Rotary electrical connection fault prediction method and system based on resistance timing characteristics

Abstract

The application provides a rotating electrical connector fault prediction method and system based on resistance timing characteristics, and relates to the technical field of computers, comprising: obtaining an original dynamic resistance sampling sequence of a rotating electrical connector and an original timing signal of a rotation angle; performing synchronous sequence segmentation on the original dynamic resistance sampling sequence and the original timing signal of the rotation angle to obtain a resistance-angle synchronous segment set corresponding to a physical rotation period; constructing an angle dimension feature based on the resistance-angle synchronous segment set to generate a timing feature matrix; modeling a health baseline based on the timing feature matrix to obtain a dynamic probability baseline; modeling abnormal propagation dynamics based on the timing feature matrix and the dynamic probability baseline to identify a deterioration evolution stage; predicting a service life and a fault range based on the deterioration evolution stage to obtain a remaining service life prediction value and an angular orientation interval of a potential fault point. The application realizes accurate prediction of rotating electrical connector faults.

Description

Technical Field

[0001] This invention relates to the field of computer technology, and more specifically, to a method and system for predicting faults in rotating electrical connectors based on resistance timing characteristics. Background Technology

[0002] Rotating electrical connectors (such as conductive rings and busbars) are core electromechanical coupling components in high-end rotating equipment such as aerospace solar panel drive mechanisms, aviation servo systems, industrial robot joints, and wind power pitch control systems. Their core function is to achieve stable, low-loss transmission of power and signals between continuously rotating components. The dynamic contact resistance formed between the brush wires and the ring wires during movement directly determines the transmission efficiency and communication reliability, thus becoming the most critical and direct technical indicator for evaluating the service health status of such devices and predicting their remaining service life.

[0003] Current mainstream methods for detecting dynamic contact resistance in rotating electrical connectors suffer from fundamental flaws in data acquisition principles and implementation. Existing methods are essentially offline, static, and manual measurements: the component under test is connected via a test cable, and an operator uses a multimeter or similar instrument to read the resistance value point-by-point while the device is stationary at several discrete angular positions. Ultimately, only the maximum and minimum values ​​at a few locations along the entire circumference are recorded. This approach introduces several methodological limitations: First, the commonly used two-wire measurement principle cannot overcome the influence of the impedance of the test leads themselves. This causes the lead resistance to directly contaminate the measurement results when measuring dynamic contact resistance in the milliohm range, resulting in a non-negligible systematic error. Second, limited by manual operation and the performance of general-purpose instruments, this method has an extremely low effective sampling rate, completely failing to capture the complete dynamic details such as instantaneous fluctuations and transient peaks in resistance during continuous rotation, thus losing crucial timing information containing early signs of degradation. Furthermore, this method acquires isolated data points that are disconnected from the continuous motion process, failing to construct a complete spatiotemporal mapping of the resistance value's continuous evolution with the rotation angle, thus fundamentally losing the ability to spatially locate faults on the mechanical circumference. Finally, each detection is merely an independent snapshot, lacking effective correlation between data from different tests, making it impossible to analyze the performance degradation trajectory and rate through accumulated historical data, and therefore completely unsupporting data-driven trend prediction and lifespan assessment.

[0004] Based on the shortcomings of the existing technology, there is an urgent need for a method and system for predicting the faults of rotating electrical connectors based on resistance timing characteristics. Summary of the Invention

[0005] The purpose of this invention is to provide a method and system for predicting faults in rotating electrical connectors based on resistance timing characteristics, in order to improve the aforementioned problems. To achieve the above objective, the technical solution adopted by this invention is as follows:

[0006] In a first aspect, this application provides a method for predicting faults in rotating electrical connectors based on resistance timing characteristics, including:

[0007] Acquire the original dynamic resistance sampling sequence generated by the multi-channel contact pair and the original rotation angle timing signal synchronized with the resistance sampling time of the rotating electrical connector in operation.

[0008] The rotation cycle synchronization sequence is segmented based on the original dynamic resistance sampling sequence and the original rotation angle timing signal. By aligning and truncating the original dynamic resistance sampling sequence and the original rotation angle timing signal based on the boundary of the complete mechanical rotation cycle, a set of resistance-angle synchronization segments corresponding to the physical rotation cycle is obtained.

[0009] Angle dimension features are constructed based on the resistance-angle synchronization segment set. By calculating the statistical deviation of the resistance value in each angle interval of the resistance-angle synchronization segment set relative to the resistance distribution of historical segments, a time-series feature matrix characterizing the abnormal level of contact state at each angle position is generated.

[0010] A health baseline is modeled based on the time-series feature matrix. A dynamic probability baseline is obtained by constructing probability density functions for the features of different angle regions in the time-series feature matrix.

[0011] Anomaly propagation dynamics modeling is performed based on the time-series feature matrix and the dynamic probability baseline. By detecting abnormal angle regions where the time-series feature matrix deviates from the dynamic probability baseline and calculating the correlation strength between the abnormal angle regions and the overall feature entropy increase of the system, the deterioration evolution stage is identified.

[0012] Based on the aforementioned degradation evolution stages, the remaining service life and failure range are predicted to obtain the predicted value of remaining service life and the angular and azimuth range of potential failure points.

[0013] Secondly, this application also provides a rotating electrical connector fault prediction system based on resistance timing characteristics, comprising:

[0014] The acquisition module is used to acquire the original dynamic resistance sampling sequence generated by the multi-channel contact pair and the original rotation angle timing signal synchronized with the resistance sampling time of the rotating electrical connector in the operating state.

[0015] The segmentation module is used to segment the rotation cycle synchronization sequence according to the original dynamic resistance sampling sequence and the original rotation angle timing signal. By aligning and truncating the original dynamic resistance sampling sequence and the original rotation angle timing signal based on the boundary of the complete mechanical rotation cycle, a set of resistance-angle synchronization segments corresponding to the physical rotation cycle is obtained.

[0016] The construction module is used to construct angular dimension features based on the resistance-angle synchronization segment set. By calculating the statistical deviation of the resistance value in each angle interval of the resistance-angle synchronization segment set relative to the resistance distribution of historical segments, a time-series feature matrix characterizing the abnormal level of contact state at each angle position is generated.

[0017] The modeling module is used to model a health baseline based on the time-series feature matrix, and to obtain a dynamic probability baseline by constructing probability density functions for the features of different angle regions in the time-series feature matrix.

[0018] The identification module is used to perform anomaly propagation dynamics modeling based on the temporal feature matrix and the dynamic probability baseline. By detecting the abnormal angle region where the temporal feature matrix deviates from the dynamic probability baseline and calculating the correlation strength between the abnormal angle region and the overall feature entropy increase of the system, the degradation evolution stage is identified.

[0019] The prediction module is used to predict the lifespan and fault range based on the degradation evolution stage, and obtain the predicted value of the remaining service life and the angular and azimuth range of potential fault locations.

