A method and system for automatically detecting electrical performance of a wiring harness

Abstract

The present application relates to the field of automation detection, and more particularly to a kind of harness electrical performance automation detection method and system.The method includes obtaining harness fluctuation signal and carrying out amplitude-frequency and phase-frequency correction, obtain original time domain data and carry out fast Fourier transform, obtain frequency component sequence and carry out cross interference calculation, obtain line interference information;Threshold comparison is carried out based on the line interference information and feature classification is carried out, obtain coupling strength label and combine with the time domain feature parameter obtained by feature extraction based on the original time domain data, obtain feature vector group, then sequence prediction is carried out, obtain future time domain data and integrate with the original time domain data, obtain signal evolution trajectory;Signal oscillation information is obtained based on the signal evolution trajectory, extract abnormal value according to the signal oscillation information, obtain key feature vector, carry out fault risk calculation according to the key feature vector, obtain risk prediction report.

Description

Technical Field

[0001] This invention relates to the field of automated testing, and in particular to an automated testing method and system for the electrical performance of wire harnesses. Background Technology

[0002] In the field of fault prediction and health management, with the increasing complexity of equipment wiring harness systems, how to effectively identify signal coupling interference between lines and improve the accuracy of electrical fault detection has become a core issue that needs to be addressed in the development of current wiring harness performance testing technology.

[0003] In a current technology, the electrical performance of wire harnesses is primarily assessed by measuring physical connection status or simple electrical parameters. This involves monitoring basic indicators such as the harness's continuity, resistance, or voltage. When a parameter exceeds a preset range, a fault is identified, thus evaluating the harness quality. However, the signal transmission environment in industrial settings is increasingly complex. Fixed parameter measurement methods are ill-suited to the actual coupling characteristics under different operating frequencies and harness structures, leading to higher false positive rates or increased risk of missed detections. Furthermore, traditional solutions are often affected by interference from line spacing, arrangement, and the external electromagnetic environment during signal transmission. High-frequency signals can generate complex coupling and interference effects between adjacent lines, making it difficult to pinpoint the location of interference sources and potentially causing inaccurate coupling strength identification due to the superposition of multiple signals.

[0004] In summary, existing technologies lack the ability to precisely quantify signal coupling between lines in complex environments and accurately separate interference characteristics, resulting in insufficient accuracy in locating electrical faults in wire harnesses. Summary of the Invention

[0005] This invention provides an automated testing method and system for the electrical performance of wire harnesses, which can accurately identify and quantify the signal coupling strength between lines in complex wire harness structures, solving the problem of insufficient accuracy in locating electrical faults in existing wire harness technologies.

[0006] In a first aspect, to solve the above-mentioned technical problems, the present invention provides an automated testing method for the electrical performance of wire harnesses, comprising: The wiring harness fluctuation signal is acquired, and after amplitude-frequency distortion correction is performed on the wiring harness fluctuation signal, phase-frequency delay compensation is performed to obtain the original time-domain data. A fast Fourier transform is performed on the original time-domain data to obtain a frequency component sequence; Cross-interference calculation is performed based on the frequency component sequence to obtain line interference information; Based on the line interference information, a threshold comparison is performed to obtain the comparison result. Based on the comparison result, the line interference information is classified according to a preset classification line to obtain a coupling strength label. Feature extraction is performed based on the original time-domain data to obtain time-domain feature parameters. The time-domain feature parameters are then combined with the coupling strength label to obtain a feature vector group. The feature vector group is sequence predicted based on a preset recurrent neural network to obtain future time domain data. The future time domain data is then integrated with the original time domain data to obtain the signal evolution trajectory. Short-time Fourier transform is performed based on the signal evolution trajectory to obtain signal frequency domain information. Second-order difference is performed based on the signal frequency domain information to obtain signal oscillation information. Outlier extraction is performed based on the signal oscillation information to obtain abnormal oscillation information. Based on the abnormal oscillation information, feature extraction is performed to obtain key feature vectors. Based on the key feature vectors, fault risk is calculated to obtain a risk prediction report.

[0007] Secondly, the present invention provides an automated testing system for the electrical performance of wire harnesses, comprising: The signal acquisition module is used to acquire the wire harness fluctuation signal, perform amplitude-frequency distortion correction on the wire harness fluctuation signal, and then perform phase-frequency delay compensation to obtain the original time-domain data. The time-frequency transformation module is used to perform a fast Fourier transform based on the original time-domain data to obtain a frequency component sequence; The interference analysis module is used to perform cross-interference calculations based on the frequency component sequence to obtain line interference information; The feature classification module is used to perform threshold comparison based on the line interference information to obtain the comparison result, and to perform feature classification on the line interference information based on the comparison result and a preset classification line to obtain a coupling strength label. The parameter extraction module is used to extract features based on the original time-domain data to obtain time-domain feature parameters, and to combine the time-domain feature parameters with the coupling strength label to obtain a feature vector group. The sequence prediction module is used to perform sequence prediction on the feature vector group based on a preset recurrent neural network to obtain future time domain data, and integrate the future time domain data with the original time domain data to obtain the signal evolution trajectory. The anomaly extraction module is used to perform a short-time Fourier transform based on the signal evolution trajectory to obtain the signal frequency domain information, perform a second-order difference based on the signal frequency domain information to obtain the signal oscillation information, and extract outliers based on the signal oscillation information to obtain abnormal oscillation information. The risk prediction module is used to extract features based on the abnormal oscillation information to obtain key feature vectors, calculate fault risk based on the key feature vectors, and obtain a risk prediction report.

[0008] Compared with the prior art, the present invention has the following beneficial effects: (1) The present invention adopts a multi-channel signal blind source separation and adaptive frequency response compensation method to achieve accurate separation and correction of coupling interference components in wire harness signals under complex electromagnetic environments. This solves the problem that the existing technology using fixed parameter measurement methods is difficult to adapt to the actual coupling characteristics under different operating frequencies and wire harness structures, leading to increased misjudgment rate or increased risk of missed detection.

[0009] (2) The present invention uses fast Fourier transform to separate frequency domain amplitude spectrum lines and phase information, thereby realizing accurate quantification of signal coupling strength between lines and deep capture of interference source location, thus effectively solving the problem that the existing technology is affected by line spacing, arrangement and external electromagnetic environment interference during signal transmission, resulting in unclear judgment of interference source location.

[0010] (3) This invention simulates the long-term evolution trajectory of the wire harness signal by inputting the feature vector group into a recurrent neural network, and performs short-time Fourier transform and second-order difference calculation on the evolution trajectory to mark the periodic abnormal oscillation and abrupt inflection points. This solves the problem of inaccurate coupling strength identification and insufficient wire harness electrical fault location accuracy caused by the superposition effect of multiple signals and complex coupling in traditional schemes.

[0011] (4) This invention uses a support vector machine classifier to determine the coupling strength level between the principal value of the coherent function and the peak amplitude of the spectral density, and combines the long short-term memory unit of the recurrent neural network to capture the long-term dependence of the signal, thereby realizing intelligent classification of discrete coupled signals and effective identification of abnormal evolution patterns. Attached Figure Description

[0012] Figure 1 This is a schematic diagram of the automated testing method for the electrical performance of wire harnesses provided in an embodiment of the present invention; Figure 2 This is a schematic diagram of the structure of the automated testing system for the electrical performance of wire harnesses provided in an embodiment of the present invention. Detailed Implementation

[0013] 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.

[0014] Reference Figure 1 This invention provides an automated testing method for the electrical performance of wire harnesses, comprising: S11, acquire the wire harness fluctuation signal, perform amplitude-frequency distortion correction on the wire harness fluctuation signal, and then perform phase-frequency delay compensation to obtain the original time domain data. S12, perform a fast Fourier transform based on the original time-domain data to obtain a frequency component sequence; S13, perform cross-interference calculation based on the frequency component sequence to obtain line interference information; S14, perform threshold comparison based on the line interference information to obtain comparison results, and perform feature classification on the line interference information based on the comparison results and a preset classification line to obtain a coupling strength label; S15, Based on the original time-domain data, feature extraction is performed to obtain time-domain feature parameters. The time-domain feature parameters are combined with the coupling strength label to obtain a feature vector group. S16, perform sequence prediction on the feature vector group based on a preset recurrent neural network to obtain future time domain data, and integrate the future time domain data with the original time domain data to obtain the signal evolution trajectory; S17. Perform a short-time Fourier transform based on the signal evolution trajectory to obtain signal frequency domain information. Perform second-order difference based on the signal frequency domain information to obtain signal oscillation information. Extract outliers based on the signal oscillation information to obtain abnormal oscillation information. S18, Based on the abnormal oscillation information, feature extraction is performed to obtain key feature vectors, and fault risk is calculated based on the key feature vectors to obtain a risk prediction report.

