A USB data line harness on-off and current detection system

Abstract

The application relates to the technical field of electric variable measurement test, and particularly discloses a USB data line bundle on-off and current detection system, which applies a step current excitation to a measured cable, synchronously collects multiple response signals on a power supply line and a communication line, performs blind source separation processing on the multiple original time sequence signals, decomposes the multiple original time sequence signals into a power supply response independent component and a communication disturbance independent component, respectively reconstructs phase spaces of the two independent components and extracts recursive quantization indexes, combines the recursive quantization indexes into a health feature set, projects the health feature set to a low-dimensional intrinsic manifold, calculates a geodesic distance between a current coordinate and a health reference coordinate, generates a health deviation degree sequence, inputs the health deviation degree sequence into a particle filtering algorithm, recursively estimates an evolution trend of the health deviation degree through sequential importance sampling, and outputs a probability estimation of a remaining service life. The application solves the problem that chip communication crosstalk under large current stress cannot be quantitatively evaluated, and realizes a technical leap from static threshold judgment to dynamic health management.

Description

Technical Field

[0001] This invention relates to the field of electrical variable measurement and testing technology, specifically to a USB data cable harness continuity and current detection system. Background Technology

[0002] USB data cables are crucial for power supply and data transmission in electronic devices, and their reliability directly impacts the operational safety of terminal devices. Current technologies primarily focus on static continuity testing and DC voltage drop measurements, using fixed thresholds to determine cable quality. However, these methods can only identify obvious faults such as physical open circuits or excessive resistance, failing to detect potential performance degradation processes such as conductor material fatigue, shielding degradation, and interference with chip communication.

[0003] As the power levels of fast charging protocols continue to increase, the dynamic response characteristics and communication stability of data cables under high current stress are gradually becoming the core factors limiting their lifespan. Existing testing methods use step-by-step testing logic, which cannot monitor the chip's communication status while applying a step current surge. This results in the inability to effectively capture and quantify the crosstalk reset or logic disorder caused by the voltage drop during high current injection on the E-Marker chip. Consequently, it is difficult to assess the reliability of the cable's collaborative operation in real fast charging scenarios, and even more difficult to make a probabilistic estimate of the remaining usable lifespan. This leads to data cables with early failure risks suddenly malfunctioning in actual use. Summary of the Invention

[0004] The purpose of this invention is to provide a USB data cable harness continuity and current detection system to solve the problems mentioned above.

[0005] The objective of this invention can be achieved through the following technical solutions:

[0006] A USB data cable harness continuity and current detection system includes:

[0007] The excitation response acquisition module applies a preset step current excitation to the USB data cable under test, and simultaneously acquires the voltage response signal, current response signal and logic level fluctuation signal on the power line inside the cable during the excitation process, to obtain multiple raw timing signals.

[0008] The signal blind source separation module performs blind source separation processing on multiple original timing signals, decomposing them into independent components of power supply response that characterize the electrical characteristics of cable conductors and independent components of communication disturbance that characterize the degree of disturbance to the communication link.

[0009] The nonlinear dynamics feature extraction module reconstructs the phase space of the independent components of power supply response and communication disturbance, obtains the phase point trajectory in the high-dimensional phase space, calculates the recursion graph of the phase point trajectory, and extracts recursive quantization indicators from the recursion graph, including recursion rate, determinism, laminar flow and average diagonal length. The recursive quantization indicators are combined into a health feature set characterizing the dynamic characteristics of the cable.

[0010] The health degradation manifold mapping module treats the health feature set as points in a high-dimensional space, projects them onto a low-dimensional intrinsic manifold, obtains the low-dimensional embedded coordinates of the health features, and calculates the geodesic distance between the current coordinates and the health baseline coordinates by monitoring the movement trajectory of the low-dimensional embedded coordinates, thereby generating a health deviation sequence.

[0011] The remaining lifetime particle filter prediction module inputs the health deviation sequence into the particle filter algorithm, updates the particle weights through sequential importance sampling, estimates the evolution trend of health deviation, and outputs a probability estimate of the remaining usable lifetime when the deviation predicted by the particles exceeds a preset threshold.

[0012] As a further aspect of the present invention: the blind source separation processing of multiple original time-series signals specifically includes:

[0013] Perform short-time Fourier transform on multiple raw time-series signals to obtain the time-spectrum matrix;

[0014] Based on the distribution differences of energy peaks and valleys in the time-frequency matrix, a first masking matrix corresponding to the power supply response characteristics and a second masking matrix corresponding to the communication disturbance characteristics are constructed in the time-frequency domain.

[0015] The first and second masking matrices are multiplied by the time-spectrum matrix respectively to obtain the separated power supply response time-spectrum and communication disturbance time-spectrum.

[0016] Inverse short-time Fourier transform is performed on the power supply response spectrum and the communication disturbance spectrum to reconstruct the independent components of the power supply response and the communication disturbance.

[0017] As a further aspect of the present invention: the construction of a first masking matrix corresponding to the power supply response characteristics and a second masking matrix corresponding to the communication disturbance characteristics in the time-frequency domain specifically includes:

[0018] Search the energy peak position of each frequency point along the time axis of the time spectrum matrix to obtain the energy accumulation trajectory of the power supply response at each frequency point;

[0019] Search along the frequency axis of the time spectrum matrix to find the energy valley position at each time point, and obtain the energy depression distribution of communication disturbance at each time.

[0020] Based on the energy accumulation trajectory and energy depression distribution, a first binary template adapted to the time-frequency form of the power supply response and a second binary template adapted to the time-frequency form of the communication disturbance are generated on the time-frequency plane.

[0021] The first binary template and the second binary template are convolved with a preset smoothing window function to obtain the first masking matrix and the second masking matrix with smooth transition bands.