[0020] The beneficial effects of this invention are as follows:

[0021] This invention constructs a time-angle two-dimensional spatiotemporal analysis framework by strictly synchronizing a multi-channel dynamic resistance sequence with a rotation angle signal and performing data segmentation and reconstruction based on a complete mechanical rotation cycle. This framework establishes an independent adaptive health probability baseline for each angular region on the circumference. By analyzing the patterns of local abnormal deviations from the baseline and their propagation dynamics in space and time, the invention ultimately achieves quantitative prediction of the remaining service life and precise angular location of potential fault points of rotating electrical connectors from early, localized performance degradation symptoms. This solves the technical problem that traditional methods cannot combine fault warning with spatial positioning, and enables accurate prediction of rotating electrical connector faults. Attached Figure Description

[0022] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as a limitation on the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0023] Figure 1 This is a flowchart illustrating a method for predicting faults in rotating electrical connectors based on resistance timing characteristics, as described in an embodiment of the present invention.

[0024] Figure 2This is a schematic diagram of the structure of a rotating electrical connector fault prediction system based on resistance timing characteristics as described in an embodiment of the present invention;

[0025] Figure 3 This is a heatmap of spatiotemporal anomalies.

[0026] The diagram is labeled as follows: 901, Acquisition Module; 902, Segmentation Module; 903, Construction Module; 904, Modeling Module; 905, Recognition Module; 906, Prediction Module. Detailed Implementation

[0027] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, 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, not all, of the embodiments of the present invention. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention.

[0028] It should be noted that similar reference numerals and letters in the following figures indicate similar items; therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures. Furthermore, in the description of this invention, terms such as "first," "second," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0029] Taking the propulsion mechanism of a long-term orbital spacecraft solar panel as an example, its conductive ring operates under harsh conditions involving multiple physical fields coupled together, such as vacuum, radiation, large temperature differences, and continuous fretting friction. The performance degradation of the contact interface is a complex spatiotemporal evolution process. A key characteristic of this process is that, due to microscopic differences such as assembly stress, material inhomogeneity, or lubricant distribution, failure modes such as wear, oxidation, or thin film formation do not occur uniformly. Instead, they always begin at certain local weak points on the circumference and gradually spread and evolve along the contact trajectory with the accumulation of rotation cycles. This gives rise to a profound technical problem that goes beyond simple anomaly alarms: how to not only identify anomalies from massive amounts of dynamic resistance time-series data that are strictly synchronized with rotation, but also proactively diagnose where (angle) the degradation originates, in what mode (e.g., static, diffuse) it evolves in space, how long it is expected to lead to functional failure, and in which angles and locations the failure is most likely to be concentrated. Traditional methods are ill-suited to this challenge because they cannot accurately anchor resistance changes to specific mechanical locations and establish a historical evolution model. Therefore, this invention aims to provide a complete technical solution: by synchronously reconstructing a one-dimensional resistance sequence and angle signal into a spatiotemporal data field, establishing an adaptive health baseline for each angle region to sense local deviations, analyzing the propagation dynamics of abnormal modes and quantifying their correlation with the overall system disorder, and finally integrating the physical degradation model and spatiotemporal extrapolation to output the predicted range of remaining service life and the angular orientation range of potential faults in an integrated manner, thereby achieving predictive maintenance from early warning to precise positioning.

[0030] Example 1:

[0031] This embodiment provides a method for predicting faults in rotating electrical connectors based on resistance timing characteristics.

[0032] See Figure 1 The figure shows that the method includes steps S100 to S600.

[0033] Step S100: Obtain the original dynamic resistance sampling sequence generated by the multi-channel contact pair and the original rotation angle timing signal synchronized with the resistance sampling time of the rotating electrical connector in the running state.

[0034] It is understood that this embodiment uses the conductive ring of the aerospace solar panel drive mechanism (SADM) as the rotating electrical connector under test. A wide-range, multi-channel dynamic resistance intelligent testing and calibration device is used to complete the raw data acquisition and preprocessing, and fault prediction is achieved through resistance timing characteristic analysis. The wide-range, multi-channel dynamic resistance intelligent testing and calibration device is used to acquire the dynamic contact resistance of the aerospace SADM conductive ring. The hardware precision measurement and control unit of this device uses an FPGA (Field Programmable Gate Array) as the core intelligent main control module, configured with 18 constant current precision excitation modules, 18 precision signal conditioning modules, and 18 ultra-high precision AD (Analog-to-Digital Converter) synchronous acquisition modules. These three components form a distributed measurement and control link through a PXI (PCI extension for instrument systems) bus high-speed communication module (equipped with a PCI9540 chip, transmission rate 132MHz), achieving zero-delay synchronous acquisition of 18 signals. The constant current precision excitation module adopts a high-stability mirror constant current source topology circuit, configured with two switchable output currents of 200mA and 20mA, respectively adaptable to measurement ranges of 10mΩ~10Ω and 10Ω~100Ω; the precision signal conditioning module uses an INA128 low offset voltage instrumentation amplifier (2x or 20x amplification selectable) in the first stage and an OP07 low drift precision operational amplifier (1x or 10x amplification selectable) in the second stage, with a built-in high-performance passive low-pass filter circuit with a 12kHz cutoff frequency; the AD synchronous acquisition module uses a 15-bit ultra-high precision conversion chip with a conversion rate of 600kSPS, and achieves an actual sampling frequency of 1.2kHz with the FPGA parallel interface. The device adopts a self-improved four-wire resistance measurement architecture. The constant current source applies a stable reference excitation current to the conductive loop, and an independent high input impedance voltage acquisition link is used to extract the real voltage drop signal across the dynamic contact resistance between the brush line and the loop line, avoiding system deviations introduced by the equivalent impedance of the adapter cable at the hardware architecture level. After the device is powered on, the integrated intelligent path self-test module automatically performs a closed-loop test on the constant current source output accuracy, signal conditioning amplification linearity, AD acquisition validity, and PXI bus communication link connectivity (including link transmission rate, data packet loss rate, and command response delay) of each channel. After confirming that there are no abnormalities, formal acquisition begins. The rotation speed of the conductive ring under test is set to 6 r / min, and the channel synchronous acquisition trigger threshold is set to 5mV, and the standardized test procedure is started. The FPGA automatically matches the 200mA constant current source output according to the preset range. When the conductive ring rotates at a constant speed of 6 r / min, a voltage drop is formed across the dynamic contact resistance generated by the sliding of the brush line relative to each ring channel (typical value range 10mΩ~800mΩ corresponds to a voltage drop of 2mV~160mV). This weak voltage drop is deeply denoised by a 12kHz low-pass filter and then amplified 20 times by an INA128 instrumentation amplifier and 10 times by an OP07 operational amplifier. Finally, the amplitude is increased to 40mV~1.6V before entering the AD conversion stage.The AD synchronous acquisition module converts the analog signal into a 15-bit digital value and transmits it to the FPGA in real time. The FPGA performs three-stage filtering and noise reduction on each digital signal (including continuous moving average filtering to eliminate electromagnetic noise), and then calculates the real-time resistance sequence based on Ohm's law and the actual output current value of the constant current source. Simultaneously, the device's host computer intelligent measurement and control unit calculates the rotation angle data in real time according to the preset rotation speed of 6 r / min, generating a raw rotation angle timing signal that is strictly synchronized with the resistance sampling time. Ultimately, two core timing inputs are obtained: one is the raw dynamic resistance sampling sequence corresponding to each of the 18 channels, and the other is the globally unified raw rotation angle timing signal.

[0035] Step S200: Perform rotation cycle synchronization sequence segmentation based on the original dynamic resistance sampling sequence and the original rotation angle timing signal. By aligning and truncating the original dynamic resistance sampling sequence and the original rotation angle timing signal based on the boundary of the complete mechanical rotation cycle, a set of resistance-angle synchronization segments corresponding to the physical rotation cycle is obtained.