[0015] In step S11, the harness fluctuation signal is acquired, and after amplitude-frequency distortion correction is performed on the harness fluctuation signal, phase-frequency delay compensation is then performed to obtain the original time-domain data, including: Acquire the wire harness fluctuation signal, and perform multi-channel signal blind source separation processing based on the wire harness fluctuation signal to obtain a pure digital signal; Each sampling point of the pure digital signal is appended with a corresponding timestamp, channel identifier, and phase information obtained through instantaneous phase analysis to construct a structured time-series dataset containing multi-dimensional information. The structured time-series dataset is then output as the original time-domain data.

[0016] It should be noted that the harness fluctuation signal is a synchronous discrete sampled numerical stream reflecting the current and voltage fluctuation status of multiple lines within the harness. It is acquired by a high-frequency analog-to-digital converter triggered by a global synchronization controller. Specifically, the sampling frequency is set to 1000kHz by default. The global synchronization controller sends a trigger pulse every 1 microsecond. Upon receiving the trigger pulse, the analog-to-digital converter of each channel synchronously acquires the analog voltage or current quantity on the corresponding line and converts it into a digital quantity to obtain the harness fluctuation signal.

[0017] It should be noted that multi-channel signal blind source separation processing recovers the source signal using a suppression matrix under the condition of a mixture of source and interference signals. The suppression matrix is ​​derived based on historical operational data and the physical layout characteristics of the lines. This suppression matrix is ​​used to decompose the observed mixed signal into independent source signal components through a linear transformation. For a harness system containing N lines, the suppression matrix is ​​an N x N matrix. Due to electromagnetic coupling effects, the observed signal is a linear combination of the source signals; this mixing relationship can be expressed using a mixing matrix, and the suppression matrix is ​​the inverse of the mixing matrix.

[0018] For example, in a simplified wiring harness system containing three lines, the actual source signals of the three lines are S1, S2, and S3, respectively. S1 and S2 are high-voltage power signals with operating voltages above 200 volts; S3 is a low-voltage control signal with operating voltages below 10 volts. Due to electromagnetic coupling between the lines, the actual observed mixed signals in each channel are X1, X2, and X3, respectively. A mixing matrix is ​​established through long-term observation and data analysis. The element in the first row and first column of this matrix is ​​0.95, indicating that the observed signal X1 mainly originates from the source signal S1, with S1 contributing 95% to X1; the element in the first row and second column is 0.03, indicating that the source signal S2 also contributes 3% to the observed signal X1; and the element in the first row and third column is 0.02, indicating that the source signal S3 contributes 2% to the observed signal X1. Further, an observed signal vector is constructed. The signal values ​​collected by each channel at the same time are arranged into a column vector. Subsequently, the suppression matrix is ​​multiplied by the observed signal vector to obtain the separated source signal vector.

[0019] The suppression matrix is ​​multiplied by the observed signal vector. Assuming the elements of the first row of the suppression matrix are 1.02, -0.01, and -0.05, the value of source signal S1 at that moment is 1.02 multiplied by 220 plus -0.01 multiplied by 215 plus -0.05 multiplied by 5.8, resulting in approximately 220.5 volts. Similarly, based on the elements of the second and third rows of the suppression matrix, the values ​​of source signals S2 and S3 at that moment are calculated to be 214.8 volts and 4.2 volts, respectively. This process is repeated for consecutive sampling moments, arranging the source signal vectors at all moments in chronological order to obtain the clean digital signals for each channel.

[0020] Furthermore, for each data point at each sampling time, not only is its amplitude value recorded, but also the precise timestamp, sampling channel number, and signal phase information corresponding to that data point are also recorded. The precise timestamp is in microseconds, starting at 0 from the beginning of sampling, with the timestamp of each subsequent data point increasing by 1 microsecond. The sampling channel number is used to identify which line the data point comes from. The phase information is obtained by analyzing the instantaneous phase of the signal, reflecting the phase offset of the signal at that moment relative to the reference moment.

[0021] The instantaneous phase is calculated as follows: For the pure digital signal sequence of each line, a Hilbert transform is first applied to obtain an analytic signal. The analytic signal consists of the original signal as the real part and the Hilbert transform result as the imaginary part. Then, the argument of the analytic signal is calculated for each sampling point, i.e., the arctangent of the imaginary and real parts is taken to obtain the instantaneous phase value at that point. The instantaneous phase is expressed in radians and ranges from -π to π. Alternatively, a continuous phase can be obtained using a phase expansion algorithm. For ease of subsequent processing, the instantaneous phase can be converted to degrees, and the accumulated phase can be modulo 360 to maintain it within the range of 0 to 360 degrees.

[0022] For example, for the pure digital signal sequence of line 1, at time 0 microseconds, the real part of the analytic signal obtained by Hilbert transform is 220.5 volts, and the imaginary part is 0 volts. The calculated phase angle is 0 degrees, therefore the initial phase angle is 0 degrees. At time 1 microseconds, the real part of the analytic signal is 220.8 volts, and the imaginary part is 0.3 volts. The calculated phase angle is arctan(0.3 / 220.8) ≈ 0.078 degrees, therefore the phase angle at this moment is 0.078 degrees. This process continues, calculating the instantaneous phase for each data point within the entire 1-second sampling time.

[0023] After calculating the phase information, the timestamp, channel number, amplitude value, and phase angle of each data point are organized into a data record. The formatted data is then integrated to obtain the original time-domain data.

[0024] In step S12, a fast Fourier transform is performed based on the original time-domain data to obtain a frequency component sequence, including: Windowing is performed on the original time-domain data to construct a sequence of window functions. Point-by-point multiplication is then performed on the window function sequence to obtain windowed time-domain data. A complex spectrum matrix is ​​obtained by performing a fast Fourier transform on the windowed time-domain data. The frequency component sequence is obtained by calculating the spectral lines and phase angles based on the complex spectral matrix.

[0025] First, the windowing process uses the Hanning window as the window function. The value of each element in the window function sequence is determined by its position index within the sequence. Let the total number of sampling points in the original time-domain data be N; then the Hanning window sequence also contains N elements. For the k-th element in the sequence, its corresponding window function value is calculated using a cosine function, with the specific value determined by the ratio of the position index k to the total number of points N. When k equals 0 or N minus 1 (i.e., elements at the two ends of the sequence), the window function value is close to zero; when k equals N minus 1 divided by 2 (i.e., elements in the middle of the sequence), the window function value reaches its maximum value of 1.

[0026] Point-by-point multiplication involves multiplying the amplitude value of each sample point in the original time-domain data sequence by the corresponding window function value to obtain the windowed amplitude value. The point-by-point multiplication is performed sequentially according to the sampling time order. For the k-th sample point, its windowed amplitude value is equal to the original amplitude value multiplied by the k-th window function value. The same windowing process is performed on the original time-domain data of all channels to obtain the windowed time-domain data for each channel.

[0027] For example, for the original sampling point at time 512 microseconds on line 1, its amplitude value is 225.6 volts. The corresponding window function value for this sampling point is approximately 0.9999, close to 1. Performing point-by-point multiplication, the windowed amplitude value is 225.6 volts multiplied by 0.9999, resulting in approximately 225.577 volts, almost identical to the original amplitude value. However, for the original sampling point at time 0 microseconds, its amplitude value is 220.5 volts, with a corresponding window function value of approximately 0.0001. Performing point-by-point multiplication, the windowed amplitude value is 220.5 volts multiplied by 0.0001, resulting in approximately 0.022 volts, much smaller than the original amplitude value. Similarly, for the original sampling point at time 1023 microseconds, its amplitude value is 221.8 volts, with a corresponding window function value of approximately 0.0001, resulting in a windowed amplitude value of approximately 0.022 volts.