[0022] As a further aspect of the present invention: the extraction process of the recursive quantification index is as follows:

[0023] The Euclidean distance between each pair of phase points is calculated for the phase point trajectories of the independent components of the power supply response and the independent components of the communication disturbance, respectively, to obtain the recursive distance matrix;

[0024] Based on the numerical distribution of each element in the recursive distance matrix, adaptive thresholds are used to determine the recursive judgment thresholds for the independent components of power supply response and communication disturbance, respectively.

[0025] The positions of elements in the recursive distance matrix that are less than or equal to the corresponding recursive decision threshold are marked as recursive points, and the positions of elements that are greater than the corresponding recursive decision threshold are marked as non-recursive points, thus generating recursive graphs of independent components of power supply response and independent components of communication disturbance.

[0026] The recursion graphs of independent components of power supply response and independent components of communication disturbance are statistically analyzed for recursion point density, diagonal structure distribution and vertical line structure distribution, respectively, to obtain recursion quantification indicators including recursion rate, determinism, laminar flow and average diagonal length.

[0027] As a further aspect of the present invention: the process for determining the recursive decision threshold is as follows:

[0028] Probability density estimation is performed on the recursive distance matrices of the independent components of power supply response and communication disturbance, respectively, to obtain the distribution curves of the recursive distance values ​​in the amplitude range;

[0029] Search for local minima on the distribution curve and use the distance values ​​corresponding to the local minima as candidate thresholds for the independent components of power supply response and communication disturbance, respectively.

[0030] Verify whether the candidate threshold satisfies the requirement that the recursion point coverage is within a preset confidence interval. If it is satisfied, the candidate threshold is determined as the final recursive decision threshold. If it is not satisfied, backtrack to the second smallest local minimum point on the distribution curve for iterative verification.

[0031] As a further aspect of the present invention: the generation of the health deviation sequence specifically includes:

[0032] Calculate the Euclidean distance between each high-dimensional point in the health feature set, determine a preset number of neighboring points for each high-dimensional point, and construct a neighborhood graph based on the connection relationships between the neighboring points.

[0033] A shortest path search algorithm is used on the neighborhood graph to calculate the geodesic distance between all pairs of high-dimensional points and obtain the geodesic distance matrix.

[0034] By performing a coordinate mapping that preserves the distance structure on the geodesic distance matrix, the low-dimensional embedded coordinates corresponding to each high-dimensional point are obtained.

[0035] The algorithm inputs the low-dimensional embedded coordinates of the current moment and the pre-stored health baseline coordinates into the geodesic tracing algorithm. It searches for the shortest path between the two points along the neighborhood graph, accumulates the Euclidean distance between adjacent points on the path as the geodesic distance at the current moment, and outputs the health deviation sequence after traversing all moments.

[0036] As a further aspect of the present invention: the step of searching for the shortest path between two points along the neighborhood graph specifically includes:

[0037] Starting with the health baseline coordinates as the search starting point and the current low-dimensional embedding coordinates as the search ending point, initialize the forward open list and the backward open list;

[0038] Expand outward from the forward open list by selecting nodes connected to the starting point, and backward from the backward open list by selecting nodes connected to the ending point. Alternately perform the expansion operation until the expansion regions in the two directions overlap.

[0039] During the expansion process, the expansion priority is dynamically adjusted based on the deviation angle between the node and the line connecting the starting point and the ending point, prioritizing the expansion of neighboring nodes with smaller directional deviations.

[0040] When the extended regions in two directions overlap, backtrack from the overlapping node to the starting point and the ending point to obtain the complete path, and accumulate the Euclidean distance between adjacent nodes on the path as the geodesic distance at the current moment.

[0041] As a further aspect of the present invention: the probability estimation of the output remaining usable lifetime specifically includes:

[0042] Using the latest value in the health deviation sequence as the initial state, an initial particle set containing a preset number of particles is generated, and each particle is assigned the same initial weight.

[0043] Based on the sequential importance sampling principle in the particle filtering algorithm, each particle is substituted into the pre-constructed state transition equation to predict the health deviation value at the next moment, and the weight of each particle is updated with the actual health deviation observation value at the current moment.

[0044] When the weight difference of each particle exceeds a preset threshold, the particle set is resampled to remove particles with too low weight and duplicate particles with too high weight, so as to keep the total number of particles constant.

[0045] The resampled particle set is continuously recursively evolved forward, and the moment when the predicted health deviation of each particle first exceeds the preset threshold is counted. The distribution of the moment when each particle exceeds the threshold is used as the probability estimate of the remaining usable lifetime.

[0046] The beneficial effects of this invention are:

[0047] (1) This invention achieves coordinated decoupling of power supply response and communication disturbance through blind source separation and recursive quantization analysis, breaking the limitation of the time sequence separation of "continuity test" and "current test" in traditional detection, and can truly reproduce the crosstalk effect of large current impact on E-Marker chip communication in fast charging scenario, thus improving the evaluation accuracy of the reliability of cable working together under dynamic stress.

[0048] (2) This invention introduces manifold learning and particle filtering algorithms to upgrade the conventional “qualified / unqualified” threshold judgment to continuous quantitative assessment of cable health status and probability prediction of remaining life. It can identify early failure risks that have not yet exceeded the internal resistance but have already shown performance degradation, providing data support for predictive maintenance of equipment and effectively avoiding charging interruption or data transmission failure caused by sudden failure of data line. Attached Figure Description

[0049] The invention will now be further described with reference to the accompanying drawings.

[0050] Figure 1 This is a system block diagram of the present invention. Detailed Implementation

[0051] 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.

[0052] Please see Figure 1 As shown, the present invention is a USB data cable harness continuity and current detection system, comprising:

[0053] The excitation response acquisition module applies a preset step current excitation to the USB data cable under test, and simultaneously acquires the voltage response signal, current response signal and logic level fluctuation signal on the power line inside the cable during the excitation process, to obtain multiple raw timing signals.