[0036] Further, step S200 includes steps S210 to S230.

[0037] Step S210: Perform signal timing alignment processing based on the original dynamic resistance sampling sequence and the original rotation angle timing signal. By aligning the initial acquisition times of the two signals, a dynamic resistance-angle signal pair with a synchronous timestamp is obtained.

[0038] This step precisely aligns the original dynamic resistance sampling sequence with the initial acquisition time of the original rotation angle timing signal. Since both signals are generated by the same PXI-FPGA acquisition link and share the same hardware clock reference, their inherent timestamp deviation does not exceed one sampling interval (approximately 0.83ms). The alignment operation uses the first valid zero-crossing point of the angle signal as the time reference, uniformly truncating the resistance sequences of all channels into the specified interval. After alignment, a set of dynamic resistance-angle signal pairs with synchronized timestamps is obtained. This set of signal pairs ensures that each subsequent resistance value can be uniquely mapped to its corresponding rotation angle position, forming the data foundation for constructing the angle dimension features.

[0039] Step S220: Perform rotation cycle boundary identification processing based on the dynamic resistance-angle signal pair. By performing periodic extreme point detection on the original rotation angle time sequence signal, extract the start and end timestamps of each complete mechanical rotation cycle to obtain the rotation cycle boundary sequence.

[0040] This step involves periodic extreme point detection of the original rotation angle time-series signal to determine the boundaries of each complete mechanical rotation cycle. First, the angle signal is detrended, subtracting the linearly growing component to obtain a zero-mean angle residual signal. Then, a Fast Fourier Transform (FFT) is applied to the residual signal to calculate the power spectral density, locating the fundamental frequency corresponding to the main peak in the frequency domain. Next, a Score function is constructed to search for local maxima near integer period lengths, precisely locating the start and end times of each cycle. The Score function is expressed as:

[0041] ;

[0042] In the formula, In time shift The autocorrelation function value at a given point is used to measure the similarity between the signal and its delayed replica; the peak value corresponds to the period of the signal. Indicates at time Angular residual signal; Indicates at time Angular residual signal; This is the time shift.

[0043] For real-world testing scenarios with slight rotational speed fluctuations, the autocorrelation peak detection using the Score function is more robust than the fixed threshold method. The final result is the rotational period boundary sequence. ,in The total number of cycles. For the first The start time of a complete mechanical rotation cycle For the first The end time of a complete mechanical rotation cycle The sequence number represents the complete mechanical rotation cycle. Each boundary pair defines a complete mechanical rotation cycle (360°).

[0044] Step S230: Perform synchronization segment extraction processing based on the rotation period boundary sequence and the dynamic resistance-angle signal pair. By extracting the corresponding resistance and angle data segments in each period from the dynamic resistance-angle signal pair according to the time scale range indicated by the rotation period boundary sequence, a set of resistance-angle synchronization segments is obtained.

[0045] This step, based on the time-scale range indicated by the rotation period boundary sequence, extracts the corresponding resistance and angle data segments within each period from the dynamic resistance-angle signal pair. The time interval of each cycle is Extract the first from it The resistance sequence of the channel within this interval and its corresponding angle subsequence ,in, For the first The measurement channel in the first The sequence of resistors within one rotation cycle For the first Each channel at the sampling time The collected dynamic contact resistance values, For the first Angle subsequence within a rotation period, To be at the sampling time The collected rotation angle value, For the first The time corresponding to each sampling point For the first The start time of a complete mechanical rotation cycle For the first The termination time of a complete mechanical rotation cycle. To eliminate the inconsistency in the number of sampling points caused by slight speed differences during different cycles, equal-length normalization is performed on all cycle segments: calculate all... Median sampling points per cycle For each shorter than The segments are padded with linear interpolation at the end, for segments longer than [a certain length]. The segments are uniformly downsampled and truncated. Each standardized segment is... A sequence of equal lengths at each sampling point. The final set of resistance-angle synchronization segments is denoted as... , which includes Each segment is a standardized periodic segment, and the data points within each segment correspond strictly one-to-one with the physical rotation period.

[0046] Step S300: Construct angle dimension features based on the resistance-angle synchronization segment set. By calculating the statistical deviation of the resistance value in each angle interval of the resistance-angle synchronization segment set relative to the resistance distribution of historical segments, generate a time-series feature matrix characterizing the abnormal level of contact state at each angle position.

[0047] Further, step S300 includes steps S310 to S330.

[0048] Step S310: Perform uniform division of angle intervals according to the resistance-angle synchronization segment set. By dividing the circumferential angle space into several continuous sectors, the resistance data in all segment sets are classified into the corresponding angle sectors according to the synchronized angle values, thus obtaining a set of resistance sub-sequences organized by angle sectors.

[0049] This step divides the circumferential angle space [0°, 360°] into equal parts. This embodiment selects consecutive sectors and categorizes the resistance data in all periodic segments into the corresponding angle sectors according to their synchronization angle values. Then the angular width of each sector ,in, The total number of sectors, The angular width for each sector. For any normalized periodic segment... Each data point is assigned to a sector number. ,in, The sector number has a value range of {0, 1, ..., 71}. For the first In the periodic segment, the The synchronous rotation angle value corresponding to each sampling point The corner width for each sector. Traverse all sectors. After collecting all data points from each period and 18 channels, a set of resistor sequences organized by angular sectors is obtained. Each of them It is a one-dimensional array containing all resistance observations of the channel across all cycles at that angular sector position. This angular-dimensional discretization transforms the continuous resistance-angle trajectory into a structured sector representation, allowing for statistical analysis of the contact state at different angular positions. This is a prerequisite step for subsequent statistical bias calculations and spatiotemporal feature matrix generation.

[0050] Step S320: Perform statistical deviation calculation processing for angle sector based on the set of resistance sub-sequences. For each angle sector, calculate the statistical distribution parameters of the resistance value at the current moment, and compare the statistical distribution parameters with the baseline distribution formed by the statistical distribution parameters corresponding to all historical moments in the angle sector to calculate the statistical deviation value that represents the current degree of deviation.

[0051] This step applies to each angle sector. and each channel The statistical distribution parameters of the internal resistance value at the current moment are calculated and compared with the baseline distribution formed by the corresponding parameters at all historical moments within the sector to obtain the statistical deviation value characterizing the current degree of deviation. Specifically, a sliding window is selected. The most recent period is used as the current window, and the sectors within the window are... aisle All resistance values ( Calculate three statistics (sample mean, sample standard deviation, and skewness coefficient) for the periodic index within the window. The vector formed by these three statistics... The resistance distribution at this sector location is comprehensively characterized. This vector is then compared with a baseline distribution: the baseline distribution is generated when the sector is in a healthy operating phase (usually taking the previous value). A multivariate reference center composed of similar statistics (per cycle). and its covariance matrix Characterization. The statistical deviation value is calculated using the Mahalanobis distance form:

[0052] ;

[0053] In the formula, Indicates the first The first channel, the first Statistical deviation values ​​for each sector; This represents the difference vector between the current statistical feature vector and the healthy reference center vector; The matrix representing the inverse of the covariance matrix of the health stage; This indicates the transpose operation. Mahalanobis distance naturally takes into account the correlation between different statistics and their respective scale differences. (This embodiment takes) When the 99.7% confidence level under the normality assumption is reached, the sector is considered to have a significant deviation.