[0028] Furthermore, for a windowed time-domain data sequence containing A sampling points, the output of the Fast Fourier Transform (FFT) is also a complex sequence containing A elements, each element corresponding to a complex spectral value at a specific frequency point. The complex spectral value consists of a real part and an imaginary part. The real part reflects the correlation between the frequency component and the reference cosine wave, while the imaginary part reflects the correlation between the frequency component and the reference sine wave. The combination of the real and imaginary parts can completely describe the amplitude and phase information of the frequency component. The amplitude information is obtained by calculating the square root of the sum of the squares of the real and imaginary parts, reflecting the energy intensity of the frequency component in the signal. The phase information is obtained by calculating the arctangent function of the imaginary and real parts, reflecting the phase shift of the frequency component relative to the reference time. The FFT is performed on the windowed time-domain data of all channels to obtain the complex spectral sequence corresponding to each channel. The complex spectral sequences of all channels are arranged in channel number order and organized into a matrix structure, which is the complex spectral matrix. The number of rows in the complex spectral matrix equals the total number of channels, and the number of columns equals the number of sampling points for each channel. The element in the i-th row and j-th column of the matrix represents the complex spectral value of the i-th channel at the j-th frequency point.

[0029] Further, the frequency component sequence is obtained by calculating the spectral lines and phase angles based on the complex spectral matrix. The spectral lines refer to the amplitude spectrum corresponding to each frequency point, which is obtained by calculating the modulus of the complex spectral values. For each element in the complex spectral matrix, its real and imaginary parts are extracted, the sum of the squares of the real and imaginary parts is calculated, and then the square root of the sum is taken to obtain the amplitude value at that frequency point. The phase angle is obtained by calculating the arctangent function of the imaginary and real parts, reflecting the phase characteristics of that frequency component.

[0030] For example, for the 10th element in the complex spectrum sequence of line 1, its real part is 3.2 and its imaginary part is 4.8. First, the square of the real part is calculated to be 10.24; the square of the imaginary part is calculated to be 23.04; the two are added together to get 33.28; the square root of 33.28 is taken to obtain an amplitude value of approximately 5.77. This means that at the 10th frequency point, corresponding to a frequency of approximately 9.77 kHz, the signal amplitude is approximately 5.77 volts. Further, the phase angle is calculated by dividing the imaginary part 4.8 by the real part 3.2, resulting in 1.5; the arctangent function of 1.5 is taken to obtain a phase angle of approximately 56.3 degrees. This means that this frequency component has a phase lead of approximately 56.3 degrees relative to the reference cosine wave.

[0031] Amplitude and phase calculations are performed on all elements of the complex spectrum matrix. The frequency value, amplitude value, and phase angle of each frequency point are organized into a record. All records are arranged in ascending order of frequency to obtain the frequency component sequence of that channel. Frequency component sequences are constructed for all channels to obtain a complete set of frequency component sequences.

[0032] In step S13, cross-interference calculation is performed based on the frequency component sequence to obtain line interference information, including: Based on the frequency component sequence, calculations are performed to obtain the cross power spectrum sequence and the coherence function value sequence; Data exceeding a preset coherence threshold is searched based on the coherence function value sequence to obtain the main coherence function value. Frequency indexing is then performed based on the main coherence function value to obtain the main crosstalk frequency point. Based on the cross-power spectrum sequence and the crosstalk main frequency point, data extraction is performed to obtain line interference information.

[0033] It should be noted that the cross-power spectrum sequence is used to describe the signal energy coupling relationship between two different lines at various frequency points, reflecting the degree of influence of the signal from one line on the other. For any two lines in the harness system, their complex spectrum sequences are extracted respectively. Let the complex spectrum sequence of the i-th line be Fi, and the complex spectrum sequence of the j-th line be Fj, with N sampling points for both. At each frequency point k, the complex spectrum value Fi(k) of the i-th line and the complex spectrum value Fj(k) of the j-th line at that frequency point are extracted. The conjugate complex number of Fi(k) is calculated and denoted as Fi. (k). Fi Multiplying Fj(k) by Fj(k) yields the cross-spectral value Cij(k) at that frequency point. Repeating the above operation for all frequency points yields the cross-spectral sequence.

[0034] Further, the magnitude of each element in the cross-spectral sequence is calculated, that is, the square root of the sum of the squares of the real and imaginary parts of each cross-spectral value is calculated, to obtain the cross-power spectrum sequence. Each element Pij(k) in the cross-power spectrum sequence represents the energy coupling strength between the i-th line and the j-th line at frequency k.

[0035] For example, in a wiring harness system containing four lines, it is necessary to analyze the coupling relationship between line 1 and line 2. At a frequency of 50 kHz, the real part of the complex spectral value F1(50) of line 1 is 8.5, and the imaginary part is 6.2; the real part of the complex spectral value F2(50) of line 2 is 3.1, and the imaginary part is 4.7. Calculate the conjugate complex F1 of F1(50). (50), the real part remains 8.5, and the imaginary part becomes -6.2. F1 Multiplying (50) by F2(50), according to the rules of complex number multiplication, the real part is 8.5×3.1-(-6.2)×4.7=55.49; the imaginary part is 8.5×4.7+(-6.2)×3.1=20.73. Therefore, the real part of the cross-frequency spectrum value C12(50) is 55.49, and the imaginary part is 20.73. Further calculation shows that the square of the real part is 3079.14, and the square of the imaginary part is 429.73. Adding the two together gives 3508.87, and the square root is approximately 59.24. This means that the cross-frequency power spectrum value of line 1 and line 2 at the 50 kHz frequency point is 59.24.

[0036] It should be noted that the coherence function value sequence is used to measure the signal correlation strength between two lines at various frequency points. For the i-th and j-th lines, at each frequency point k, the calculation of the coherence function value Gij(k) requires the simultaneous use of the cross power spectrum value, the self-power spectrum value of the i-th line, and the self-power spectrum value of the j-th line. The self-power spectrum refers to the spectral energy distribution of a single line itself, obtained by multiplying the complex spectral value of the line by its conjugate complex value and taking the modulus. Specifically, the self-power spectrum value Pii(k) of the i-th line at frequency point k is calculated, the self-power spectrum value Pjj(k) of the j-th line at frequency point k is calculated, the two are multiplied to obtain the product value, and then the square root operation is performed on the product value to obtain the normalization factor. The cross power spectrum value Pij(k) is divided by the normalization factor to obtain the coherence function value Gij(k). The coherence function calculation is repeated for all frequency points to obtain the coherence function value sequence.

[0037] For example, the self-power spectral density (P11(50)) of line 1 at 50 kHz is 125.8, and the self-power spectral density (P22(50)) of line 2 at 50 kHz is 48.3. Multiplying the two together gives 6076.14, and the square root is approximately 77.95. This value is the normalization factor. Dividing the cross-power spectral density (59.24) by the normalization factor (77.95) yields the coherence function value (G12(50)) of approximately 0.76.

[0038] Further, data exceeding a preset coherence threshold are searched based on the coherence function value sequence to obtain the principal value of the coherence function. The preset coherence threshold is a criterion determined based on engineering experience and statistical analysis of historical data, used to distinguish between normal independent signal transmission and abnormal states with significant coupling interference. The default coherence threshold is set to 0.4. The search process is performed sequentially according to frequency. For each element in the coherence function value sequence, its value is compared with the preset coherence threshold of 0.4. If the value of the element is greater than 0.4, the element is marked as the principal value of the coherence function, and its position index in the sequence is recorded. The position index directly corresponds to the frequency value represented by the element; a set of relevant frequency points is obtained by collecting all marked position indices.

[0039] It should be noted that the frequency index integration involves clustering and merging adjacent related frequency points. All collected location indices are sorted by numerical value, and the intervals between adjacent indices are checked. Based on historical engineering data, the common coupling interference bandwidth in harness systems typically does not exceed three times the frequency resolution, meaning the interference energy is mainly concentrated within one to two frequency points before and after the crosstalk center frequency. Therefore, the threshold for the difference between adjacent indices is set to 3. If the difference between two indices is less than or equal to 3, they are considered to belong to the same related frequency region and should be merged into a single crosstalk dominant frequency point. During merging, the arithmetic mean of all indices within that region is taken as the representative index. Coherence function principal value search and frequency index integration are performed on all line pair combinations to obtain the set of crosstalk dominant frequency points between each line pair.