[0054] The signal blind source separation module performs blind source separation processing on multiple original timing signals, decomposing them into independent components of power supply response that characterize the electrical characteristics of cable conductors and independent components of communication disturbance that characterize the degree of disturbance to the communication link.

[0055] The nonlinear dynamics feature extraction module reconstructs the phase space of the independent components of power supply response and communication disturbance, obtains the phase point trajectory in the high-dimensional phase space, calculates the recursion graph of the phase point trajectory, and extracts recursive quantization indicators from the recursion graph, including recursion rate, determinism, laminar flow and average diagonal length. The recursive quantization indicators are combined into a health feature set characterizing the dynamic characteristics of the cable.

[0056] The health degradation manifold mapping module treats the health feature set as points in a high-dimensional space, projects them onto a low-dimensional intrinsic manifold, obtains the low-dimensional embedded coordinates of the health features, and calculates the geodesic distance between the current coordinates and the health baseline coordinates by monitoring the movement trajectory of the low-dimensional embedded coordinates, thereby generating a health deviation sequence.

[0057] The remaining lifetime particle filter prediction module inputs the health deviation sequence into the particle filter algorithm, updates the particle weights through sequential importance sampling, estimates the evolution trend of health deviation, and outputs a probability estimate of the remaining usable lifetime when the deviation predicted by the particles exceeds a preset threshold.

[0058] In the excitation response acquisition module, a preset step current excitation is applied to the USB data cable under test. Simultaneously, the voltage response signal and current response signal on the power line inside the cable, as well as the logic level fluctuation signal on the communication line, are acquired during the excitation process to obtain multiple raw timing signals, specifically including:

[0059] The arbitrary waveform generator in the testing device is activated, and electrically connected to the power and ground pins of the USB data cable under test. This arbitrary waveform generator outputs a preset step current signal. The amplitude of this step current signal is set to match the current level of the fast charging protocol to be simulated, and the rise time is controlled in the microsecond range to simulate the current surge in a real fast charging scenario. During the duration of the applied step current, a first data acquisition card connected in parallel with the power pin of the data cable under test synchronously acquires the voltage response signal on the power line; a current probe connected in series with the power pin of the data cable under test, in conjunction with a second data acquisition card, synchronously acquires the current response signal on the power line; simultaneously, a logic analyzer connected to the communication pin of the data cable under test synchronously acquires the logic level fluctuation signal on the communication line. The first data acquisition card, the second data acquisition card, and the logic analyzer are triggered by the same clock source to ensure strict alignment of the three signals on the time axis. The acquired analog voltage signals, analog current signals, and digital logic level signals are filtered and amplitude adjusted by the signal conditioning circuit, and then converted into digital discrete sequences by the analog-to-digital converter. Finally, multiple original timing signals containing timestamps are formed and stored in the buffer of the test device.

[0060] In the signal blind source separation module, blind source separation processing is performed on multiple original timing signals, decomposing them into independent components of power supply response characterizing the electrical characteristics of the cable conductor and independent components of communication disturbance characterizing the degree of disturbance to the communication link. Specifically, these include:

[0061] Multiple raw timing signals are read from the buffer. These raw timing signals include discrete sequences of voltage response, current response, and logic level fluctuations. First, each discrete sequence is windowed using a Hanning window with a length of 256 sampling points and an overlap of 50% between adjacent windows. A Fast Fourier Transform is then performed on each windowed frame to obtain its spectrum. The spectra of all frames are then arranged chronologically and combined into a three-dimensional time-spectrum matrix. Rows in this matrix correspond to frequency points, columns to time frames, and element values ​​represent the energy amplitude at the corresponding frequency and time point.

[0062] A masking matrix is ​​constructed based on the differences in energy distribution within the time-spectrum matrix. For extracting power supply response features, along the time axis of the time-spectrum matrix, for each fixed frequency point, the energy amplitude across all time frames corresponding to that frequency point is traversed. The location where the energy amplitude reaches a local maximum is searched, and the times when local maximums occur at all frequency points are recorded. These time points are then connected to form the energy accumulation trajectory of the power supply response at each frequency point. For extracting communication disturbance features, along the frequency axis of the time-spectrum matrix, for each fixed time frame, the energy amplitude across all frequency points corresponding to that time frame is traversed, and the frequency location where the energy amplitude reaches a local minimum is searched. The frequency points where local minimums occur at all time frames are recorded, and these frequency points are then connected to form the energy dip distribution of the communication disturbance at each time point.

[0063] A binary template is constructed on the time-frequency plane based on the energy accumulation trajectory and energy depression distribution. For the power supply response, all time-frequency grid points near the energy accumulation trajectory are marked with a value of 1, and the remaining grid points are marked with a value of 0, generating a first binary template adapted to the time-frequency pattern of the power supply response. For communication disturbances, all time-frequency grid points near the energy depression distribution are marked with a value of 1, and the remaining grid points are marked with a value of 0, generating a second binary template adapted to the time-frequency pattern of the communication disturbances. The "nearby" range is defined as a region three grid units to the left and right of the trajectory or distribution location.

[0064] A two-dimensional convolution operation is performed between the first binary template and a preset smoothing window function. The smoothing window function is a two-dimensional Gaussian window with a window size of 5x5 grid and a standard deviation of 1.2. The convolution operation involves sliding the Gaussian window across the binary template, performing a weighted average of the values ​​covered within the window, and using the resulting value as the output value at the center point. After convolution, the regions with values ​​of 1 in the first binary template diffuse outward and smoothly transition with the surrounding regions, forming a first masking matrix. The elements in this matrix are continuously distributed between 0 and 1. The same two-dimensional convolution operation is performed on the second binary template using the same Gaussian window function to obtain the second masking matrix.

