Wire harness test data processing method and system based on artificial intelligence

Abstract

The present application relates to the technical field of data processing, in particular to a harness test data processing method and system based on artificial intelligence. The method comprises: obtaining test signals of a harness in a set time period under multiple different test times, and preprocessing to obtain preprocessed test signals; extracting feature vectors of the test signals; the feature vectors include multiple dimension features; the same dimension feature under different test times constitutes a feature sequence of the corresponding dimension feature; all features in each feature vector are fused to obtain a corresponding dynamic health index; a sequence of the dynamic health index of the harness is input into a network prediction model to obtain the remaining life of the harness under the current test time, realizing test evaluation of the harness. That is, the scheme of the present application can accurately test the life of the harness.

Description

Technical Field

[0001] This invention relates to the field of data processing technology. More specifically, this invention relates to a method and system for processing wire harness test data based on artificial intelligence. Background Technology

[0002] A wire harness is a complex assembly composed of multiple wires, connectors, terminals, and insulation materials, used to transmit power or signals. It is widely used in the automotive, aerospace, home appliance, and industrial equipment industries. As the neural network of an electrical system, the quality of the wire harness directly affects the safety and reliability of the entire device.

[0003] Traditional wire harness testing mainly includes continuity testing, insulation testing, withstand voltage testing, and component testing. Testing equipment collects basic electrical parameters of the wire harness under test, such as resistance, voltage, and capacitance, and compares them with preset thresholds to obtain a binary judgment of pass or fail. This method can only determine whether a wire harness has a fault, but it cannot accurately diagnose the specific type and location of the fault. When critical or complex faults occur, the diagnostic results are ambiguous, heavily relying on experienced test engineers for manual troubleshooting, which is inefficient.

[0004] With the development of artificial intelligence, especially machine vision and deep learning, the efficiency and accuracy of wire harness quality inspection have been significantly improved in recent years.

[0005] In related technologies, such as the Chinese patent document CN117471227B entitled "Automotive Wiring Harness Parameter Performance Testing Method and Testing System," it discloses a method for measuring electrical parameters and testing signal transmission of automotive wiring harnesses by combining different environmental parameters. This method studies the influence weight of environmental parameters on electrical parameters. Furthermore, through feature weighting and vector transformation, various test features are integrated into a more comprehensive electrical measurement feature vector and signal transmission feature. By inputting the feature vector into a preset automotive wiring harness parameter performance analysis model, it is possible not only to comprehensively analyze wiring harness performance but also to predict wiring harness faults, thereby improving the accuracy of automotive wiring harness parameter performance testing.

[0006] The above method only considers the impact of different environmental parameters on the electrical performance data of the wiring harness. However, it does not delve into the impact of changes in the electrical performance data itself, making it difficult to truly and accurately test the wiring harness. Summary of the Invention

[0007] The purpose of this invention is to propose a method and system for processing wire harness test data based on artificial intelligence, in order to solve the problem of inaccuracy in existing wire harness testing technologies; to this end, this invention provides solutions in the following two aspects.

[0008] In a first aspect, the present invention provides an artificial intelligence-based wire harness test data processing method, comprising:

[0009] The test signals of the wire harness within a set time period under multiple different test numbers are acquired and preprocessed to obtain the preprocessed test signals;

[0010] Extract the feature vector of the test signal; the feature vector includes features of multiple dimensions; features of the same dimension under different test numbers constitute the feature sequence of the corresponding dimension feature.

[0011] By fusing all features from each feature vector, the corresponding dynamic health index is obtained.

[0012] The sequence of dynamic health index of the wire harness is input into the network prediction model to obtain the remaining life of the wire harness under the current number of tests, thereby realizing the test evaluation of the wire harness.

[0013] The weights of each feature during fusion are normalized feature sensitivities; the feature sensitivities are positively correlated with the absolute value of the rank correlation coefficient and the coefficient of variation of the local segment containing any dimension feature in the current test; the local segment is a sequence segment extracted from the feature sequence, starting from the feature corresponding to the current test and containing multiple test times.