[0054] Step S330: Generate a spatiotemporal feature matrix based on the statistical deviation values ​​of all angle sectors. This is done by arranging the statistical deviation values ​​of all angle sectors at the same time in angular order to form a row of feature vectors, and stacking the feature vectors from different times along the time dimension to generate a time-series feature matrix.

[0055] This step arranges the statistical deviation values ​​of all angle sectors at the same time into a row of feature vectors in angular order, and then stacks the feature vectors from different times along the time dimension to generate a time-series feature matrix. Specifically, at the end time of each sliding window (corresponding to the first) (Evaluation time points), organize the deviation values ​​of 18 channels across 72 sectors into a 72-column row vector (if multiple channels need to be fused, the maximum channel value or weighted average can be used), and stack them along the time axis. After obtaining such row vectors, we get dimensional time series feature matrix , of which OK , It is a row vector with dimensions 1×72. Indicates the first The first evaluation time point, the first Statistical deviation values ​​for each angle sector The value ranges from 0 to 71. Each row of the matrix describes the spatial distribution pattern of anomalous deviations across the entire angular range at a given moment, while each column tracks the evolution of a specific angular position over time. Simultaneously encoding spatial information in the angular dimension and evolutionary information in the temporal dimension, it constitutes a unified input data structure for health baseline modeling and anomaly propagation dynamics analysis.

[0056] Step S400: Model a health baseline based on the time series feature matrix. By constructing probability density functions for the features of different angle regions in the time series feature matrix, a dynamic probability baseline is obtained.

[0057] Further, step S400 includes steps S410 to S430.

[0058] Step S410: Initialize the angle region baseline distribution based on the time series feature matrix. By fitting the initial probability density function to the feature data of each angle region in the time series feature matrix during the healthy operation phase, the initial angle region probability density function is obtained.

[0059] This step involves the time series feature matrix. Each angle region During the healthy operation phase, the feature data is fitted with an initial probability density function (PDF). Before selection... Each assessment time point serves as a health baseline period, from which each angular region is extracted. The sequence of deviation values ​​during this period. Considering the non-normal characteristics of the actual data (such as skewed or multimodal distributions), kernel density estimation (KDE) is used instead of simple parametric normal fitting to construct the initial PDF:

[0060] ;

[0061] In the formula, For the first The initial probability density function fitted to each angular region; The independent variable is the probability density function; Indicates the first The first evaluation time point, the first Statistical deviation values ​​for each angular region; This represents the total number of assessment time points; For bandwidth parameters; The Gaussian kernel function; Index for the evaluation time point. Bandwidth. Determined using Silverman's rule of thumb:

[0062] ;

[0063] In the formula, For bandwidth parameters; For the first Sample standard deviation of the deviation value sequence of each angular region during the healthy period; For the first Interquartile range of deviation values ​​in each angular region during the healthy period; This represents the total number of evaluation time points. Silverman's rule approaches optimal bandwidth when the data is nearly normal, while also exhibiting good robustness in skewed conditions. Finally, the initial probability density function for each angular region is obtained, collectively forming the initial baseline model.

[0064] Step S420: Calculate the regional confidence weights based on the time series feature matrix. By analyzing the volatility and stability of historical feature data for each angle region and combining the theoretical mechanical wear intensity of the rotating electrical connector in the angle region, the confidence weight coefficients are calculated.

[0065] The volatility and stability of historical characteristic data for each angular region are analyzed, and confidence weight coefficients are calculated based on the theoretical mechanical wear intensity of the rotating electrical connector in that region. The confidence weight reflects the reliability of the baseline PDF for that angular region in subsequent fusion—regions with more stable data and less mechanical wear have higher weights. The confidence weight consists of two weighted components:

[0066] The first part is the data stability weight, based on the coefficient of variation. ( For the first The coefficient of variation for each angular region and The first Calculation of the mean and standard deviation of the deviation value sequence for each angular region during the healthy period:

[0067] ;

[0068] In the formula, Weights for data stability; For the attenuation coefficient, this embodiment takes... ; For the first The coefficient of variation for each angular region. When When approaching zero (data is highly stable) The weight approaches 1; when the data fluctuates drastically, the weight decays rapidly.

[0069] The second part is the mechanical wear weight, which is set based on the theoretical wear intensity of this angle region on the conductive ring. For SADM conductive rings, the areas where the brush enters and exits the bridge experience faster wear due to stress concentration, and the mechanical wear intensity in these areas is set to a higher value. The remaining regular gliding areas showed less wear. .but , For the first The theoretical mechanical wear intensity of each angular region, i.e., the more severe the wear, the lower the confidence weight.

[0070] The final confidence weight coefficient is the normalized weighted sum of the two:

[0071] ;

[0072] In the formula, For the first Confidence weight coefficients for each angular region; Weights for data stability; The mechanical wear weight is calculated based on theoretical wear intensity; It is a balancing factor.

[0073] Step S430: Perform dynamic probability baseline fusion processing based on the initial angle region probability density function and confidence weight coefficient. By using the confidence weight coefficient to perform weighted fusion and adaptive update of the initial angle region probability density function, a dynamic probability baseline is obtained.

[0074] This step utilizes confidence weight coefficients to perform weighted fusion and adaptive updates on the probability density function of the initial angle region, resulting in a dynamic probability baseline. The dynamic update mechanism of the baseline employs an incremental learning strategy in the form of an exponential moving average (EMA): whenever a new evaluation time point arrives, the latest deviation value is obtained and used to update the current baseline. The specific update rules are as follows:

[0075]

[0076] First, calculate the contribution weight of the new observation point to the current PDF. ,in The global learning rate, This represents the confidence weight coefficient for that angular region.

[0077] Secondly, the updated PDF after incorporating the new information is as follows:

[0078] ;

[0079] In the formula, For the updated version Dynamic probability density function for each angular region; For the first time before the update Dynamic probability density function for each angular region; To be at the latest assessment point in time, the first The latest observation deviation values ​​for each angular region; The independent variable is the probability density function; For the first The contribution weight of new observation points in each angular region; This is the Dirac delta function. The intuitive meaning of the above update rule is that regions with high confidence weights (stable data, low wear) are more receptive to new information, and the baseline can quickly track real changes; regions with low confidence weights tend to maintain the original baseline unchanged, avoiding oversensitivity due to inherent large fluctuations.

[0080] go through After each iteration, the final dynamic probability baseline for each angular region is obtained. This baseline combines the characteristics of a "static reference" (from a KDE fit during the healthy period) and an "adaptive tracking" (from incremental updates of the EMA), enabling it to remain stable within the normal drift range while responding promptly to the emergence of real degradation trends.

[0081] Step S500: Based on the time series feature matrix and the dynamic probability baseline, perform anomaly propagation dynamics modeling. By detecting the abnormal angle region where the time series feature matrix deviates from the dynamic probability baseline and calculating the correlation strength between the abnormal angle region and the overall feature entropy increase of the system, the deterioration evolution stage can be identified.

[0082] Further, step S500 includes steps S510 to S530.