[0065] The first masking matrix is ​​multiplied by the original time-frequency spectrum matrix. Multiplication involves multiplying the element values ​​at the same positions in both matrices, resulting in a new matrix that serves as the separated power supply response time-frequency spectrum. The second masking matrix is ​​then multiplied by the original time-frequency spectrum matrix to obtain the separated communication disturbance time-frequency spectrum. After the multiplication, the power supply response time-frequency spectrum retains components near the energy accumulation trajectory, while the communication disturbance time-frequency spectrum retains components near the energy depression distribution.

[0066] An inverse short-time Fourier transform (ISFT) is performed on the separated power supply response time spectrum. First, the spectrum of each frame is conjugate-symmetrically expanded to satisfy the real signal condition. Then, an inverse fast Fourier transform (IFFT) is performed on each frame to obtain time-domain frame data. Finally, the frame data are arranged in chronological order and summed according to the overlap rate during windowing to reconstruct the independent power supply response components in time-series form. The same IFT steps are then applied to the separated communication disturbance time spectrum to reconstruct the independent communication disturbance components in time-series form.

[0067] In the nonlinear dynamics feature extraction module, phase space reconstruction is performed on the independent components of power supply response and communication disturbance, respectively, to obtain the phase point trajectories in the high-dimensional phase space. A recursive graph of the phase point trajectories is calculated, and recursive quantization indicators, including recursion rate, determinism, laminar flow, and average diagonal length, are extracted from the recursive graph. These recursive quantization indicators are combined into a health feature set characterizing the dynamic properties of the cable, specifically including:

[0068] First, phase space reconstruction is performed on the independent components of power supply response and communication disturbance. The embedding dimension is set to 6, and the delay time is 10 sampling points. For an independent component time series of length N, starting from the first sampling point, six consecutive sampling points are extracted to form a 6-dimensional vector, which serves as the first phase point in the high-dimensional phase space. Then, one sampling point is moved forward, and another six consecutive sampling points are extracted to form the second phase point. This process continues until the entire time series is traversed, ultimately obtaining a phase point trajectory matrix consisting of N minus 5 phase points. The rows of the matrix correspond to the phase points at different times, and the columns correspond to the coordinates of each dimension of the phase points.

[0069] For the phase trajectory matrix of the independent components of the power supply response, calculate the Euclidean distance between any two phase points. For the a-th and b-th phase points, extract their coordinate values ​​in six dimensions, calculate the difference between the coordinate values ​​in each dimension, square the differences in each dimension, sum them, and then take the square root of the sum to obtain the Euclidean distance between the two phase points. Iterate through all combinations of a and b, and fill the calculation results into a matrix whose number of rows and columns are equal to the number of phase points. The element located in row a and column b is the Euclidean distance between the a-th and b-th phase points, thus obtaining the recursive distance matrix corresponding to the independent components of the power supply response. Using the same calculation method, process the phase trajectory matrix of the independent components of the communication disturbance to obtain the recursive distance matrix corresponding to the independent components of the communication disturbance.

[0070] For the recursive distance matrix of the independent components of the power supply response, a recursive decision threshold is determined. All off-diagonal elements in the recursive distance matrix are extracted to form a set of distance value samples. Kernel density estimation is used to estimate the probability density of this set of samples. The distance values ​​are divided into 200 equally spaced intervals from the minimum to the maximum value. The number of samples falling into each interval is counted, and the number is normalized by dividing by the total number of samples to obtain the probability density distribution curve of the distance value as a function of amplitude. Local minima are searched on this distribution curve by comparing the ordinate values ​​of three adjacent points from left to right. If the ordinate value of the middle point is simultaneously less than the ordinate values ​​of the two adjacent points on either side, the distance value corresponding to the middle point is marked as a local minimum. The distance values ​​corresponding to all local minima are sorted from smallest to largest, and the distance value corresponding to the smallest local minimum is used as the candidate threshold for the independent components of the power supply response.

[0071] Verify whether the candidate threshold meets the recursion point coverage requirement. Count the number of elements in the recursion distance matrix that are less than or equal to the candidate threshold, divide this number by the total number of all off-diagonal elements in the recursion distance matrix, and obtain the recursion point coverage. The preset confidence interval has a lower limit of 5% and an upper limit of 15%. If the calculated recursion point coverage is within the interval of 5% to 15%, then the candidate threshold is determined as the final recursion decision threshold. If the recursion point coverage is less than 5%, then the distance value corresponding to the second smallest local minimum is selected as the new candidate threshold, and the recursion point coverage is recalculated for verification; if the recursion point coverage is greater than 15%, then the distance value corresponding to the larger local minimum is selected as the new candidate threshold for verification, until the recursion point coverage meets the preset confidence interval requirement.

[0072] Using the same probability density estimation, local minimum point search, candidate threshold selection, and recursive point coverage verification steps, the recursive distance matrix of the independent components of communication disturbance is processed to obtain the recursive decision threshold corresponding to the independent components of communication disturbance.

[0073] In the recursive distance matrix of the independent power supply response components, all elements with values ​​less than or equal to the recursion decision threshold are marked as recursive points, and elements with values ​​greater than the recursion decision threshold are marked as non-recursive points. This generates a binary matrix with the same size as the recursive distance matrix, which is the recursive graph of the independent power supply response components. Elements at recursive points have a value of 1, and elements at non-recursive points have a value of 0. Using the same method, a recursive graph of the independent communication disturbance components is generated based on the recursive distance matrix and its recursion decision threshold.