[0014] The above-described solution involves testing the wire harness to obtain test signals, analyzing these signals to identify their features, and assigning weights to each feature for feature fusion. This fusion is then used for subsequent analysis of the wire harness's lifespan. In other words, the solution of this invention can accurately analyze wire harness data and accurately predict its lifespan.

[0015] Optionally, the feature sensitivity is:

[0016] ;

[0017] in, Features at the t-th test Feature sensitivity, Features at the t-th test The absolute value of the rank correlation coefficient of the local segment. Features at the t-th test The coefficient of variation of the local segment.

[0018] The aforementioned sensitivity characteristics can characterize the trend changes of the test signal during the testing process of the wire harness.

[0019] Optionally, the features of the multiple dimensions include the mean, variance, peak factor, waveform factor, frequency standard deviation, centroid frequency, frequency band energy ratio, and wavelet packet energy entropy of the test signal.

[0020] The multi-domain characteristics of the test signals obtained above provide data support for subsequent harness life prediction.

[0021] Optionally, it also includes: selecting key features from all features of the feature vector to obtain the final feature vector, specifically:

[0022] Obtain the historical feature vectors and historical lifetimes from all historical test counts of the harness, and obtain the historical feature time series of each historical feature according to the test time sequence;

[0023] Calculate the correlation between any historical feature time series and the corresponding historical lifetime series;

[0024] Sort the features by relevance from largest to smallest, and select the top m features as key features, where m is greater than 1.

[0025] By selecting key features, we can identify those that have a significant impact on the wiring harness, thereby improving the accuracy of subsequent wiring harness testing.

[0026] Optionally, it also includes a step of correcting the feature sensitivity, specifically:

[0027] Obtain the trend guidance factor of the local segment, and use the product of the trend guidance factor and the feature sensitivity as the final feature sensitivity;

[0028] The trend guiding factor is the ratio of the square of the difference between the values ​​of the first and last features of a local segment to the sum of the squares of the differences between the values ​​of all adjacent features within the local segment.

[0029] Optionally, the Softmax function is used to normalize the final feature sensitivity.

[0030] Optionally, the network prediction model is an LSTM model.

[0031] Optionally, before fusing all features in each feature vector, a step of normalizing the feature sequences of each dimension is also included.

[0032] Optionally, the preprocessing includes: denoising the test signal using wavelet threshold denoising and standardization.

[0033] The above preprocessing of the test signal can avoid errors caused by noise, amplitude differences under different test conditions, etc.

[0034] In the second aspect, the AI-based wire harness test data processing system includes:

[0035] processor;

[0036] The memory stores computer instructions for processing wire harness test data based on artificial intelligence, which, when executed by the processor, cause the system to perform the aforementioned artificial intelligence-based wire harness test data processing method.

[0037] The beneficial effects of this invention are as follows:

[0038] The solution of this invention can obtain the feature vector of the wire harness by performing in-depth analysis of the test signal of the wire harness, and then fuse the features in the feature vector to obtain a dynamic health index that can guarantee different test numbers, thereby achieving accurate prediction of the life of the wire harness. Attached Figure Description

[0039] Figure 1 The flowchart illustrating the steps of the AI-based wire harness test data processing method in this embodiment is shown in the schematic diagram.

[0040] Figure 2 The schematic diagram illustrates the structural block diagram of the AI-based wire harness test data processing system in this embodiment. Detailed Implementation

[0041] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention.

[0042] like Figure 1 As shown, the AI-based wire harness test data processing method in this embodiment includes the following steps:

[0043] Step S1: Obtain the test signals of the wire harness within a set time period under multiple different test counts, and perform preprocessing to obtain the preprocessed test signals.

[0044] In this embodiment, time-domain reflectometry is used to perform multiple tests on the wire harness under test. That is, a rapidly changing pulse signal is injected into the wire harness under test. The pulse signal propagates along the wire harness. When it encounters a discontinuity where the characteristic impedance changes (e.g., open circuit, short circuit, poor crimping, insulation layer damage, etc.), part of the signal energy will be reflected back to the injection port. At this time, the reflected signal at the injection port is collected to obtain the test signal within the set time period of the current test.