[0040] Finally, data extraction is performed based on the cross-power spectrum sequence and the crosstalk dominant frequency point to obtain line interference information. For each identified crosstalk dominant frequency point, the corresponding cross-power spectrum sequence is traced back to extract the peak amplitude of the spectral density at that frequency point. The peak amplitude of the spectral density is the element value of the cross-power spectrum sequence at the crosstalk dominant frequency point index. The frequency value of the crosstalk dominant frequency point, the corresponding line pair number, and the peak amplitude of the spectral density are organized into an interference record, which fully describes a specific line-to-line coupling interference event. The above process is repeated for all crosstalk dominant frequencies of all line pairs to obtain line interference information.

[0041] For example, for the crosstalk dominant frequency point identified at 50.80 kHz between lines 1 and 2, it is necessary to extract the value of the cross-power spectrum sequence P12 at the corresponding position. This crosstalk dominant frequency point is obtained by integrating indices 51, 52, and 53, with index 52 representing the dominant frequency. The 52nd element of the cross-power spectrum sequence P12 is found, and its value is 61.37. The interference record is constructed as follows: line pair numbered 1 and 2, crosstalk dominant frequency point frequency is 50.80 kHz, and peak amplitude of spectral density is 61.37.

[0042] In step S14, a threshold comparison is performed based on the line interference information to obtain a comparison result. Based on the comparison result, the line interference information is classified according to a preset classification line to obtain a coupling strength label, including: If the line interference information exceeds the preset amplitude reference, then the line interference information is used to construct a vector to obtain a two-dimensional feature vector; the comparison result is the case where the line interference information exceeds the preset amplitude reference. The two-dimensional feature vector is mapped to a preset feature space to obtain vector feature data. The relative position of the vector feature data and the preset classification line is calculated to obtain vector position data. Based on the vector position data, the coupling strength level of the two-dimensional feature vector is determined, and the coupling strength level is labeled with the corresponding tag to obtain the coupling strength tag.

[0043] It should be noted that the preset amplitude benchmark is set based on the statistical characteristics of the background noise level under normal operating conditions of the wiring harness system, and is set to 20 by default. For each interference record in the line interference information set, the value of its peak spectral density amplitude field is extracted and compared with the preset amplitude benchmark 20. If the value is less than 20, the record is marked as background interference. If the value is greater than or equal to 20, the interference event is considered to require fine-grained classification, and vector construction continues.

[0044] For example, for the interference record between lines 1 and 2 at 50.80 kHz, its peak spectral density amplitude is 61.37, which is greater than the preset amplitude benchmark of 20, so vector construction is required. For another interference record between lines 3 and 4 at 120.5 kHz, assuming its peak spectral density amplitude is 15.6, which is less than the preset amplitude benchmark of 20, it is determined to be background interference and no further classification processing is performed.

[0045] It should be noted that the vector construction involves organizing the key feature parameters in the interference records into a two-dimensional feature vector. The peak amplitude of the spectral density of each interference record is used as the first element of the corresponding two-dimensional feature vector, and the principal value of the coherence function is used as the second element, forming a vector containing two values, which is the two-dimensional feature vector of the interference event. For example, for the interference record between lines 1 and 2 at 50.80 kHz, its peak amplitude of the spectral density is 61.37. This principal frequency point is obtained by integrating three frequency points with indices 51, 52, and 53, with corresponding coherence function values ​​of 0.76, 0.79, and 0.71, respectively. The average of these three values ​​is calculated to obtain 0.753. Therefore, the principal value of the coherence function for this interference event is 0.753. A two-dimensional feature vector is constructed, with the first element being 61.37 and the second element being 0.753, represented as the vector [61.37, 0.753].

[0046] Furthermore, the preset feature space is a two-dimensional planar coordinate system, where the horizontal axis represents the peak amplitude of the spectral density and the vertical axis represents the principal value of the coherence function. Each two-dimensional feature vector corresponds to a coordinate point in this coordinate system. For example, for the two-dimensional feature vector [61.37, 0.753], mapping it to the preset feature space yields a coordinate point with a horizontal coordinate of 61.37 and a vertical coordinate of 0.753.

[0047] There are two preset classification lines, which are classification decision boundaries pre-trained using machine learning methods. These lines divide the feature plane into three regions, each corresponding to a coupling strength level. The preset classification lines are trained using a support vector machine (SVM) algorithm, utilizing historically accumulated data from the wiring harness system, including interference information collected under different operating conditions and corresponding manually labeled coupling strength levels. After training, an optimal classification line is obtained, which can be used to accurately classify sample points of different levels in the feature plane. Since the peak amplitude of the spectral density and the principal value of the coherence function are not negative, only the classification of the first quadrant region in the feature plane needs to be considered. There are three coupling strength levels: Level 1, Level 2, and Level 3, with the interference level indicating the coupling strength.

[0048] Furthermore, the coordinates of the feature points are substituted into the expressions of the two classification lines to calculate their relative positions. The expressions for the two classification lines are -0.02x-y+0.6=0 and -0.01x-y+0.8=0, respectively. Subsequently, for any coordinate point (x0, y0) in the feature space, its relative position with the classification lines is calculated. For each classification line, the coordinates of the feature points are substituted into the line equation; if the result is greater than zero, the point is determined to be below the line; if the result is less than or equal to zero, the point is determined to be above the line. The calculation results of the coordinates of each feature vector relative to the two classification lines are then organized to obtain the vector position data. For example, in the vector feature data, there is a coordinate point (61.37, 0.753). For the first classification line -0.02x-y+0.6=0, substituting the point, the calculation result is -0.02×61.37-0.753+0.6≈-1.38, indicating that the point is located above the first classification line. For the second classification line -0.01x-y+0.8=0, substituting the point (61.37, 0.753), the calculation result is -0.01×61.37-0.753+0.8≈-0.56, indicating that the point is located above the second classification line.

[0049] Further, the coupling strength level of the two-dimensional feature vector is determined based on the vector position data. Two classification lines divide the feature plane into four regions, each corresponding to a different coupling strength level. If a feature point is simultaneously above both lines, it is marked as level three coupling; if a feature point is above the first line and below the second line, it is marked as level two coupling; if a feature point is simultaneously below both lines, it is marked as level one coupling; if a feature point is below the first line and above the second line, it is marked as level two coupling, or defined as another level according to actual physical meaning. Labeling the coupling strength levels with corresponding tags yields the coupling strength labels.

[0050] In step S15, feature extraction is performed based on the original time-domain data to obtain time-domain feature parameters. These time-domain feature parameters are then combined with the coupling strength label to obtain a feature vector group, including: The slope is calculated based on the original time-domain data to obtain the rising edge slope. Based on the original time-domain data, a Hilbert transform is performed to obtain an analytical signal. Based on the analytical signal, the difference between the extreme points of the envelope is extracted to obtain the envelope fluctuation depth. By combining the rising edge slope and the envelope fluctuation depth, temporal feature parameters are obtained. The temporal feature parameters are then combined with the coupling strength label to obtain a feature vector group.

[0051] First, the slope is calculated based on the original time-domain data to obtain the rising edge slope. Starting from the first sampling point of the time-domain sequence, differential calculations are performed sequentially to obtain the amplitude difference between each sampling point and its predecessor. To identify sampling points in the rising edge phase, a sliding window detection method is used. Specifically, starting from the fourth sampling point, the difference values ​​of the current point and its two preceding points (a total of three consecutive sampling points) are checked each time. When all three consecutive difference values ​​are positive and their sum exceeds 2 volts, the first sampling point of the current detection window is determined to be the starting point of the rising edge. After determining the starting point, detection continues to proceed. As long as the difference values ​​of subsequent sampling points remain positive, these points are continuously marked as points within the rising edge region. This continues until the first sampling point with a difference value of zero or negative is encountered, at which point the preceding sampling point is marked as the end point of the rising edge.

[0052] For example, for the original time-domain sequence of line 1, the sampling point amplitude at time 98 microseconds is 50 volts, at time 99 microseconds it is 51 volts, at time 100 microseconds it is 52.5 volts, and at time 101 microseconds it is 55 volts. The amplitude difference is then calculated: the amplitude difference at 99 microseconds is 51-50=1 volt, at 100 microseconds it is 52.5-51=1.5 volts, and at 101 microseconds it is 55-52.5=2.5 volts. When the detection window moves to 101 microseconds, it is found that the first three difference values ​​(1 volt, 1.5 volts, and 2.5 volts) are all positive and their sum is 5 volts, exceeding the 2-volt threshold. Therefore, time 99 microseconds is marked as the rising edge start point. Starting from 99 microseconds, the difference values ​​at subsequent times of 102 microseconds, 103 microseconds, and 104 microseconds are 3 volts, 5 volts, and 7 volts, respectively, all positive, until the amplitude reaches 120 volts at time 105 microseconds. The amplitude remains at 120 volts at time 106 microseconds, at which point the difference value is 0 volts and no longer positive. Therefore, time 106 microseconds is marked as the end of the rising edge, and the time series interval from time 98 microseconds to time 106 microseconds is marked as the rising edge. Further, a linear fit is performed on the time series marked as rising edges. The linear fit refers to using the least squares method to find a straight line that minimizes the overall deviation between this line and the data points in the original amplitude sequence. The slope of this line reflects the average rising rate trend throughout the entire rising edge process.