[0074] Statistical feature extraction is performed on the recurrence graph of the independent components of the power supply response. The recurrence rate is obtained by counting the number of elements with a value of 1 in the recurrence graph and dividing it by the total number of elements in the recurrence graph. Line segments consisting of recurrence points arranged diagonally are searched in the recurrence graph. The lengths of all line segments with a length greater than or equal to 2 are recorded, and the total number of these line segments and the total number of recurrence points contained in all line segments are counted. The total number of recurrence points is divided by the total number of line segments to obtain the average diagonal length. The length of each vertical line segment is recorded, and the total number of recurrence points contained in vertical line segments with a length greater than or equal to 2 is divided by the total number of vertical line segments to obtain laminarity. The determinism is obtained by dividing the total number of recurrence points contained in all line segments consisting of diagonally arranged recurrence points in the recurrence graph by the total number of recurrence points in the recurrence graph. The calculated recurrence rate, average diagonal length, laminarity, and determinism are combined into a four-dimensional vector as the health feature subset corresponding to the independent components of the power supply response.

[0075] Repeat the above calculations of recurrence rate, average diagonal length, laminarity, and determinism for the recursive graph of the independent components of the communication disturbance to obtain the health feature subsets corresponding to the independent components of the communication disturbance. Concatenate the health feature subsets of the two independent components to form an eight-dimensional vector, which is the health feature set characterizing the current dynamic properties of the cable.

[0076] In the health degradation manifold mapping module, the health feature set is treated as points in a high-dimensional space and projected onto a low-dimensional intrinsic manifold to obtain the low-dimensional embedded coordinates of the health features. By monitoring the movement trajectory of the low-dimensional embedded coordinates, the geodesic distance between the current coordinates and the health baseline coordinates is calculated, generating a health deviation sequence, specifically including:

[0077] First, a neighborhood graph in high-dimensional space is constructed. All health feature sets acquired during historical monitoring are arranged chronologically, with each health feature set representing a data point in an eight-dimensional space. All data points are denoted as a high-dimensional point set. For each high-dimensional point in the set, the Euclidean distance between it and all other high-dimensional points is calculated. Specifically, for any two eight-dimensional points A and B, their coordinate values ​​in each of the eight dimensions are extracted. The difference between the coordinate values ​​in each dimension is calculated, and the squares of the eight differences are summed. The square root of the sum is then taken to obtain the Euclidean distance between the two points. For the currently processed high-dimensional point, its Euclidean distances to all other points are sorted in ascending order, and the K points with the smallest distances are selected as its neighbors, where K is set to 12. To ensure the connectivity and local structure of the neighborhood graph are maintained, an adaptive adjustment mechanism is introduced: For each high-dimensional point, based on its distance to its Kth neighbor, an exponential decay function is used to calculate the connection weight between the point and each of its neighbors. The calculation formula is as follows: ;In the formula, Indicates the first The high-dimensional point and the first Connection weights between high-dimensional points Indicates the first An eight-dimensional coordinate vector of a high-dimensional point Indicates the first An eight-dimensional coordinate vector of a high-dimensional point Indicates the first The high-dimensional point and the first Euclidean distance between high-dimensional points Indicates the first The local scale parameter is the distance between a high-dimensional point and all its neighbors, specifically taking the value of the i-th high-dimensional point. The numerical value of the distance between a high-dimensional point and its Kth neighbor. Indicates the first The local scale parameter is the distance between a high-dimensional point and all its neighbors, specifically taking the value of the i-th high-dimensional point. The numerical value of the distance between a high-dimensional point and its Kth neighbor, where e represents the base of the natural logarithm.

[0078] For connections between non-nearest points, the weight is set to 0. Based on the proximity relationships between all high-dimensional points and their corresponding connection weights, a weighted undirected graph is constructed. The nodes in the graph are the various high-dimensional points, and the edges between nodes exist only between their nearest neighbors. The weight of the edge is the calculated connection weight value. This weighted undirected graph is called the neighborhood graph.

[0079] After the neighborhood graph is constructed, the geodesic distance between all pairs of high-dimensional points is calculated. The geodesic distance is defined as the sum of the weights of all edges traversed along the shortest path in the neighborhood graph. For any two high-dimensional points, the shortest path search algorithm is used to calculate the distance. The distance matrix is ​​initialized, setting the distance between directly connected neighboring points to the value of the connection weight, and setting the distance between non-neighboring points to infinity. Then, all high-dimensional points are traversed as intermediate nodes. For each pair of source and target nodes, it is checked whether there exists a path through the current intermediate node that shortens the distance. If so, the distance value between the source and target nodes is updated. The above traversal process is repeated until the distance between all node pairs no longer changes. The final distance matrix is ​​the geodesic distance matrix, where the distance between the nodes is calculated as follows: Line number The elements of the column represent the first... The high-dimensional point and the first Geodesic distance between high-dimensional points.

[0080] A coordinate mapping that preserves the distance structure is performed on the geodesic distance matrix to obtain the low-dimensional embedded coordinates corresponding to each high-dimensional point. The target low-dimensional space is set to three dimensions. First, the geodesic distance matrix is ​​bicentered by subtracting the row mean, column mean, and overall mean from the squared distances to obtain the inner product matrix. Then, eigenvalue decomposition is performed on the inner product matrix to extract the three largest eigenvalues ​​and their corresponding eigenvectors. Each eigenvector is multiplied by the square root of the corresponding eigenvalue to obtain the projection values ​​on the three coordinate axes. The three projection values ​​are combined into a three-dimensional vector, which is the low-dimensional embedded coordinate of the high-dimensional point. The same mapping operation is performed on all high-dimensional points to obtain the low-dimensional embedded coordinate sequence of all historical moments. The low-dimensional embedded coordinates corresponding to the optimal health state (e.g., the initial state when the cable is first put into use) are extracted and stored as the health baseline coordinates.