[0045] The above multiple tests are tests conducted at different times under the same set interval, such as testing the wire harness at the same time every day, or testing the wire harness at the same time after a set number of days.

[0046] In order to obtain high-resolution impedance characteristics, a step pulse signal with a fast rising edge is injected into the harness port as an excitation signal. The amplitude and shape of the excitation signal should be sufficient to generate a clearly measurable reflected signal in the harness, i.e., a test signal.

[0047] The sampling rate and sampling duration for acquiring the reflected signal are determined according to the following principles:

[0048] The sampling rate should be much higher than the equivalent bandwidth corresponding to the rising edge of the excitation pulse, i.e. ,in The sampling rate of the harness. The oversampling factor is greater than 2, and its value is determined based on the required measurement accuracy. This is the equivalent bandwidth during the rise time of the excitation pulse.

[0049] The sampling frequency mentioned above is used to accurately capture the rising edge characteristics of fast-reflected signals caused by small impedance discontinuities.

[0050] The acquisition duration should ensure that the entire round-trip propagation process of the test signal within the entire harness under test is fully recorded, i.e. ,in The acquisition time for the wire harness. The total length of the wire harness to be tested. It is the speed at which a signal propagates in the medium of the wire harness, which can be calculated from the material parameters of the wire harness or obtained through calibration.

[0051] To improve the accuracy and robustness of subsequent analysis, the acquired test signals are preprocessed in this embodiment, including denoising and normalization.

[0052] The acquired test signal is denoised by using wavelet thresholding to decompose the test signal into wavelets. Soft or hard thresholding functions are applied to the high-frequency detail coefficients after decomposition to suppress noise. Then, the denoised test signal is reconstructed through inverse wavelet transform.

[0053] Furthermore, the collected test signals are standardized by using maximum-minimum normalization to map the numerical range of the signal to a fixed interval, thus obtaining the final test signal and simplifying subsequent calculations.

[0054] Step S2: Extract the feature vector of the test signal; the feature vector includes features of multiple dimensions; the same dimension features under different test numbers constitute the feature sequence of the corresponding dimension features.

[0055] In order to comprehensively and deeply extract hidden information related to the health status of the harness from the acquired test signals, this embodiment performs multi-domain feature extraction on the preprocessed test signals to construct feature vectors.

[0056] The multi-domain features include the extracted time-domain features, frequency-domain features, and time-frequency-domain features of the test signal. Specifically, time-domain features are those that describe the overall statistical distribution and time series shape of the signal, including the mean, variance, peak factor, and waveform factor of the test signal. Frequency-domain features include obtaining the spectrum of the test signal by performing a Fast Fourier Transform, and extracting the frequency standard deviation, centroid frequency, and band energy ratio from the spectrum.

[0057] The time-frequency domain feature is the wavelet packet energy entropy, specifically using wavelet packet transform to process the test signal. Layer decomposition is performed to calculate the energy in each sub-band, and the energy is normalized to obtain the energy proportion of each band. Based on the definition of information entropy, the entropy value of the normalized energy distribution is calculated as the wavelet packet energy entropy.

[0058] The aforementioned time-frequency domain characteristics are primarily used to measure the complexity or uncertainty of signal energy distribution in the frequency domain, reflecting early signs of harness degradation. Theoretically, a healthy harness exhibits relatively uniform time-domain reflected signals, with energy concentrated in a few frequency bands, resulting in low wavelet packet energy entropy. When early degradation or minor defects occur in the harness, weak high-frequency or transient disturbances are introduced, causing signal energy to diffuse to more frequency bands, increasing the uncertainty of energy distribution and thus significantly raising the wavelet packet energy entropy. Therefore, wavelet packet energy entropy is a highly sensitive indicator for early harness failures.

[0059] All the features extracted above are combined into a feature vector.

[0060] Furthermore, each feature in the aforementioned feature vector may contain information that is not critical. Therefore, in order to obtain features that better reflect the harness, in this embodiment, critical features are selected from the feature vector to form a critical feature vector.