[0083] Step S510: Perform abnormal angle region detection processing based on the temporal feature matrix and dynamic probability baseline. Calculate the probability density by substituting the feature values ​​of each time and angle region in the temporal feature matrix into the corresponding dynamic probability baseline, and mark regions with probability densities lower than a set threshold as abnormal to obtain a spatiotemporal anomaly map.

[0084] This step involves substituting the feature value of each angular region corresponding to each evaluation time point in the time-series feature matrix into the currently maintained dynamic probability baseline function for that angular region to calculate the probability density value of that feature value under the probability distribution. This probability density value quantitatively characterizes the degree to which the currently observed deviation value conforms to the expected distribution of the baseline for that angular region. The higher the probability density value, the more likely the deviation value originates from normal system fluctuations; the lower the probability density value, the more likely it is an abnormal deviation.

[0085] To implement anomaly detection, a relative anomaly detection threshold is set. For each spatiotemporal location—a specific evaluation time point and a specific angular region—an anomaly is assessed. If the probability density value is below the threshold, the location is marked as an anomaly. After traversing all evaluation time points and all angular regions to complete the anomaly detection, a binary spatiotemporal anomaly map is obtained. This map is a matrix structure, where rows correspond to angular regions and columns correspond to evaluation time points. Each element in the matrix takes a value of 1 or 0; a value of 1 indicates that an anomaly was detected at the corresponding spatiotemporal location, and a value of 0 indicates that the state is normal.

[0086] To further improve the robustness of the detection results and suppress false alarms caused by random noise, mathematical morphological opening filtering was performed on this binarized spatiotemporal anomaly map. This operation sequentially performs erosion and dilation operations on a 3x3 structuring element: first, erosion is performed to eliminate small anomaly clusters with areas smaller than three adjacent units (i.e., considered isolated noise points); then, dilation is performed to restore the boundary morphology of meaningful anomaly regions that may have been reduced in size due to erosion. After this morphological filtering, a purified spatiotemporal anomaly map is obtained. This map filters out isolated noise points, more clearly revealing the true distribution and structure of anomaly patterns in time and angular space, and will serve as input data for quantifying anomaly correlation strength and the degradation evolution stage of the identification system in subsequent analysis steps. The heatmap of the spatiotemporal anomaly map is shown below. Figure 3 As shown.

[0087] Step S520: Quantify the correlation strength of anomalies based on the spatiotemporal anomaly map. By analyzing the connectivity of the anomalous regions in the spatiotemporal anomaly map, the spatial displacement and area changes in the continuous rotation cycle, and calculating the overall complexity and disorder change index of the spatiotemporal anomaly map, the correlation strength between the anomalous angle region and the overall characteristic entropy increase of the system is quantified.

[0088] Further, step S520 includes steps S521 to S523.

[0089] Step S521: Extract the spatiotemporal characteristics of the anomalous region based on the spatiotemporal anomaly map. By identifying and tracking the marked anomalous regions in the spatiotemporal anomaly map, calculate the spatial centroid displacement vector, the rate of change of the region area, and the spatial overlap between adjacent anomalous regions in consecutive rotation cycles to obtain the first set of quantitative indicators.

[0090] This step identifies and tracks marked anomalous regions in the spatiotemporal anomaly map, calculates their three basic kinematic indices between consecutive rotation cycles, and obtains the first set of quantitative indices. Specifically, these include:

[0091] Spatial centroid displacement vector: For anomaly regions at two adjacent evaluation times, calculate their centroid angular positions. The centroid angular position at the previous time step is equal to the sum of the products of all anomaly sector numbers and the anomaly flag values ​​at their corresponding positions, divided by the total area of ​​the anomaly region at that time. The centroid angular position at the current time step is calculated in the same way, but using the anomaly flag values ​​at the current time step. Therefore, the centroid displacement is equal to the absolute value of the difference between the current and previous centroid angular positions, expressed in sector number differences. Multiplying this difference by the angular width of each sector converts it to the actual angular displacement. This displacement measures the migration range of the anomaly region's "center of gravity" between adjacent time steps.

[0092] Regional area change rate: Let the total area of ​​the anomalous region at the previous time step be equal to the sum of all anomalous indicator values ​​at that time step, and the total area of ​​the anomalous region at the current time step be equal to the sum of all anomalous indicator values ​​at that time step. Then the area change rate is equal to the absolute value of the difference between the total areas of the anomalous region at the two time steps, divided by the total area of ​​the anomalous region at the previous time step, plus a constant to prevent division by zero. A rate of change greater than zero indicates that the anomalous region is expanding, and a rate of change less than zero indicates that it is contracting.

[0093] Spatial overlap between adjacent periods: This is calculated as the ratio of the intersection to the union of the anomalous regions at two adjacent time points. The intersection size equals the number of positions where an element-wise logical AND operation of the anomalous region vectors at both time points results in a true value, and the union size equals the number of positions where an element-wise logical OR operation of the anomalous region vectors at both time points results in a true value. The spatial overlap ranges from zero to one, measuring the temporal persistence of the anomalous pattern—a spatial overlap approaching one indicates that the anomalous region hardly changes over time, exhibiting a highly stable, fixed anomalous pattern; a spatial overlap approaching zero indicates that the anomalous region is jumping or flickering violently.

[0094] The above three indicators constitute the first set of quantitative indicators, which includes the triplet sequence calculated for each adjacent time from the second evaluation time to the last evaluation time. Each triplet includes centroid displacement, area change rate, and spatial overlap.

[0095] Step S522: Perform correlation strength index calculation processing based on the first set of quantitative indicators. Calculate the consistency of the main direction of anomaly propagation based on the spatial centroid displacement vector, calculate the acceleration of anomaly expansion based on the regional area change rate, and calculate the temporal persistence of the anomaly pattern based on spatial overlap to obtain the second set of quantitative indicators.

[0096] Based on the three fundamental indicators of the first layer, three derived composite indicators are calculated to obtain the second set of quantitative indicators. Specifically, these include:

[0097] Main Directional Consistency Index: This index is used to assess the directional consistency of centroid displacement in anomaly regions over a continuous period. It is calculated as follows: First, a centroid displacement sequence containing several consecutive evaluation time windows (e.g., ten windows) is selected. Next, the sign (positive or negative) of each displacement in the sequence relative to the median displacement of the entire sequence is calculated, and the average of these signs is taken. Finally, the absolute value is calculated. The index ranges from zero to one. When the index value approaches one, it indicates that the centroid movement direction in the anomaly region is highly consistent within the observation window (e.g., continuous clockwise or counterclockwise diffusion), which usually indicates the presence of a clear, directional driving force of physical degradation (such as continuous wear of a single-sided brush). When the index value approaches zero, it indicates that the displacement direction is chaotic and may be more caused by random noise or multiple confounding factors.

[0098] Expansion Acceleration Index: This index captures the changing trend of the rate of expansion of an abnormal region, i.e., the "acceleration" of area change. It is calculated based on a time series of the area change rate, achieved by calculating the second difference of that series. When the index value is greater than zero and remains positive, it indicates that the total area of ​​the abnormal region is not only expanding, but the rate of expansion is also accelerating, a typical characteristic of the degradation process entering a rapid development stage. Conversely, when the index value is less than zero, it indicates that the expansion rate is slowing down, or even potentially turning into a contraction state, which may stem from temporary self-repair effects or transient disturbances in the measurement.