[0081] For the current moment to be evaluated, obtain the corresponding low-dimensional embedded coordinates, denoted as the current coordinates. Input the health baseline coordinates and the current coordinates into the geodesic tracing algorithm to search for the shortest path between the two points along the neighborhood graph. The specific search process adopts a bidirectional progressive expansion strategy. Using the health baseline coordinates as the search starting point, initialize a forward open list containing the starting point and all nodes directly reachable from the starting point via an edge, and record the path length from the starting point to these nodes as the weight of the corresponding edge. Simultaneously, using the current coordinates as the search ending point, initialize a backward open list containing the ending point and all nodes directly reachable from the ending point via an edge, and record the path length from the ending point to these nodes as the weight of the corresponding edge. Set the expansion priority for the forward search and the expansion priority for the backward search.

[0082] During the expansion process, the node with the highest priority is selected from the current forward open list for expansion. The priority is determined by the deviation angle between the node and the line connecting the start and end points. First, the direction vector of the line between the start and end points is calculated. For the node to be expanded, the cosine of the angle between the path direction formed after passing through the node from the start point and the direction of this line is calculated. The expression for calculating the cosine of the angle is: In the formula, The cosine value represents the deviation angle. The three-dimensional coordinate vector representing the starting point. The three-dimensional coordinate vector representing the endpoint. This represents the three-dimensional coordinate vector of the node to be expanded. This represents a vector pointing from the end point to the beginning point. This represents the vector pointing from the starting point to the current node. This represents the result of the dot product of two vectors. This represents the length of the vector from the end point to the start point. This represents the length of the vector pointing from the starting point to the current node.

[0083] The forward and backward expansion processes are performed alternately. Before each expansion, it is checked whether there are identical nodes in the open lists of both directions; that is, a node in the forward open list and a node in the backward open list are the same node in the neighborhood graph. If identical nodes exist, it indicates that the expansion regions in the two directions overlap, and the expansion stops. Starting from the overlapping node, the path information recorded in the forward open list is used to backtrack to the starting point, and the path information recorded in the backward open list is used to backtrack to the ending point. The forward and backward paths are then concatenated to obtain the complete path from the starting point to the ending point. The Euclidean distances between adjacent nodes on the path are accumulated to obtain the geodesic distance between the starting point and the ending point. The accumulated Euclidean distance here refers to the Euclidean distance between adjacent nodes in the original eight-dimensional space, rather than the edge weights in the neighborhood graph, to ensure that the distance metric is consistent with the physical meaning of the original feature space.

[0084] The geodesic distance is used as the health deviation value at the current moment. The entire process from obtaining the current coordinates to calculating the geodesic distance is repeated for all moments to be evaluated, resulting in a series of health deviation values ​​arranged by time. This sequence of values ​​is the health deviation sequence.

[0085] A larger cosine value indicates a smaller directional deviation, and a higher expansion priority for that node. Starting from the selected node, traverse all neighboring nodes in the neighborhood graph, calculate the cumulative path length from the starting point through the current node to a neighboring node. If this length is less than the existing record of the neighboring node in the open list, update the path length and path information of the neighboring node, and add the neighboring node to the forward open list. After expanding the current node, remove it from the forward open list. The backward search expansion process is symmetrical to the forward search. Select the node with the highest priority from the backward open list for expansion, calculate the cosine value of the angle between the reverse path formed from the endpoint through this node and the straight line direction from the endpoint to the starting point, determine the priority based on the cosine value, and update the node information in the backward open list after expansion.

[0086] In the remaining lifetime particle filter prediction module, the health deviation sequence is input into the particle filter algorithm. Particle weights are updated through sequential importance sampling to estimate the evolution trend of the health deviation. When the predicted deviation exceeds a preset threshold, a probability estimate of the remaining usable lifetime is output, specifically including:

[0087] First, extract the latest value from the health deviation sequence and use it as the initial state. Set the total number of particles to 500. Centering on the initial state, generate 500 initial particles in a normal distribution around it, each representing a possible current true health deviation value. Assign each particle the same initial weight, 1 / 500, ensuring all particles have equal importance at the initial moment.

[0088] Particle state recursion and weight updates are performed based on the sequential importance sampling principle. A state transition equation is pre-constructed, describing the evolution of health deviation over time. Specifically, it adopts a first-order autoregressive form, where the predicted value for the next time step equals the current value multiplied by the autoregressive coefficient, plus a random disturbance term with a mean of 0 and a preset process noise variance. The autoregressive coefficient is set to 0.98, and the process noise variance is set to 0.01 based on the fluctuation range of historical data. For each particle, its current value is substituted into the state transition equation to predict one time step forward, obtaining the predicted health deviation value for that particle. The actual observed health deviation value at the current time step is obtained, and the absolute difference between the predicted value and the actual observed value for each particle is calculated. The weight update factor for the particle is determined based on the magnitude of this difference; the difference is inversely proportional to the weight update factor, with a larger difference resulting in a larger weight update factor. The original weight of each particle is multiplied by the corresponding weight update factor to obtain the updated weight. The updated weights of all particles are summed, and then the weight of each particle is divided by the sum to complete the weight normalization process.

[0089] It should be noted that in the remaining lifespan prediction process, the pre-constructed state transition equation adopts a first-order autoregressive function to describe the evolution trend of health deviation over time. The input parameter of this function is the health deviation value at the previous time step, and the output is the predicted health deviation value at the current time step. The function contains two key parameters: the first parameter is the first-order autoregressive coefficient, with a value of 0.98, used to control the correlation strength between adjacent time steps of health deviation. This coefficient is determined through autocorrelation analysis of the historical health deviation sequence. The second parameter is the process noise, which follows a normal distribution with a mean of zero and a variance of 0.01, used to characterize the random uncertainty in the system evolution process. Its variance is calibrated based on the fluctuation range of the historical health deviation difference sequence and combined with accelerated aging test data. By multiplying the health deviation value at the previous time step by the autoregressive coefficient and adding the process noise following the above normal distribution, the predicted health deviation value at the current time step can be obtained.