[0061] Specifically, the process of selecting key features from the feature vector is as follows:

[0062] Obtain the historical feature vectors and historical lifetimes from all historical tests of the harness, and obtain the historical feature time series of each historical feature according to the test time sequence;

[0063] Calculate the correlation between any historical feature time series and the corresponding historical lifetime series;

[0064] Sort the features by relevance from largest to smallest, and select the top m features as key features, where m is greater than 1.

[0065] Step S3: Fuse all features in the feature vectors of each test run to obtain the corresponding dynamic health index.

[0066] Considering that wire harness degradation is a complex nonlinear process, the sensitivity of different physical characteristics of the wire harness varies at different degradation stages. For example, capacitance changes may be most pronounced in the early stages of degradation, while contact resistance changes dominate in the later stages. Therefore, the corresponding test signals will inevitably differ during wire harness testing. Thus, this embodiment constructs a dynamic health index to accurately reflect the changes in the intrinsic degradation degree of the wire harness.

[0067] Specifically, the process of obtaining the dynamic health index is as follows:

[0068] First, obtain the feature sensitivity of each feature in the feature vector.

[0069] In this embodiment, the feature sensitivity is used to quantify the ability of each feature in the feature vector of the harness to characterize the current degradation stage.

[0070] The process of obtaining feature sensitivity is as follows:

[0071] First, obtain the absolute value of the rank correlation coefficient and the coefficient of variation of the local segment of the feature sequence; where the local segment is a sequence segment extracted from the feature sequence, starting from the feature corresponding to the current test and containing multiple test times.

[0072] In this embodiment, the rank correlation coefficient is the Spearman rank correlation coefficient, which is obtained by calculating the feature sequence and the number of tests corresponding to each feature in the feature sequence. Since the specific calculation process of the Spearman rank correlation coefficient is existing technology, it will not be described in detail here.

[0073] It should be noted that the time interval between two consecutive tests is the same, such as testing the wiring harness at the same time every day, or testing the wiring harness at the same time after a set number of days.

[0074] The number of features in the aforementioned local segment can be either half or two-thirds of the total number of feature sequences. In this embodiment, not all feature sequences are selected because data from historical test counts that are far removed from the current test are not very useful and may even cause interference.

[0075] Then, based on the absolute value of the rank correlation coefficient and the coefficient of variation, the feature sensitivity of the corresponding feature is obtained.

[0076] Specifically, feature sensitivity for:

[0077] ;

[0078] in, For feature sensitivity, Features at the t-th test The absolute value of the rank correlation coefficient of the local segment. Features at the t-th test The coefficient of variation of the local segment.

[0079] The coefficient of variation is the ratio of the standard deviation to the mean within a local segment.

[0080] in, The value measures the local trend of the corresponding feature, that is, it should show a clear monotonic trend during the degradation process. Therefore, when The closer the value is to 1, the better the local monotonicity; The larger the value, the more dispersed the feature is relative to its mean. In the early and late stages of degradation, the numerical change should be significant enough, that is, the data points should be dispersed to avoid passivation.

[0081] Furthermore, in order to more accurately obtain the sensitivity of each feature, a step of correcting the feature sensitivity is also included, specifically:

[0082] First, identify trend-guiding factors.

[0083] A high-quality trend inevitably involves high energy efficiency, which will be a key characteristic. The evolutionary process within a local segment is quantified as a trend-guiding factor. The smoothness of the trend is reflected in local segments.

[0084] Specifically: ;

[0085] in, and It is a feature The values ​​of the first and nth tests within the local segment; n is the total number of features within the local segment. It represents the effective work required to achieve the net displacement from the first value to the last value. This work is only related to the difference between the first and last values ​​and represents the overall change in the trend of the local segment. , Representing features respectively The first in the local segment The values ​​of the i-th test and the (i-1)th test. It represents the sum of the squares of the differences between features across all two adjacent test runs. It is a tiny constant to prevent the denominator from being zero, such as 0.001.