[0099] Temporal Persistence Index: This index aims to comprehensively assess the historical coherence and stability of anomalous patterns over time. It is calculated using an exponentially decaying memory integral method: within a defined cumulative time window (e.g., twenty evaluation points), the spatial overlap of each historical moment is weighted and summed. The weighting coefficient decays exponentially with the interval between that historical moment and the current moment (e.g., a decay constant is set so that the half-life of the memory is approximately ten time units). This weighting method gives greater weight to recent high persistence while gradually "forgetting" the effects of longer-term events. A higher index value indicates stronger historical coherence and persistence of the anomalous pattern, suggesting a long-term structural degradation rather than a transient transient disturbance.

[0100] The three derived indices mentioned above together constitute the second set of quantitative indices. This set is a sequence whose starting time is determined by the window length required to calculate the principal direction consistency and the cumulative window length required to calculate the time series persistence. The larger of the two values ​​is incremented by one, and the sequence continues until the last evaluation time. The sample at each time point contains the values ​​of the three indices corresponding to that time point: principal direction consistency, expansion acceleration, and time series persistence.

[0101] Step S523: Based on the spatiotemporal anomaly map and the second set of quantitative indicators, perform system entropy increase correlation analysis. By performing multiple regression analysis on the main direction consistency, expansion acceleration and time series persistence indicators in the second set of quantitative indicators and the values ​​of the overall system characteristic entropy increase, the correlation strength characterizing the correlation between local anomaly dynamics and the disordered growth of the overall system is calculated.

[0102] This step involves performing a multiple regression analysis on the three composite indicators from the second set of quantitative indicators and the overall system entropy increase to calculate the correlation strength, which characterizes the degree of correlation between local abnormal dynamics and the disordered growth of the overall system. The specific calculation process is as follows:

[0103] First, the overall entropy increase of the system is calculated. For the full-angle deviation vector corresponding to each evaluation moment, the absolute value of each component is divided by the sum of the absolute values ​​of all components, thus normalizing it to a probability distribution. Then, Shannon entropy is calculated based on this probability distribution. The formula is the sum of the natural logarithms of each negative probability value. To prevent taking the logarithm of zero, a very small positive number is added to each probability value. The system entropy increase is the Shannon entropy value at the current moment minus the baseline Shannon entropy value during the healthy period. This baseline value is the average of the Shannon entropies at several evaluation moments before the healthy phase. An increase in system entropy greater than zero indicates that the angular deviation distribution of the system becomes more uniform or disordered in space, meaning that abnormal patterns are dispersed to more angular locations. This is a macroscopic statistical marker of the degradation process.

[0104] Secondly, a multiple linear regression model is established, with system entropy increase as the dependent variable and three composite indicators—main direction consistency, expansion acceleration, and time series persistence—as independent variables. The mathematical expression is:

[0105] ;

[0106] In the formula, For the first The overall entropy of the system increases at each evaluation time. This is the intercept term of the regression model; , , These are the regression coefficients; , , These are, respectively, the main direction consistency variable, the expansion acceleration variable, and the time series persistence variable; For the first The regression residuals of each sample are used. Ordinary least squares is used to estimate the regression coefficients based on all time series data. The adjusted coefficient of determination of this regression model is the desired association strength.

[0107] Step S530: Based on the correlation strength and spatiotemporal anomaly map, identify the degradation evolution stage. By matching the evolution trend of the correlation strength and the diffusion pattern of the spatiotemporal anomaly map with the preset degradation pattern library, the degradation evolution stage is identified.

[0108] The evolution trend of correlation strength and the diffusion pattern of spatiotemporal anomaly maps are matched with a pre-defined degradation pattern library to identify degradation evolution stages. This embodiment constructs a degradation pattern library containing four stages, each defined by a set of feature fingerprints, including isolated anomaly period, regional diffusion period, global degradation period, and near-failure period.

[0109] This step uses a feature vector template matching method, combined with hysteresis logic, to determine the current stage of system degradation. The specific process is as follows:

[0110] For the current moment to be evaluated, its real-time feature vector is first calculated. This vector consists of seven indicators: the total area of ​​the current abnormal region, the number of connected regions, consistency of the main direction, expansion acceleration, temporal persistence, overall system entropy increase, and the correlation strength between the above local indicators and the overall entropy increase. Subsequently, this real-time feature vector is matched against the standard feature templates corresponding to four predefined degradation stages (stages I, II, III, and IV, in sequence). The matching method calculates the Euclidean distance between the real-time feature vector and the standard template of each stage, which measures the overall similarity between the two. By comparing these four distance values, the stage with the smallest distance is taken as the preliminary matching result. To prevent frequent stage jumps between adjacent evaluation points due to measurement noise or instantaneous fluctuations, this method introduces a hysteresis constraint mechanism. The rule is: once the system is determined to have entered a higher stage, to revert to a lower stage, the real-time feature vectors at at least three consecutive evaluation time points must be closer to the standard template of the lower stage before a degradation rollback is allowed. This mechanism ensures the stability of the stage identification results and avoids oscillations in the results. Finally, the system outputs a stable degradation evolution stage determination result after hysteresis constraint processing.

[0111] Step S600: Based on the deterioration evolution stage, predict the service life and failure range to obtain the remaining service life prediction value and the angle and azimuth range of potential failure points.

[0112] Further, step S600 includes steps S610 to S630.

[0113] Step S610: Perform remaining lifetime curve mapping processing according to the degradation evolution stage. By inputting the dynamic parameters of the current degradation mode into the physical empirical degradation model describing the relationship between performance degradation and time, the preliminary remaining lifetime prediction value and its statistical confidence interval are calculated.

[0114] Step S620: Perform probability extrapolation of the fault angle range based on the dynamic probability baseline. By analyzing the evolution of the probability density function of different angle regions in the dynamic probability baseline over time, calculate the estimated time for the characteristic value of each angle region to drop to the failure threshold, and obtain the set of potential fault angle regions and their temporal sequence relationship based on time consistency clustering.

[0115] Step S630: Based on the preliminary remaining life prediction value, statistical confidence interval, angle region set and time sequence relationship, perform prediction result integration and output processing. By calibrating and fusing the time point of life prediction with the expected failure time point of angle region set, the remaining life prediction value and the angle azimuth interval of potential failure points are obtained.

[0116] Specifically, this process comprehensively utilizes multiple physical empirical degradation models, spatiotemporal extrapolation techniques based on dynamic probability baselines, and a prediction result fusion and calibration mechanism to ultimately generate complete prediction results containing information on remaining service life and potential failure locations. First, based on the currently identified degradation stage, an appropriate degradation model is selected for remaining service life mapping: a linear model is used to calculate the preliminary remaining service life and confidence interval during the early to mid-stages of degradation; an exponential model is used during the accelerated degradation phase to capture the accelerated performance decline trend; and a survival analysis model is used to assess risk near the failure stage. Next, by analyzing the historical evolution patterns of the dynamic probability baselines for each angular region, time series prediction methods are used to extrapolate the estimated time when the characteristic values ​​of each angular region will decrease to the failure threshold. Based on the similarity of the estimated times, temporal consistency clustering is performed on all angular regions to identify potential failure clusters driven by the same physical mechanism and expected to fail at similar times, and their temporal sequence relationships are established. Finally, the lifetime prediction and spatial prediction results obtained from the previous two steps are cross-calibrated and fused: the global remaining lifetime prediction value based on the dynamic model is compared and weighted with the estimated time of the earliest failure cluster to obtain the final remaining lifetime prediction value and its confidence interval after time consistency verification; at the same time, based on the relationship between the estimated failure time and the final remaining lifetime of each failure cluster, they are classified into near-term risks and long-term risks, thereby generating a list of potential failure points containing specific angular azimuth intervals, estimated failure times, and confidence levels. The final output of the entire process integrates the remaining lifetime prediction value and its statistical interval, degradation stage identifiers, the classified list of potential failure points, and prediction quality assessment indicators, forming decision support information that takes into account both time warning and spatial positioning.