[0053] It should be noted that the Hilbert transform is a mathematical operation that converts real-valued signals into analytic signals in complex form. The real part of the analytic signal is the original time-domain signal itself, while the imaginary part is the result of the original signal after a specific phase transformation. By calculating the magnitude of the analytic signal, the instantaneous envelope of the original signal can be obtained. First, a Fast Fourier Transform is performed on the original time-domain data to obtain its frequency domain representation. Then, the frequency-domain data is processed, keeping all positive frequency components unchanged and setting all negative frequency components to zero. Subsequently, the amplitude of the positive frequency components is multiplied by 2. The processed frequency-domain data is then subjected to an inverse Fast Fourier Transform to convert it back to the time domain, and the resulting complex sequence is the analytic signal.

[0054] For example, the original time-domain sequence of line 3 is transformed by Hilbert to obtain an analytic signal sequence. At time 200 microseconds, the real part of the analytic signal is 110 volts and the imaginary part is -30 volts. The envelope value at this time is calculated as the square root of the sum of the squares of the real and imaginary parts, which equals 114.2 volts. At time 201 microseconds, the real part is 112 volts and the imaginary part is -28 volts, with an envelope value of approximately 115.5 volts. The envelope value is calculated for all sampling points throughout the entire time domain to obtain the complete envelope sequence.

[0055] The extreme points of the envelope refer to the locations of local maximum and local minimum values ​​in the envelope sequence. The method for detecting extreme points is as follows: for each sampling point in the envelope sequence, compare the envelope values ​​of that point with those of its five neighboring points. If the envelope value of that point is simultaneously greater than the envelope values ​​of the five sampling points before and after it, it is determined to be a local maximum point; if the envelope value of that point is simultaneously less than the envelope values ​​of the five sampling points before and after it, it is determined to be a local minimum point. The envelope fluctuation depth is quantified by calculating the amplitude difference between extreme points. Specifically, find the point with the largest envelope value from the set of local maximum points and record its envelope value; find the point with the smallest envelope value from the set of local minimum points and record its envelope value; calculate the difference between the largest and smallest envelope values ​​to obtain the envelope fluctuation depth. This parameter reflects the overall fluctuation degree of the signal envelope; a larger value indicates a deeper modulation depth. Finally, for each line, construct a time-domain feature parameter vector from its corresponding rising edge slope and envelope fluctuation depth, and collect these vectors together to obtain a feature vector group.

[0056] In step S16, the feature vector group is sequence predicted based on a preset recurrent neural network to obtain future time-domain data. The future time-domain data is then integrated with the original time-domain data to obtain the signal evolution trajectory, including: The feature vector group is input into a preset recurrent neural network for sequence prediction to obtain future time-domain data. The future time-domain data is concatenated with the original time-domain data to obtain a complete time series. Based on the complete time series, the amplitude values ​​and corresponding timestamps of each sampling point are integrated and aligned to obtain the signal evolution trajectory.

[0057] It should be noted that each line corresponds to a feature vector at each sampling time. These feature vectors are packaged into a data packet every 100 microseconds from the earliest to the latest sampling time, resulting in a feature vector sequence. For example, for line 1, the feature vector sequence between time 100 microseconds and time 200 microseconds contains three elements: rising edge slope, envelope ripple depth, and coupling strength label. At time 100 microseconds, the feature vector is [8.5 V / µs, 45.2 V, level 2]; at time 101 microseconds, the feature vector is [9.1 V / µs, 47.8 V, level 2]; and so on, until the feature vector at time 200 microseconds is [12.3 V / µs, 58.6 V, level 3]. These 100 feature vectors, arranged chronologically, constitute the data packet for this line between time 100 microseconds and time 200 microseconds.

[0058] It's worth noting that the pre-defined recurrent neural network uses a long short-term memory (LSTM) network structure, comprising a sequentially connected input layer, two LSTM hidden layers, and a fully connected output layer. The dimension of the input layer is consistent with the dimension of the feature vector at each time step, containing three features: rising edge slope, envelope ripple depth, and coupling strength label. The first LSTM hidden layer has 128 memory units, and the second LSTM hidden layer has 64 memory units, both using the tanh activation function, and a dropout mechanism is introduced to prevent overfitting, with a dropout ratio set to 0.2. The fully connected output layer has 10 neurons, corresponding to the predicted voltage amplitude at each sampling time within the next 10 microseconds. The output layer does not use an activation function and directly outputs the linear predicted value.

[0059] During the training phase, a training dataset is first constructed. Historical operating data of the harness system under various working conditions are collected, including raw time-domain data under normal and various fault conditions, and corresponding feature vector groups are extracted according to the aforementioned steps. The feature vector groups are divided into input sequences every 100 microseconds (i.e., 100 consecutive time steps), and the raw time-domain voltage amplitude 10 microseconds after the corresponding time window is used as the label sequence. For the rising edge slope and envelope ripple depth in the feature vector, Z-score normalization is used, and the mean and standard deviation are obtained statistically from all samples in the training set. For the coupling strength label, since it is a categorical discrete value, it is directly converted into a three-dimensional vector through one-hot encoding, and concatenated with the normalized continuous features to form the final input feature vector. For the label sequence, i.e., voltage amplitude, Z-score normalization is also used, and its mean and standard deviation are statistically independent of the continuous values ​​in the features.

[0060] During training, mean squared error was used as the loss function, Adam was selected as the optimizer, and the initial learning rate was set to 0.001. Every 10 epochs, if the validation set loss no longer decreased, the learning rate was reduced to 0.5. The maximum number of training epochs was set to 200, and an early stopping strategy was adopted: training stopped when the validation set loss did not decrease for 20 consecutive epochs. During training, the training and validation sets were randomly divided in an 8:2 ratio to monitor the model's generalization ability.

[0061] In the prediction phase, the feature vector sequence acquired and processed in real time is packaged into 100-microsecond increments and input into a trained recurrent neural network. The network output is a normalized sequence of voltage amplitude values ​​for the next 10 microseconds. This sequence needs to be denormalized, i.e., multiplied by the standard deviation of the training set labels and added to the mean, to restore the actual voltage amplitude physical quantity, thus obtaining the future time-domain data. This future time-domain data is then concatenated with the original time-domain data in chronological order to form a complete signal evolution trajectory.

[0062] In step S17, a short-time Fourier transform is performed based on the signal evolution trajectory to obtain the signal frequency domain information. Second-order difference is then performed based on the signal frequency domain information to obtain signal oscillation information. Outlier extraction is then performed based on the signal oscillation information to obtain abnormal oscillation information, including: A short-time Fourier transform is performed based on the signal evolution trajectory to obtain the main frequency amplitude distribution. Second-order difference calculation is then performed based on the main frequency amplitude distribution to obtain the signal oscillation information. Based on the signal oscillation information, inflection point analysis is performed to obtain candidate inflection point positions. Based on the candidate inflection point positions, the standard deviation and peak-to-peak value of the local window are extracted from the signal evolution trajectory to obtain abnormal oscillation information.

[0063] The Fourier transform employed is the Short-Time Fourier Transform. First, the long-duration signal is divided into multiple short-time windows. Frequency analysis is performed on each window separately, with a window length of 50 microseconds and a shift step of 30 microseconds between windows. The top five frequency components with the largest amplitudes for each window are extracted, and their frequency values ​​and corresponding amplitudes are recorded. Then, the component with the largest amplitude from these five components is selected as the dominant frequency component for that window, its frequency value recorded as the dominant frequency value, and its corresponding amplitude recorded as the dominant frequency amplitude. The dominant frequency components of all windows are arranged in chronological order to form a time-frequency distribution matrix, where each row represents a time window, and each column represents the dominant frequency value and dominant frequency amplitude for that window.