[0086] When features When it is a smooth linear degradation, the trajectory is a straight line. The value and The closer the trend is to the trend leader, the higher the trend leader, and the better the trend quality; when the feature... It is a noisy, steady signal that jitters violently around a horizontal line without any obvious trend. When the trend indicator approaches 0, the trend quality is poor; when the feature... It is a degradation trend with noise. It captured the energy of the overall decrease and the energy of noise fluctuations.

[0087] Therefore, the final feature sensitivity for:

[0088] .

[0089] The aforementioned final feature sensitivity ensures that only those features with clear direction, significant changes, and smooth processes can be used to construct the final dynamic health index.

[0090] Secondly, the weights of the corresponding features are determined based on feature sensitivity.

[0091] Specifically, the weights are the normalized feature sensitivities.

[0092] Normalization can be achieved using the Softmax function, i.e., the weights. for:

[0093] ;

[0094] in, Let J be the weight of the j-th feature in the t-th test. , These are the j-th features at the t-th test. The kth feature The final feature sensitivity, It is a hyperparameter used to adjust the weight allocation to enhance or weaken the dominant role of the most sensitive feature.

[0095] Then, the dynamic health index of each feature vector is calculated.

[0096] Specifically, first, each feature sequence is normalized, and then the weights are used to sum all the normalized features for the same number of tests.

[0097] The dynamic health index is: ;

[0098] in, This represents the dynamic health index at the t-th test. It is the j-th feature in the t-th test, and K is the total number of features fused.

[0099] It should be noted that before calculating the dynamic health index, the features of each extracted dimension (such as mean, variance, centroid frequency, etc.) are standardized or normalized (for example, using Z-score standardization or max-min normalization) to make them all dimensionless values, so as to avoid dimensional problems during fusion.

[0100] In this embodiment, each feature is assigned a different weight, which can automatically adjust the contribution of each feature according to different degradation stages. For example, in the early stage of insulation aging, a certain capacitance-related feature may be the most sensitive; while in the later stage, the resistance-related feature changes more drastically, and the dynamic health index can dynamically capture this change; when a feature is temporarily distorted due to noise or specific operating conditions, its sensitivity will decrease, and the corresponding weight will also decrease, thereby reducing the negative impact of the outlier on the overall health index and improving the robustness of the model.

[0101] Step S4: Input the sequence of dynamic health index of the harness into the network prediction model to output the remaining life of the harness, thereby realizing the test and evaluation of the harness.

[0102] In this embodiment, a network prediction model is constructed, which can be a Long Short-Term Memory (LSTM) network. The input of the model is a sequence of dynamic health indices under multiple test runs, and the output is the remaining lifespan of the harness.

[0103] It should be noted that for training the network prediction model, the training set is a collection of sequences of historical dynamic health indices for different number of historical tests of the harness, with labels consisting of sequences of remaining lifetime of the harness at each test number. Since this is existing technology, it will not be elaborated further here.

[0104] In this embodiment, the dynamic health index is input into the LSTM prediction model, and the remaining service life of the harness is output.

[0105] The present invention decomposes the complex task of harness condition assessment into two interrelated but technically different sub-tasks: fault classification and health assessment. The optimal deep learning model is configured for each sub-task to maximize the overall performance.

[0106] This invention also provides an artificial intelligence-based wire harness test data processing system. For example... Figure 2 As shown, the system includes a processor and a memory, the memory storing computer program instructions, which, when executed by the processor, implement the artificial intelligence-based wire harness test data processing method according to the present invention.

[0107] The system also includes other components well known to those skilled in the art, such as communication buses and communication interfaces, the settings and functions of which are known in the art and therefore will not be described in detail here.

[0108] In this invention, the aforementioned memory can be any tangible medium containing or storing a program that can be used or combined with an instruction execution system, apparatus, or device. For example, a computer-readable storage medium can be any suitable magnetic or magneto-optical storage medium, such as Resistive Random Access Memory (RRAM), Dynamic Random Access Memory (DRAM), Static Random Access Memory (SRAM), Enhanced Dynamic Random Access Memory (EDRAM), High-Bandwidth Memory (HBM), Hybrid Memory Cube (HMC), etc., or any other medium that can be used to store desired information and can be accessed by an application, module, or both. Any such computer storage medium can be part of a device or accessible to or connected to a device. Any application or module described in this invention can be implemented by computer-readable / executable instructions stored or otherwise maintained on such a computer-readable medium.