[0090] The system monitors the distribution of weights for all particles and calculates the current number of effective particles. The number of effective particles is equal to the reciprocal of the sum of the squares of the weights of all particles. When the number of effective particles is less than a preset threshold of 250, it is determined that the weight differences among particles are too large, triggering a resampling operation. The resampling process uses a system resampling method. First, a random number between 0 and 1 / 500 is generated as the starting point. Then, 500 selection pointers are generated at intervals of 1 / 500. Based on the pointer position, the corresponding particle is selected on the accumulated weight distribution for replication. Particles with high weights are replicated multiple times, while particles with low weights are eliminated. Finally, a new set of 500 particles is generated, and all particles in the new set are reassigned the same weight of 1 / 500.

[0091] The resampled particle set is continuously recursively evolved forward. A failure threshold for health deviation is set, which is 0.85 based on the cable's technical specifications at the time of manufacture and accelerated aging test data. For each particle, its current value is substituted into the state transition equation, and the health deviation value at each future time step by step is predicted forward until the predicted value of the particle first exceeds 0.85. The number of prediction steps that the particle takes from the current time to exceeding the threshold is recorded as the sampled value of the particle's remaining usable lifetime. The above forward recursive process is repeated for all 500 particles to obtain 500 sampled values ​​of remaining usable lifetime. The distribution of these 500 sampled values ​​is statistically analyzed, and the number of particles appearing in different lifetime intervals is calculated. The number of particles in each interval is divided by 500 to obtain the probability estimate corresponding to that interval. Finally, the probability distribution of remaining usable lifetime is output, which describes the probability of cable failure at different points in the future.

[0092] The working principle of this invention is as follows: By applying a step current excitation to the cable under test and simultaneously acquiring multiple response signals from the power line and communication line, the aliased signals are decomposed into independent components of power supply response characterizing the electrical characteristics of the conductor and independent components of communication disturbance characterizing the degree of disturbance to the communication link using blind source separation technology. Then, the phase space of the two independent components is reconstructed and recursive quantization indicators are extracted to construct a high-dimensional health feature set that can characterize the dynamic characteristics of the cable. The high-dimensional feature set is projected onto a low-dimensional intrinsic manifold through manifold learning, and the geodesic distance between the current moment and the health baseline state is calculated to generate a health deviation sequence reflecting the performance degradation of the cable. Finally, the health deviation sequence is input into a particle filtering algorithm, and the evolution trend of the health deviation is estimated by sequential importance sampling recursively. Based on the distribution of the moment when the particle prediction value exceeds a preset threshold, the probability estimate of the remaining usable lifetime is output, thereby realizing a technological leap from traditional continuity and voltage drop detection to cable health status assessment and lifetime prediction.

[0093] The foregoing has provided a detailed description of one embodiment of the present invention, but this description is merely a preferred embodiment and should not be construed as limiting the scope of the invention. All equivalent variations and modifications made within the scope of the claims of this invention should still fall within the patent coverage of this invention.

Claims

1. A USB data cable harness continuity and current detection system, characterized in that, include: The excitation response acquisition module applies a preset step current excitation to the USB data cable under test, and simultaneously acquires the voltage response signal, current response signal and logic level fluctuation signal on the power line inside the cable during the excitation process, to obtain multiple raw timing signals. The signal blind source separation module performs blind source separation processing on multiple raw timing signals, decomposing them into independent components of power supply response characterizing the electrical characteristics of cable conductors and independent components of communication disturbance characterizing the degree of interference in the communication link. Specifically, the blind source separation processing on the multiple raw timing signals includes: Perform short-time Fourier transform on multiple raw time-series signals to obtain the time-spectrum matrix; Based on the distribution differences of energy peaks and valleys in the time-frequency matrix, a first masking matrix corresponding to the power supply response characteristics and a second masking matrix corresponding to the communication disturbance characteristics are constructed in the time-frequency domain. The first and second masking matrices are multiplied by the time-spectrum matrix respectively to obtain the separated power supply response time-spectrum and communication disturbance time-spectrum. Inverse short-time Fourier transform is performed on the power supply response time spectrum and the communication disturbance time spectrum to reconstruct the independent components of the power supply response and the communication disturbance. The construction of a first masking matrix corresponding to the power supply response characteristics and a second masking matrix corresponding to the communication disturbance characteristics in the time-frequency domain specifically includes: Search the energy peak position of each frequency point along the time axis of the time spectrum matrix to obtain the energy accumulation trajectory of the power supply response at each frequency point; Search along the frequency axis of the time spectrum matrix to find the energy valley position at each time point, and obtain the energy depression distribution of communication disturbance at each time. Based on the energy accumulation trajectory and energy depression distribution, a first binary template adapted to the time-frequency form of the power supply response and a second binary template adapted to the time-frequency form of the communication disturbance are generated on the time-frequency plane. The first binary template and the second binary template are convolved with a preset smoothing window function to obtain the first masking matrix and the second masking matrix with smooth transition bands. The nonlinear dynamics feature extraction module reconstructs the phase space of the independent components of power supply response and communication disturbance, obtains the phase point trajectory in the high-dimensional phase space, calculates the recursion graph of the phase point trajectory, and extracts recursive quantization indicators from the recursion graph, including recursion rate, determinism, laminar flow and average diagonal length. The recursive quantization indicators are combined into a health feature set characterizing the dynamic characteristics of the cable. The health degradation manifold mapping module treats the health feature set as points in a high-dimensional space, projects them onto a low-dimensional intrinsic manifold, obtains the low-dimensional embedded coordinates of the health features, and calculates the geodesic distance between the current coordinates and the health baseline coordinates by monitoring the movement trajectory of the low-dimensional embedded coordinates, thereby generating a health deviation sequence. The remaining lifetime particle filter prediction module inputs the health deviation sequence into the particle filter algorithm, updates the particle weights through sequential importance sampling, estimates the evolution trend of health deviation, and outputs a probability estimate of the remaining usable lifetime when the particle prediction deviation exceeds a preset threshold. Specifically, this includes: Using the latest value in the health deviation sequence as the initial state, an initial particle set containing a preset number of particles is generated, and each particle is assigned the same initial weight. Based on the sequential importance sampling principle in the particle filtering algorithm, each particle is substituted into the pre-constructed state transition equation to predict the health deviation value at the next moment, and the weight of each particle is updated with the actual health deviation observation value at the current moment. When the weight difference of each particle exceeds a preset threshold, the particle set is resampled to remove particles with too low weight and duplicate particles with too high weight, so as to keep the total number of particles constant. The resampled particle set is continuously recursively evolved forward, and the moment when the predicted health deviation of each particle first exceeds the preset threshold is counted. The distribution of the moment when each particle exceeds the threshold is used as the probability estimate of the remaining usable lifetime.