[0064] Second-order difference calculation is performed on the amplitude time series in the signal evolution trajectory, rather than directly on the time-frequency distribution matrix. First, an amplitude sequence ordered by time is extracted from the signal evolution trajectory; this sequence is a one-dimensional time series. A first-order difference operation is performed on this amplitude sequence, starting from the second sampling point, subtracting the amplitude value of the previous sampling point from the current sampling point's amplitude value to obtain the first-order difference sequence. Subsequently, a difference operation is performed again on this first-order difference sequence, starting from its second element, subtracting the previous element from the current element to obtain the second-order difference sequence. The second-order difference values ​​are used to characterize the changing trend of the signal amplitude rate. Simultaneously, the time-frequency distribution matrix obtained from the short-time Fourier transform is used to extract the dominant frequency value and corresponding dominant frequency amplitude for each time window. Finally, the center time, dominant frequency value, dominant frequency amplitude, and the second-order difference value corresponding to the sampling point closest to the center time for each time window are time-aligned and organized to form signal oscillation information. This second-order difference value can be obtained by finding the sampling points closest to and before the center time.

[0065] For example, in line 3, the signal amplitude at time 100 microseconds is 120 volts, at time 101 microseconds it is 123 volts, and at time 102 microseconds it is 128 volts. Therefore, the first-order difference value at time 101 microseconds is 123 - 120 = 3 volts. The first-order difference value at time 102 microseconds is 128 - 123 = 5 volts. Further calculation yields the second-order difference value at time 102 microseconds as 5 - 3 = 2 volts. This positive value indicates that the signal rise rate is accelerating.

[0066] It should be noted that the inflection point refers to the position where the trend of signal change reverses. In a second-order difference sequence, the inflection point corresponds to the position where the numerical sign changes. When the second-order difference changes from a positive value to a negative value, it indicates that the signal changes from accelerating upward to decelerating upward, or from decelerating downward to accelerating downward. When the second-order difference changes from a negative value to a positive value, it indicates that the signal changes from decelerating upward to accelerating upward, or from accelerating downward to decelerating downward.

[0067] Furthermore, the process begins by examining each sampling point from the starting position of the second-order difference sequence. For each sampling point, the sign of its second-order difference value is compared with that of the previous sampling point. When the signs are opposite, the previous sampling point is considered a candidate inflection point. Subsequently, valid inflection points are selected based on preset screening criteria. These criteria, set based on engineering experience, are used to filter out false inflection points caused by minor fluctuations. The preset screening criterion is that the absolute difference between the second-order difference value at the inflection point and the second-order difference value at the previous moment must be greater than 1 volt.

[0068] Furthermore, abnormal oscillation features are extracted near the inflection point. For each valid inflection point, 20 sampling points are traced back and 20 sampling points are extended forward, forming a 41-point analysis window centered on the inflection point. Within this window, the standard deviation and peak-to-peak value of the signal amplitude are calculated. Subsequently, a dynamic threshold mechanism is used to screen for abnormal inflection points. The dynamic threshold is adaptively adjusted based on the historical statistical characteristics of the current line signal. Specifically, firstly, all valid inflection point window data within the past 10,000 sampling points of the current line are extracted, and the mean standard deviation and mean peak-to-peak value of these historical windows are calculated as the baseline fluctuation level of the line signal. The dynamic threshold for standard deviation is set to 1.2 times the mean historical standard deviation. The dynamic threshold for peak-to-peak value is set to 1.5 times the mean historical peak-to-peak value. When the standard deviation of the inflection point window exceeds the dynamic threshold for standard deviation of the line, or the peak-to-peak value exceeds the dynamic threshold for peak-to-peak value of the line, the inflection point is determined to correspond to abnormal oscillation. Finally, the inflection point positions marked as abnormal, the corresponding standard deviations, peak-to-peak values, and the dominant frequency value and dominant frequency amplitude of the time window in which the inflection point is located are summarized to form an abnormal oscillation information set. If the inflection point falls within multiple time windows, such as in an overlapping region, then the dominant frequency information of the time window whose center time is closest to the inflection point is taken.

[0069] In step S18, feature extraction is performed based on the abnormal oscillation information to obtain key feature vectors. Fault risk is then calculated based on these key feature vectors to obtain a risk prediction report, including: Based on the abnormal oscillation information, spectrum analysis and time-domain waveform scanning are performed to obtain key feature vectors. The Euclidean distance between the key feature vectors and each record in the preset historical fault feature database is calculated to obtain the fault similarity value. Based on the historical fault feature database, corresponding weight coefficients are assigned to the fault similarity values. The weight coefficients and the fault similarity values ​​are weighted and summed to obtain the fault risk probability. Based on the fault risk probability, risk intervals are classified to obtain risk level data. Based on the risk level data, time-series monitoring is performed to obtain the duration of the anomaly; if the duration of the anomaly exceeds a preset duration threshold, regional spatial labeling is performed and the risk level is updated to obtain a risk prediction report.

[0070] It should be noted that the spectral analysis described herein is a process of extracting the frequency domain characteristics of each anomalous oscillation location. For each record in the anomalous oscillation information set, the time position of the anomalous oscillation is first located. Then, 100 sampling points are extended forward and backward from this time position in the signal evolution trajectory, forming an analysis window containing 201 sampling points. Subsequently, a Fast Fourier Transform is performed on the signal data within the analysis window to convert the time-domain signal to a frequency-domain representation. After the transformation, the frequency component distribution of the window is obtained, and the three frequency components with the largest amplitudes are extracted and their frequency values ​​and corresponding amplitude values ​​are recorded. Further, the arithmetic mean, standard deviation, peak-to-peak value, and skewness coefficient of the signal amplitude within the analysis window are calculated. The skewness coefficient measures the symmetry of the signal amplitude distribution relative to the mean; positive skewness indicates that the signal has a tail in the direction of higher voltage, and negative skewness indicates that the signal has a tail in the direction of lower voltage. These four time-domain statistical parameters, together with the aforementioned six frequency-domain characteristic parameters, constitute a key feature vector containing ten elements.

[0071] Furthermore, the historical fault feature database is a dataset of known fault cases accumulated over a long period of operation. Each record contains a ten-dimensional key feature vector and corresponding fault label information. The fault label information includes fault type encoding, fault severity level, and description of the fault's impact range. Subsequently, the key feature vector of the current abnormal oscillation and the feature vectors of all records in the historical fault feature database are Z-score standardized. Then, the Euclidean distance between the current anomaly and each record in the historical database is calculated to obtain the fault similarity value. The calculation process involves subtracting the corresponding elements of the two ten-dimensional vectors to obtain ten differences. Each difference is squared, the ten squared values ​​are summed, and finally, the square root of the sum is calculated. The result is the Euclidean distance between the two vectors.

[0072] For example, the key feature vector for the abnormal oscillation detected at 5280 microseconds on Line 2 is a ten-element vector A. The historical fault feature database contains a fault record with the number H087, whose key feature vector is a ten-element vector B. The first element of vector A is 48.5 kHz, and the first element of vector B is 49.2 kHz; the difference is -0.7 kHz, which squares to 0.49. The second element of vector A is 35.6 volts, and the second element of vector B is 34.1 volts; the difference is 1.5 volts, which squares to 2.25. The squares of the differences of all ten elements are calculated sequentially, and the sum is 125.8. Taking the square root of this sum gives 11.2. This value is the Euclidean distance between the current abnormal oscillation and the historical fault case H087.

[0073] Furthermore, a Gaussian kernel function is used for mapping. This Gaussian kernel function maps distance values ​​to similarity values ​​between 0 and 1; the smaller the distance, the closer the similarity is to 1, and the larger the distance, the closer the similarity is to 0. The calculation of the Gaussian kernel function involves a scaling parameter, which is determined based on the average distance between all feature vectors in the historical fault feature database. The Euclidean distance between any two records in the historical database is pre-calculated, and the median of these distance values ​​is calculated and used as the scaling parameter. The Euclidean distance calculated between the current abnormal oscillation and each historical record is substituted into the Gaussian kernel function for mapping, yielding the corresponding fault similarity value. This calculation is repeated for all records in the historical fault feature database to obtain a sequence of fault similarity values ​​between the current abnormal oscillation and each historical record.

[0074] For each record in the historical fault feature database, its weight is determined by two parts: the fault similarity value between the record and the current anomaly, and the severity level of the fault corresponding to the record. The higher the severity level of the fault, the greater the weight coefficient of the record in risk calculation. Specifically, the weight coefficient for level 1 faults is set to 0.3, the weight coefficient for level 2 faults is set to 0.6, and the weight coefficient for level 3 faults is set to 1.0.