[0117] Example 2:

[0118] like Figure 2 As shown, this embodiment provides a rotating electrical connector fault prediction system based on resistance timing characteristics. The system includes:

[0119] The acquisition module 901 is used to acquire the original dynamic resistance sampling sequence generated by the multi-channel contact pair and the original rotation angle timing signal synchronized with the resistance sampling time of the rotating electrical connector in the running state.

[0120] The segmentation module 902 is used to segment the rotation cycle synchronization sequence based on the original dynamic resistance sampling sequence and the original rotation angle timing signal. By aligning and truncating the original dynamic resistance sampling sequence and the original rotation angle timing signal based on the boundary of the complete mechanical rotation cycle, a set of resistance-angle synchronization segments corresponding to the physical rotation cycle is obtained.

[0121] Module 903 is used to construct angular dimension features based on the resistance-angle synchronization segment set. By calculating the statistical deviation of the resistance value in each angle interval of the resistance-angle synchronization segment set relative to the resistance distribution of historical segments, a time-series feature matrix characterizing the abnormal level of contact state at each angle position is generated.

[0122] Modeling module 904 is used to model the health baseline based on the time series feature matrix. By constructing probability density functions for the features of different angle regions in the time series feature matrix, a dynamic probability baseline is obtained.

[0123] The identification module 905 is used to perform anomaly propagation dynamics modeling based on the temporal feature matrix and the dynamic probability baseline. By detecting the abnormal angle region where the temporal feature matrix deviates from the dynamic probability baseline and calculating the correlation strength between the abnormal angle region and the overall feature entropy increase of the system, the degradation evolution stage is identified.

[0124] The prediction module 906 is used to predict the lifespan and failure range based on the deterioration evolution stage, and obtain the remaining service life prediction value and the angular and azimuth range of potential failure points.

[0125] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for predicting faults in rotating electrical connectors based on resistance timing characteristics, characterized in that, include: Acquire the original dynamic resistance sampling sequence generated by the multi-channel contact pair and the original rotation angle timing signal synchronized with the resistance sampling time of the rotating electrical connector in operation. The rotation cycle synchronization sequence is segmented based on the original dynamic resistance sampling sequence and the original rotation angle timing signal. By aligning and truncating the original dynamic resistance sampling sequence and the original rotation angle timing signal based on the boundary of the complete mechanical rotation cycle, a set of resistance-angle synchronization segments corresponding to the physical rotation cycle is obtained. Angle dimension features are constructed based on the resistance-angle synchronization segment set. By calculating the statistical deviation of the resistance value in each angle interval of the resistance-angle synchronization segment set relative to the resistance distribution of historical segments, a time-series feature matrix characterizing the abnormal level of contact state at each angle position is generated. A health baseline is modeled based on the time-series feature matrix. A dynamic probability baseline is obtained by constructing probability density functions for the features of different angle regions in the time-series feature matrix. Anomaly propagation dynamics modeling is performed based on the time-series feature matrix and the dynamic probability baseline. By detecting abnormal angle regions where the time-series feature matrix deviates from the dynamic probability baseline and calculating the correlation strength between the abnormal angle regions and the overall feature entropy increase of the system, the deterioration evolution stage is identified. Based on the aforementioned degradation evolution stages, lifespan and failure range are predicted to obtain the remaining lifespan prediction value and the angular and azimuth range of potential failure points. The health baseline modeling based on the time-series feature matrix includes: The angular region baseline distribution initialization process is performed based on the time-series feature matrix. The initial angular region probability density function is obtained by fitting the initial probability density function to the feature data of each angular region in the time-series feature matrix during the healthy operation phase. Based on the time-series feature matrix, regional confidence weights are calculated. By analyzing the volatility and stability of historical feature data for each angular region and combining the theoretical mechanical wear intensity of the rotating electrical connector in the angular region, the confidence weight coefficients are calculated. Dynamic probability baseline fusion processing is performed based on the initial angle region probability density function and the confidence weight coefficient. By using the confidence weight coefficient to perform weighted fusion and adaptive update of the initial angle region probability density function, a dynamic probability baseline is obtained.

2. The method for predicting faults in rotating electrical connectors based on resistance timing characteristics according to claim 1, characterized in that, The rotation period synchronization sequence is segmented based on the original dynamic resistance sampling sequence and the original rotation angle timing signal, including: The original dynamic resistance sampling sequence and the original rotation angle timing signal are aligned to obtain a dynamic resistance-angle signal pair with a synchronous timestamp by aligning the initial acquisition times of the two signals. Based on the dynamic resistance-angle signal pair, the rotation cycle boundary identification process is performed. By periodically detecting the extreme points of the original rotation angle time sequence signal, the start and end timestamps of each complete mechanical rotation cycle are extracted to obtain the rotation cycle boundary sequence. Synchronization segment extraction is performed based on the rotation period boundary sequence and the dynamic resistance-angle signal pair. By extracting the corresponding resistance and angle data segments within each period from the dynamic resistance-angle signal pair according to the time scale range indicated by the rotation period boundary sequence, the resistance-angle synchronization segment set is obtained.

3. The method for predicting faults in rotating electrical connectors based on resistance timing characteristics according to claim 1, characterized in that, Based on the set of resistance-angle synchronization segments, angular dimension features are constructed, including: The resistance-angle synchronization segment set is uniformly divided into angle intervals. By dividing the circumferential angle space into several continuous sectors, the resistance data in all segment sets are classified into the corresponding angle sectors according to the synchronized angle values, resulting in a set of resistance subsequences organized by angle sectors. The statistical deviation calculation of the angle sector is performed based on the set of resistance subsequences. For each angle sector, the statistical distribution parameters of the resistance value at the current moment are calculated, and the statistical distribution parameters are compared with the baseline distribution formed by the statistical distribution parameters corresponding to all historical moments in the angle sector. The statistical deviation value representing the current degree of deviation is then calculated. The spatiotemporal feature matrix is ​​generated based on the statistical deviation values ​​of all angle sectors. By arranging the statistical deviation values ​​of all angle sectors at the same time in angular order, a row of feature vectors is formed. The feature vectors at different times are stacked along the time dimension to generate a time-series feature matrix.

4. The method for predicting faults in rotating electrical connectors based on resistance timing characteristics according to claim 1, characterized in that, Anomaly propagation dynamics modeling is performed based on the temporal feature matrix and the dynamic probability baseline, including: Anomaly region detection processing is performed based on the temporal feature matrix and the dynamic probability baseline. The probability density is calculated by substituting the feature values ​​of each time and each angle region in the temporal feature matrix into the corresponding dynamic probability baseline, and regions with probability density lower than a set threshold are marked as anomalies to obtain a spatiotemporal anomaly map. The correlation strength of the anomaly is quantified based on the spatiotemporal anomaly map. By analyzing the connectivity of the abnormal regions in the spatiotemporal anomaly map, the spatial displacement and area changes in continuous rotation cycles, and calculating the overall complexity and disorder change index of the spatiotemporal anomaly map, the correlation strength between the abnormal angle region and the overall characteristic entropy increase of the system is quantified. The degradation evolution stage is identified by matching the evolution trend of the correlation strength and the diffusion pattern of the spatiotemporal anomaly map with a preset degradation pattern library.