[0109] In the description of this specification, "multiple" means at least two, such as two, three or more, etc., unless otherwise expressly and specifically defined.

[0110] While various embodiments of the invention have been shown and described in this specification, it will be apparent to those skilled in the art that such embodiments are provided by way of example only. Many modifications, alterations, and alternatives will occur to those skilled in the art without departing from the spirit and essence of the invention.

Claims

1. A method for processing wire harness test data based on artificial intelligence, characterized in that, include: The test signals of the wire harness within a set time period under multiple different test numbers are acquired and preprocessed to obtain the preprocessed test signals; Extract the feature vector of the test signal; the feature vector includes features of multiple dimensions; features of the same dimension under different test numbers constitute the feature sequence of the corresponding dimension feature. By fusing all features from each feature vector, the corresponding dynamic health index is obtained. The sequence of dynamic health index of the wire harness is input into the network prediction model to obtain the remaining life of the wire harness under the current number of tests, thereby realizing the test evaluation of the wire harness. The weights of each feature during fusion are normalized feature sensitivities; the feature sensitivities are positively correlated with the absolute value of the rank correlation coefficient and the coefficient of variation of the local segment containing any dimension feature in the current test; the local segment is a sequence segment extracted from the feature sequence, starting from the feature corresponding to the current test and containing multiple test times.

2. The method for processing wire harness test data based on artificial intelligence according to claim 1, characterized in that, The feature sensitivity is: ； in, Features at the t-th test Feature sensitivity, Features at the t-th test The absolute value of the rank correlation coefficient of the local segment. Features at the t-th test The coefficient of variation of the local segment.

3. The method for processing wire harness test data based on artificial intelligence according to claim 2, characterized in that, The features in these multiple dimensions include the mean, variance, peak factor, waveform factor, frequency standard deviation, centroid frequency, frequency band energy ratio, and wavelet packet energy entropy of the test signal.

4. The method for processing wire harness test data based on artificial intelligence according to claim 1, characterized in that, This also includes: selecting key features from all features of the feature vector to obtain the final feature vector, specifically: Obtain the historical feature vectors and historical lifetimes from all historical test counts of the harness, and obtain the historical feature time series of each historical feature according to the test time sequence; Calculate the correlation between any historical feature time series and the corresponding historical lifetime series; Sort the features by relevance from largest to smallest, and select the top m features as key features, where m is greater than 1.

5. The method for processing wire harness test data based on artificial intelligence according to claim 1, characterized in that, It also includes a step of correcting the sensitivity of the features, specifically: Obtain the trend guidance factor of the local segment, and use the product of the trend guidance factor and the feature sensitivity as the final feature sensitivity; The trend guiding factor is the ratio of the square of the difference between the values ​​of the first and last features of a local segment to the sum of the squares of the differences between the values ​​of all adjacent features within the local segment.

6. The method for processing wire harness test data based on artificial intelligence according to claim 5, characterized in that, The final feature sensitivity is obtained by using the Softmax function when normalizing it.

7. The method for processing wire harness test data based on artificial intelligence according to claim 1, characterized in that, The network prediction model is an LSTM model.

8. The method for processing wire harness test data based on artificial intelligence according to claim 1, characterized in that, Before fusing all features in each feature vector, the process also includes a step of normalizing the feature sequences of each dimension.

9. The method for processing wire harness test data based on artificial intelligence according to claim 1, characterized in that, The preprocessing includes: denoising the test signal using wavelet threshold denoising and standardization.

10. A wire harness test data processing system based on artificial intelligence, characterized in that, include: processor; A memory storing computer instructions for processing wire harness test data based on artificial intelligence, which, when executed by the processor, cause the system to perform the wire harness test data processing method based on artificial intelligence according to any one of claims 1-9.

Images