[0075] For each historical record, its fault similarity value is multiplied by the corresponding severity weighting coefficient to obtain the risk contribution value of that record to the current anomaly. The risk contribution values ​​of all historical records are summed to obtain the total risk score. Then, the total risk score is normalized using minmax, and the number of historical records with a fault similarity value exceeding 0.5 with the current anomaly is counted and recorded as the number of valid reference records. The total risk score is divided by the number of valid reference records to obtain the fault risk probability. Further, risk interval classification is performed based on the fault risk probability to obtain risk level data. The risk interval classification adopts a three-level standard. When the fault risk probability is less than 0.4, it is judged as low risk, and the risk level is marked as Level 1, indicating that the possibility of the current abnormal oscillation causing an actual fault is small, and normal operation and monitoring can continue. When the fault risk probability is greater than or equal to 0.4 and less than 0.7, it is judged as medium risk, and the risk level is marked as Level 2, indicating that the current abnormal oscillation has a certain potential for fault, and it is recommended to check the relevant lines in the next maintenance cycle. When the probability of failure risk is greater than or equal to 0.7, it is judged as high risk and the risk level is marked as level three. This indicates that the current abnormal oscillation is highly similar to historical serious failures and technical personnel need to be arranged immediately for on-site inspection and handling.

[0076] Furthermore, time-series monitoring is performed based on the risk level data to obtain the duration of the anomaly. This time-series monitoring is the process of tracking and analyzing consecutive abnormal oscillation events on the same line. Once an abnormal oscillation is detected on a line and its risk level is calculated, a continuous monitoring mechanism for that line is immediately initiated. During subsequent sampling periods, the line is continuously monitored for any recurrence of abnormal oscillations. If, within the next 1000 sampling points (1000 microseconds), the line experiences another abnormal oscillation with a risk level equal to or higher than the previous one, these two abnormal oscillations are considered continuous manifestations of the same potential fault, and they are linked as an abnormal event sequence. The time of the first abnormal oscillation in this sequence is recorded as the start time, and the time of the last abnormal oscillation is recorded as the end time; the difference between the two is the duration of the anomaly. The duration of the anomaly reflects the stability and persistence of the potential fault over time. A longer duration indicates that the potential fault is not an occasional transient disturbance but a systemic problem with a persistent impact, and its risk level should be correspondingly increased.

[0077] Furthermore, if the duration of the anomaly exceeds a preset duration threshold, spatial labeling is performed and the risk level is updated. The preset duration threshold is determined based on the operating characteristics of the wiring harness system and historical fault data statistics, and is set to 1500 microseconds by default. When the duration of an abnormal event sequence on a certain line exceeds 1500 microseconds, it indicates that the line has a persistent potential fault, and its actual risk level is higher than the initially determined risk level. The risk level of the abnormal event sequence is automatically increased by one level. If the initial risk level is level one, it is increased to level two; if the initial risk level is level two, it is increased to level three; if the initial risk level is already level three, it remains at level three but an emergency indicator is added.

[0078] The spatial labeling of the region is the process of marking the spatial location of the line experiencing a persistent anomaly and its surrounding related lines. Pre-defined harness topology information stores the physical adjacency relationships and electrical connections between lines in the form of an adjacency list. This information is used to identify other lines that are physically adjacent to the abnormal line or have a direct electrical connection with it, and these lines are marked as potential risk diffusion areas. Finally, the location information of the abnormal oscillation, key feature vectors, fault similarity value distribution, risk level data, anomaly duration, regional spatial labeling, and suggested remedial measures are integrated to generate a structured risk prediction report.

[0079] In summary, this invention provides an automated method and system for detecting the electrical performance of wire harnesses, which can accurately identify and quantify the signal coupling strength between lines in complex wire harness structures, thus solving the problem of insufficient accuracy in locating electrical faults in existing wire harness technologies.

[0080] Reference Figure 2 This invention provides an automated testing system for the electrical performance of wire harnesses, characterized in that it includes: The signal acquisition module is used to acquire the wire harness fluctuation signal, perform amplitude-frequency distortion correction on the wire harness fluctuation signal, and then perform phase-frequency delay compensation to obtain the original time-domain data. The time-frequency transformation module is used to perform a fast Fourier transform based on the original time-domain data to obtain a frequency component sequence; The interference analysis module is used to perform cross-interference calculations based on the frequency component sequence to obtain line interference information; The feature classification module is used to perform threshold comparison based on the line interference information to obtain the comparison result, and to perform feature classification on the line interference information based on the comparison result and a preset classification line to obtain a coupling strength label. The parameter extraction module is used to extract features based on the original time-domain data to obtain time-domain feature parameters, and to combine the time-domain feature parameters with the coupling strength label to obtain a feature vector group. The sequence prediction module is used to perform sequence prediction on the feature vector group based on a preset recurrent neural network to obtain future time domain data, and integrate the future time domain data with the original time domain data to obtain the signal evolution trajectory. The anomaly extraction module is used to perform a short-time Fourier transform based on the signal evolution trajectory to obtain the signal frequency domain information, perform a second-order difference based on the signal frequency domain information to obtain the signal oscillation information, and extract outliers based on the signal oscillation information to obtain abnormal oscillation information. The risk prediction module is used to extract features based on the abnormal oscillation information to obtain key feature vectors, calculate fault risk based on the key feature vectors, and obtain a risk prediction report.

[0081] It should be noted that the automated testing system for electrical performance of wire harnesses provided in this embodiment of the invention is used to execute all the process steps of the automated testing method for electrical performance of wire harnesses in the above embodiment. The working principles and beneficial effects of the two are one-to-one, so they will not be described again.

[0082] This invention also provides an electronic device. The electronic device includes a processor, a memory, and a computer program stored in the memory and executable on the processor, such as an automated wire harness electrical performance testing program. When the processor executes the computer program, it implements the steps described in the various automated wire harness electrical performance testing method embodiments above, for example... Figure 1 The step S11 shown. Alternatively, when the processor executes the computer program, it implements the functions of each module / unit in the above-described device embodiments, such as the signal acquisition module.

[0083] For example, the computer program may be divided into one or more modules / units, which are stored in the memory and executed by the processor to complete the present invention. The one or more modules / units may be a series of computer program instruction segments capable of performing a specific function, which describe the execution process of the computer program in the electronic device.

[0084] The electronic device may be a desktop computer, laptop, handheld computer, or smart tablet, etc. The electronic device may include, but is not limited to, a processor and memory. Those skilled in the art will understand that the above components are merely examples of electronic devices and do not constitute a limitation on the electronic device. It may include more or fewer components than described above, or combine certain components, or different components. For example, the electronic device may also include input / output devices, network access devices, buses, etc.

[0085] The processor can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor can be a microprocessor or any conventional processor. The processor is the control center of the electronic device, connecting all parts of the electronic device via various interfaces and lines.

[0086] The memory can be used to store the computer programs and / or modules. The processor implements various functions of the electronic device by running or executing the computer programs and / or modules stored in the memory and by calling data stored in the memory. The memory may mainly include a program storage area and a data storage area. The program storage area may store the operating system, at least one application program required for a function (such as sound playback function, image playback function, etc.), etc.; the data storage area may store data created according to the use of the mobile phone (such as audio data, phonebook, etc.). In addition, the memory may include high-speed random access memory, and may also include non-volatile memory, such as hard disk, RAM, plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, at least one disk storage device, flash memory device, or other volatile solid-state storage device.

[0087] Wherein, if the modules / units integrated in the electronic device are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the methods of the above embodiments of the present invention can also be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include: any entity or device capable of carrying the computer program code, recording media, USB flash drives, portable hard drives, magnetic disks, optical disks, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media, etc. It should be noted that the content included in the computer-readable medium can be appropriately added or removed according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, computer-readable media do not include electrical carrier signals and telecommunication signals.

[0088] It should be noted that the device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Furthermore, in the accompanying drawings of the device embodiments provided by this invention, the connection relationships between modules indicate that they have communication connections, which can be specifically implemented as one or more communication buses or signal lines. Those skilled in the art can understand and implement this without any creative effort.

[0089] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above descriptions are merely specific embodiments of the present invention and are not intended to limit the scope of protection of the present invention. In particular, it should be noted that any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention for those skilled in the art.