5. The method for predicting faults in rotating electrical connectors based on resistance timing characteristics according to claim 1, characterized in that, Based on the aforementioned degradation evolution stages, lifetime and failure range predictions are performed, including: Based on the degradation evolution stage, the remaining lifetime curve mapping process is performed. By inputting the dynamic parameters of the current degradation mode into the physical empirical degradation model describing the relationship between performance degradation and time, the preliminary remaining lifetime prediction value and its statistical confidence interval are calculated. Based on the dynamic probability baseline, the probability extrapolation of the fault angle range is performed. By analyzing the evolution of the probability density function of different angle regions in the dynamic probability baseline over time, the estimated time for the characteristic value of each angle region to drop to the failure threshold is calculated. Then, based on time consistency clustering, the set of potential fault angle regions and their temporal sequence relationship are obtained. Based on the preliminary remaining life prediction value, the statistical confidence interval, the angle region set, and the time sequence relationship, the prediction results are integrated and output. By calibrating and fusing the time point of the life prediction with the expected failure time point of the angle region set, the remaining life prediction value and the angle azimuth interval of the potential failure point are obtained.

6. The method for predicting faults in rotating electrical connectors based on resistance timing characteristics according to claim 4, characterized in that, The anomaly correlation strength is quantified based on the spatiotemporal anomaly map, including: Based on the spatiotemporal anomaly map, the spatiotemporal characteristics of the anomaly region are extracted. By identifying and tracking the marked anomaly region in the spatiotemporal anomaly map, the spatial centroid displacement vector, the rate of change of the region area, and the spatial overlap between adjacent anomaly regions in consecutive rotation cycles are calculated to obtain the first set of quantitative indicators. The first set of quantitative indicators is used to calculate the correlation strength index. The consistency of the main direction of anomaly propagation is calculated based on the spatial centroid displacement vector, the acceleration of anomaly expansion is calculated based on the regional area change rate, and the temporal persistence of the anomaly pattern is calculated based on the spatial overlap, to obtain the second set of quantitative indicators. Based on the spatiotemporal anomaly map and the second set of quantitative indicators, a system entropy increase correlation analysis is performed. By performing multiple regression analysis on the main direction consistency, expansion acceleration and time series persistence indicators in the second set of quantitative indicators and the values ​​of the overall system characteristic entropy increase, the correlation strength characterizing the degree of correlation between local anomaly dynamics and the disordered growth of the overall system is calculated.

7. A fault prediction system for rotating electrical connectors based on resistance timing characteristics, characterized in that, include: The acquisition module is used to acquire the original dynamic resistance sampling sequence generated by the multi-channel contact pair and the original rotation angle timing signal synchronized with the resistance sampling time of the rotating electrical connector in the operating state. The segmentation module is used to segment the rotation cycle synchronization sequence according to the original dynamic resistance sampling sequence and the original rotation angle timing signal. By aligning and truncating the original dynamic resistance sampling sequence and the original rotation angle timing signal based on the boundary of the complete mechanical rotation cycle, a set of resistance-angle synchronization segments corresponding to the physical rotation cycle is obtained. The construction module is used to construct angular dimension features based on the resistance-angle synchronization segment set. By calculating the statistical deviation of the resistance value in each angle interval of the resistance-angle synchronization segment set relative to the resistance distribution of historical segments, a time-series feature matrix characterizing the abnormal level of contact state at each angle position is generated. The modeling module is used to model a health baseline based on the time-series feature matrix, and to obtain a dynamic probability baseline by constructing probability density functions for the features of different angle regions in the time-series feature matrix. The identification module is used to perform anomaly propagation dynamics modeling based on the temporal feature matrix and the dynamic probability baseline. By detecting the abnormal angle region where the temporal feature matrix deviates from the dynamic probability baseline and calculating the correlation strength between the abnormal angle region and the overall feature entropy increase of the system, the degradation evolution stage is identified. The prediction module is used to predict the lifespan and fault range based on the degradation evolution stage, and obtain the remaining service life prediction value and the angle and orientation range of potential fault locations. The health baseline modeling based on the time-series feature matrix includes: The angular region baseline distribution initialization process is performed based on the time-series feature matrix. The initial angular region probability density function is obtained by fitting the initial probability density function to the feature data of each angular region in the time-series feature matrix during the healthy operation phase. Based on the time-series feature matrix, regional confidence weights are calculated. By analyzing the volatility and stability of historical feature data for each angular region and combining the theoretical mechanical wear intensity of the rotating electrical connector in the angular region, the confidence weight coefficients are calculated. Dynamic probability baseline fusion processing is performed based on the initial angle region probability density function and the confidence weight coefficient. By using the confidence weight coefficient to perform weighted fusion and adaptive update of the initial angle region probability density function, a dynamic probability baseline is obtained.

8. The rotating electrical connector fault prediction system based on resistance timing characteristics according to claim 7, characterized in that, The segmentation module includes: The first segmentation unit is used to perform signal timing alignment processing based on the original dynamic resistance sampling sequence and the original rotation angle timing signal. By aligning the initial acquisition times of the two signals, a dynamic resistance-angle signal pair with a synchronous timestamp is obtained. The second segmentation unit is used to perform rotation cycle boundary identification processing based on the dynamic resistance-angle signal. By performing periodic extreme point detection on the original rotation angle time sequence signal, the start and end timestamps of each complete mechanical rotation cycle are extracted to obtain the rotation cycle boundary sequence. The third segmentation unit is used to perform synchronization segment extraction processing based on the rotation period boundary sequence and the dynamic resistance-angle signal pair. By extracting the corresponding resistance and angle data segments in each period from the dynamic resistance-angle signal pair according to the time scale range indicated by the rotation period boundary sequence, the set of resistance-angle synchronization segments is obtained.

9. The rotating electrical connector fault prediction system based on resistance timing characteristics according to claim 7, characterized in that, The building module includes: The first construction unit is used to perform uniform division of angle intervals according to the resistance-angle synchronization segment set. By dividing the circumferential angle space into several continuous sectors, the resistance data in all segment sets are classified into the corresponding angle sectors according to the synchronized angle values, thereby obtaining a set of resistance sub-sequences organized by angle sectors. The second construction unit is used to perform angular sector statistical deviation calculation processing based on the set of resistance sub-sequences. For each angular sector, the statistical distribution parameters of the resistance value at the current moment are calculated, and the statistical distribution parameters are compared with the baseline distribution formed by the statistical distribution parameters corresponding to all historical moments in the angular sector to calculate the statistical deviation value characterizing the current degree of deviation. The third construction unit is used to generate a spatiotemporal feature matrix based on the statistical deviation values ​​of all angle sectors. By arranging the statistical deviation values ​​of all angle sectors at the same time in angular order, a row of feature vectors is formed, and the feature vectors at different times are stacked along the time dimension to generate a time-series feature matrix.

Images