustering characteristics are generated. The high-frequency energy ratio, peak count, and aggregation characteristics are weighted and fused to obtain the arcing index of the monitored event. Based on the abnormal rise of the arcing index, an arcing anomaly candidate set is constructed. By taking the intersection of the candidate set of impedance anomalies and the candidate set of arcing anomalies, the fault phase and the fault location can be obtained; The method for obtaining the fault phase and fault location includes: Perform an intersection operation on the candidate set of impedance anomalies and the candidate set of arcing anomalies, and record the result as a fault mapping event; Obtain the wiring topology description data of the fault mapping event, map it to maintainable component nodes, and output the fault location and fault phase; The wiring topology description data uses nodes and connection relationships to express the electrical paths of plug contacts, contactor main contacts, terminal blocks and cable joints. The channel segment corresponds to the topology edge or edge set. During mapping, the nearest maintainable component node is traced from the corresponding edge of the channel segment as the fault location. If there are multiple mapping candidates for the same phase, the component node with the largest product of intersection confidence and consistency score is selected as the final location result.

2. The fault diagnosis method for a shore power transfer device according to claim 1, characterized in that, The method for identifying drift regions includes: Calculate the difference between the smoothed residual and the noise threshold to obtain the effective offset. If the effective offset is not greater than 0, then the effective offset is set to 0. The effective offset of the residual window is accumulated according to the event sequence of the monitored events. The accumulated offset is obtained and updated. When the accumulated offset does not increase under the monitoring event corresponding to the preset fallback confirmation length, it is determined that the offset has disappeared and the accumulated offset is set to 0. Calculate the smoothing slope between the equivalent contact impedances of adjacent monitoring events on the impedance smoothing sequence, and identify the smoothing slope region where the smoothing slope is continuously positive. Mark several consecutive monitoring events corresponding to the smooth slope region as target events, and concatenate the cumulative offsets in the residual window according to the time order of the target events to obtain the cumulative offset sequence. Determine if the cumulative offset sequence is increasing. If the cumulative offset is increasing, then generate a drift region based on consecutive target events.

3. The fault diagnosis method for a shore power transfer device according to claim 2, characterized in that, The method for identifying step abrupt change regions of cumulative offset includes: For the same phase impedance input sequence, calculate the difference between the equivalent contact impedance value of each monitoring event and the exponentially weighted moving average value under the adjacent previous monitoring event, and record the result as the single-event surge amplitude of the previous monitoring event; The robustness dispersion and noise threshold are fused to obtain the surge threshold. When the surge amplitude of a single event is greater than the surge threshold, the corresponding monitoring event is marked as a trigger event. After the event is triggered and within the monitoring event range of the preset fallback confirmation length, retrieve the cumulative offset that is greater than the noise threshold and mark it as a cross offset; Mark the monitoring event corresponding to the last valid offset that has been generated across offsets as the triggering event; A step mutation region is generated based on the triggering event, the inducing event, and the monitoring event that is between the triggering event and the inducing event in terms of timestamp.

4. The fault diagnosis method for a shore power transfer device according to claim 3, characterized in that, The method for generating peak counts and clustering characteristics includes: For each phase current waveform, extract the zero-crossing time sequence within the transient window, and extract symmetrical short time segments based on the zero-crossing times; The local maxima of the high-frequency component sequence within the zero-crossing segment are statistically analyzed to obtain the peak amplitude. The peak amplitudes are then spliced ​​together in chronological order to obtain the peak amplitude sequence. The maximum value of the peak amplitude under recent no-alarm monitoring events is obtained as the amplitude threshold. Peak amplitudes exceeding the amplitude threshold are marked as over-threshold peaks. The number of over-threshold peaks is counted and recorded as the peak count. All peaks exceeding the threshold are sorted by time, peak clusters formed by several adjacent peaks are identified, the number of peaks in each peak cluster is calculated as the cluster length, and the maximum cluster length is identified. Determine whether there is a threshold peak within a short time segment; if so, mark the threshold peak as the target peak. Determine whether different target peaks belong to the same peak cluster. If different target peaks belong to the same peak cluster, then determine that the short segment is contaminated by the peak cluster and mark it as a contaminated segment. The proportion of the number of contaminated fragments to the total number of short-time fragments in the transient window is denoted as the cluster duty cycle. This is then combined with the maximum cluster length to obtain the clustering characteristic.

5. The fault diagnosis method for a shore power transfer device according to claim 4, characterized in that, The method for constructing the candidate set of arcing anomalies includes: Obtain the arcing index for each monitoring event. When the increase in the arcing index of adjacent monitoring events exceeds the surge threshold, both adjacent monitoring events are marked as arcing abnormal events. Obtain the arc index of historical monitoring events, calculate the arc index histogram, sort the arc index intervals according to the median of the arc index intervals, and obtain a list of arc index intervals; Mark the first three arc index intervals in the arc index interval list as the fusion interval, normalize the distribution probability of the fusion interval, and obtain the interval fusion weight. The anomaly detection threshold is obtained by weighting the median of the fusion interval based on the interval fusion weight. When the arcing index of a monitored event exceeds the anomaly determination threshold, it is marked as an arcing abnormal event, and an arcing abnormal candidate set is constructed based on the abnormal events.

Images