Claims

1. An automated testing method for the electrical performance of wire harnesses, characterized in that, include: The wiring harness fluctuation signal is acquired, and after amplitude-frequency distortion correction is performed on the wiring harness fluctuation signal, phase-frequency delay compensation is performed to obtain the original time-domain data. A fast Fourier transform is performed on the original time-domain data to obtain a frequency component sequence; Cross-interference calculation is performed based on the frequency component sequence to obtain line interference information; Based on the line interference information, a threshold comparison is performed to obtain the comparison result. Based on the comparison result, the line interference information is classified according to a preset classification line to obtain a coupling strength label. Feature extraction is performed based on the original time-domain data to obtain time-domain feature parameters. The time-domain feature parameters are then combined with the coupling strength label to obtain a feature vector group. The feature vector group is sequence predicted based on a preset recurrent neural network to obtain future time domain data. The future time domain data is then integrated with the original time domain data to obtain the signal evolution trajectory. Short-time Fourier transform is performed based on the signal evolution trajectory to obtain signal frequency domain information. Second-order difference is performed based on the signal frequency domain information to obtain signal oscillation information. Outlier extraction is performed based on the signal oscillation information to obtain abnormal oscillation information. Based on the abnormal oscillation information, feature extraction is performed to obtain key feature vectors. Based on the key feature vectors, fault risk is calculated to obtain a risk prediction report.

2. The automated testing method for the electrical performance of wire harnesses according to claim 1, characterized in that, The process of acquiring the harness fluctuation signal, correcting the amplitude-frequency distortion of the signal, and then compensating for the phase-frequency delay to obtain the original time-domain data includes: Acquire the wire harness fluctuation signal, and perform multi-channel signal blind source separation processing based on the wire harness fluctuation signal to obtain a pure digital signal; A corresponding timestamp, channel identifier, and phase information obtained through instantaneous phase analysis are added to each sampling point of the pure digital signal to construct a structured time-series dataset containing multi-dimensional information, and the structured time-series dataset is output as the raw time-domain data.

3. The automated testing method for the electrical performance of wire harnesses according to claim 1, characterized in that, The step of performing a Fast Fourier Transform based on the original time-domain data to obtain a frequency component sequence includes: Windowing is performed on the original time-domain data to construct a sequence of window functions. Point-by-point multiplication is then performed on the window function sequence to obtain windowed time-domain data. A complex spectrum matrix is ​​obtained by performing a fast Fourier transform on the windowed time-domain data. The frequency component sequence is obtained by calculating the spectral lines and phase angles based on the complex spectral matrix.

4. The automated testing method for the electrical performance of wire harnesses according to claim 1, characterized in that, The calculation of cross-interference based on the frequency component sequence to obtain line interference information includes: Based on the frequency component sequence, calculations are performed to obtain the cross power spectrum sequence and the coherence function value sequence; Data exceeding a preset coherence threshold is searched based on the coherence function value sequence to obtain the main coherence function value. Frequency indexing is then performed based on the main coherence function value to obtain the main crosstalk frequency point. Based on the cross-power spectrum sequence and the crosstalk main frequency point, data extraction is performed to obtain line interference information.

5. The automated testing method for the electrical performance of wire harnesses according to claim 1, characterized in that, The process involves threshold comparison based on the line interference information to obtain a comparison result. Based on the comparison result, feature classification of the line interference information is performed using a preset classification line to obtain a coupling strength label, including: If the line interference information exceeds the preset amplitude reference, then the line interference information is used to construct a vector to obtain a two-dimensional feature vector; the comparison result is the case where the line interference information exceeds the preset amplitude reference. The two-dimensional feature vector is mapped to a preset feature space to obtain vector feature data. The relative position of the vector feature data and the preset classification line is calculated to obtain vector position data. Based on the vector position data, the coupling strength level of the two-dimensional feature vector is determined, and the coupling strength level is labeled with the corresponding tag to obtain the coupling strength tag.

6. The automated testing method for the electrical performance of wire harnesses according to claim 1, characterized in that, The step involves extracting features from the original time-domain data to obtain time-domain feature parameters. These time-domain feature parameters are then combined with the coupling strength label to obtain a feature vector group, including: The slope is calculated based on the original time-domain data to obtain the rising edge slope. Based on the original time-domain data, a Hilbert transform is performed to obtain an analytical signal. Based on the analytical signal, the difference between the extreme points of the envelope is extracted to obtain the envelope fluctuation depth. By combining the rising edge slope and the envelope fluctuation depth, temporal feature parameters are obtained. The temporal feature parameters are then combined with the coupling strength label to obtain a feature vector group.

7. The automated testing method for the electrical performance of wire harnesses according to claim 1, characterized in that, The step of performing sequence prediction on the feature vector group based on a preset recurrent neural network to obtain future time-domain data, and integrating the future time-domain data with the original time-domain data to obtain the signal evolution trajectory includes: The feature vector group is input into a preset recurrent neural network for sequence prediction to obtain future time-domain data. The future time-domain data is concatenated with the original time-domain data to obtain a complete time series. Based on the complete time series, the amplitude values ​​and corresponding timestamps of each sampling point are integrated and aligned to obtain the signal evolution trajectory.

8. The automated testing method for the electrical performance of wire harnesses according to claim 1, characterized in that, The process involves performing a short-time Fourier transform based on the signal evolution trajectory to obtain signal frequency domain information, performing a second-order difference based on the signal frequency domain information to obtain signal oscillation information, and extracting outliers based on the signal oscillation information to obtain abnormal oscillation information, including: A short-time Fourier transform is performed based on the signal evolution trajectory to obtain the main frequency amplitude distribution. Second-order difference calculation is then performed based on the main frequency amplitude distribution to obtain the signal oscillation information. Based on the signal oscillation information, inflection point analysis is performed to obtain candidate inflection point positions. Based on the candidate inflection point positions, the standard deviation and peak-to-peak value of the local window are extracted from the signal evolution trajectory to obtain abnormal oscillation information.

9. The automated testing method for the electrical performance of wire harnesses according to claim 1, characterized in that, The process of extracting features based on the abnormal oscillation information to obtain key feature vectors, and calculating fault risk based on the key feature vectors to obtain a risk prediction report includes: Based on the abnormal oscillation information, spectrum analysis and time-domain waveform scanning are performed to obtain key feature vectors. The Euclidean distance between the key feature vectors and each record in the preset historical fault feature database is calculated to obtain the fault similarity value. Based on the historical fault feature database, corresponding weight coefficients are assigned to the fault similarity values. The weight coefficients and the fault similarity values ​​are weighted and summed to obtain the fault risk probability. Based on the fault risk probability, risk intervals are classified to obtain risk level data. Based on the risk level data, time-series monitoring is performed to obtain the duration of the anomaly; if the duration of the anomaly exceeds a preset duration threshold, regional spatial labeling is performed and the risk level is updated to obtain a risk prediction report.

10. An automated testing system for the electrical performance of wire harnesses, characterized in that, include: The signal acquisition module is used to acquire the wire harness fluctuation signal, perform amplitude-frequency distortion correction on the wire harness fluctuation signal, and then perform phase-frequency delay compensation to obtain the original time-domain data. The time-frequency transformation module is used to perform a fast Fourier transform based on the original time-domain data to obtain a frequency component sequence; The interference analysis module is used to perform cross-interference calculations based on the frequency component sequence to obtain line interference information; The feature classification module is used to perform threshold comparison based on the line interference information to obtain the comparison result, and to perform feature classification on the line interference information based on the comparison result and a preset classification line to obtain a coupling strength label. The parameter extraction module is used to extract features based on the original time-domain data to obtain time-domain feature parameters, and combine the time-domain feature parameters with the coupling strength label to obtain a feature vector group; The sequence prediction module is used to perform sequence prediction on the feature vector group based on a preset recurrent neural network to obtain future time domain data, and integrate the future time domain data with the original time domain data to obtain the signal evolution trajectory. The anomaly extraction module is used to perform a short-time Fourier transform based on the signal evolution trajectory to obtain signal frequency domain information, perform second-order difference based on the signal frequency domain information to obtain signal oscillation information, and extract outliers based on the signal oscillation information to obtain abnormal oscillation information. The risk prediction module is used to extract features based on the abnormal oscillation information to obtain key feature vectors, calculate fault risk based on the key feature vectors, and obtain a risk prediction report.

Images