2. The USB data cable harness continuity and current detection system according to claim 1, characterized in that, The extraction process of the recursive quantification index is as follows: The Euclidean distance between each pair of phase points is calculated for the phase point trajectories of the independent components of the power supply response and the independent components of the communication disturbance, respectively, to obtain the recursive distance matrix; Based on the numerical distribution of each element in the recursive distance matrix, adaptive thresholds are used to determine the recursive judgment thresholds for the independent components of power supply response and communication disturbance, respectively. The positions of elements in the recursive distance matrix that are less than or equal to the corresponding recursive decision threshold are marked as recursive points, and the positions of elements that are greater than the corresponding recursive decision threshold are marked as non-recursive points, thus generating recursive graphs of independent components of power supply response and independent components of communication disturbance. The recursion graphs of independent components of power supply response and independent components of communication disturbance are statistically analyzed for recursion point density, diagonal structure distribution and vertical line structure distribution, respectively, to obtain recursion quantification indicators including recursion rate, determinism, laminar flow and average diagonal length.

3. The USB data cable harness continuity and current detection system according to claim 2, characterized in that, The process for determining the recursive decision threshold is as follows: Probability density estimation is performed on the recursive distance matrices of the independent components of power supply response and communication disturbance, respectively, to obtain the distribution curves of the recursive distance values ​​in the amplitude range; Search for local minima on the distribution curve and use the distance values ​​corresponding to the local minima as candidate thresholds for the independent components of power supply response and communication disturbance, respectively. Verify whether the candidate threshold satisfies the requirement that the recursion point coverage is within a preset confidence interval. If it is satisfied, the candidate threshold is determined as the final recursive decision threshold. If it is not satisfied, backtrack to the second smallest local minimum point on the distribution curve for iterative verification.

4. The USB data cable harness continuity and current detection system according to claim 1, characterized in that, The generation of the health deviation sequence specifically includes: Calculate the Euclidean distance between each high-dimensional point in the health feature set, determine a preset number of neighboring points for each high-dimensional point, and construct a neighborhood graph based on the connection relationships between the neighboring points. A shortest path search algorithm is used on the neighborhood graph to calculate the geodesic distance between all pairs of high-dimensional points and obtain the geodesic distance matrix. By performing a coordinate mapping that preserves the distance structure on the geodesic distance matrix, the low-dimensional embedded coordinates corresponding to each high-dimensional point are obtained. The algorithm inputs the low-dimensional embedded coordinates of the current moment and the pre-stored health baseline coordinates into the geodesic tracing algorithm. It searches for the shortest path between the two points along the neighborhood graph, accumulates the Euclidean distance between adjacent points on the path as the geodesic distance at the current moment, and outputs the health deviation sequence after traversing all moments.

5. A USB data cable harness continuity and current detection system according to claim 4, characterized in that, The search for the shortest path between two points along the neighborhood graph specifically includes: Starting with the health baseline coordinates as the search starting point and the current low-dimensional embedding coordinates as the search ending point, initialize the forward open list and the backward open list; Expand outward from the forward open list by selecting nodes connected to the starting point, and backward from the backward open list by selecting nodes connected to the ending point. Alternately perform the expansion operation until the expansion regions in the two directions overlap. During the expansion process, the expansion priority is dynamically adjusted based on the deviation angle between the node and the line connecting the starting point and the ending point, prioritizing the expansion of neighboring nodes with smaller directional deviations. When the extended regions in two directions overlap, backtrack from the overlapping node to the starting point and the ending point to obtain the complete path, and accumulate the Euclidean distance between adjacent nodes on the path as the geodesic distance at the current moment.

6. The USB data cable harness continuity and current detection system according to claim 1, characterized in that, The probability estimate of the remaining usable lifetime output specifically includes: Using the latest value in the health deviation sequence as the initial state, an initial particle set containing a preset number of particles is generated, and each particle is assigned the same initial weight. Based on the sequential importance sampling principle in the particle filtering algorithm, each particle is substituted into the pre-constructed state transition equation to predict the health deviation value at the next moment, and the weight of each particle is updated with the actual health deviation observation value at the current moment. When the weight difference of each particle exceeds a preset threshold, the particle set is resampled to remove particles with too low weight and duplicate particles with too high weight, so as to keep the total number of particles constant. The resampled particle set is continuously recursively evolved forward, and the moment when the predicted health deviation of each particle first exceeds the preset threshold is counted. The distribution of the moment when each particle exceeds the threshold is used as the probability estimate of the remaining usable lifetime.

Images