Wire harness processing quality traceability management method based on big data

Abstract

The invention discloses a wire harness processing quality traceability management method based on big data, and relates to the technical field of wire harness processing management, and the method comprises the steps: selecting a principal component of which the cumulative contribution rate is greater than a contribution rate threshold after calculating a feature value and a feature vector of a covariance matrix, and projecting a decentralized data matrix to a principal component space to obtain feature data; and a plurality of batches of historical feature data are substituted into the gating circulation unit, the gating circulation unit performs time sequence prediction to predict the quality trend of the wire harness in the future, the SHAP algorithm is utilized to analyze the influence of each piece of feature data on quality reduction, and a quality management strategy is output based on the prediction result of the gating circulation unit and the analysis result of the SHAP algorithm. According to the management method, high-dimensional data are projected to a low-dimensional principal component space, the calculation complexity is reduced, time sequence prediction is carried out by adopting a gating circulation unit, possible quality degradation is warned in advance, and the generation efficiency of the automobile wire harness is guaranteed.

Description

Technical Field

[0001] The present invention relates to the technical field of wire harness processing management, and in particular to a method for tracing and managing the quality of wire harness processing based on big data. Background Art

[0002] Automobile wire harnesses are key components of the vehicle's electrical system, responsible for connecting various electrical devices such as electronic control units, sensors, and actuators in the vehicle to achieve signal transmission and power supply. With the upgrade of automotive technology, especially the development of new energy and intelligent connected vehicles, the structure of wire harnesses has become increasingly complex, with increased length, more branches, and a variety of wire types. The processing process involves multiple links such as wire cutting, terminal crimping, connector assembly, bundling, and detection. Any quality problem in any link may affect the safety and stability of the entire vehicle.

[0003] The existing technology has the following defects:

[0004] 1. Traditional quality inspection methods are usually based on fixed threshold rules or offline analysis, and can only detect abnormalities after quality problems occur, unable to predict in advance the quality problems that may occur in future batches. This may lead to the discovery of problems only after the product has entered the next process or even after leaving the factory, increasing the rework cost and thus affecting production efficiency;

[0005] 2. Since the large database of wire harnesses involves multiple quality parameters (such as insulation layer thickness, terminal crimping force, conductor resistance, etc.), these data have high-dimensionality, non-linear correlation, and redundant information. Traditional methods directly use the original data for analysis, resulting in a large amount of calculation and being easily interfered by noise, making it difficult to find the truly key influencing factors.

[0006] Based on this, the present invention proposes a method for tracing and managing the quality of wire harness processing based on big data. By projecting high-dimensional data into a low-dimensional principal component space, the calculation complexity is reduced, and a gated recurrent unit is used for time series prediction to early warn of possible quality deterioration and ensure the production efficiency of automotive wire harnesses. Summary of the Invention

[0007] The purpose of the present invention is to provide a method for tracing and managing the quality of wire harness processing based on big data to solve the deficiencies in the background art.

[0008] To achieve the above purpose, the present invention provides the following technical solution: A method for tracing and managing the quality of wire harness processing based on big data, the management method includes the following steps:

[0009] S1: Clean and process the original multi-source data and then construct an original data matrix;

[0010] S2: Generate a covariance matrix based on the original data matrix and calculate the eigenvalues and eigenvectors of the covariance matrix;

[0011] S3: Select the principal components with cumulative contribution rates greater than the contribution rate threshold, and project the de-centered data matrix into the principal component space to obtain feature data;

[0012] S4: The gated recurrent unit performs time series prediction, and the SHAP algorithm is used to analyze the impact of each feature data on the quality decline;

[0013] S5: Output the quality management strategy based on the prediction results of the gated recurrent unit and the analysis results of the SHAP algorithm.

[0014] In a preferred embodiment, generating a covariance matrix based on the original data matrix includes the following steps:

[0015] Construct the original data matrix X with the terminal crimping force, insulation layer thickness, conductor resistance, crimping die wear, equipment operation time, temperature, humidity, vibration amplitude, unqualified rate, and rework rate parameters after normalization processing, and calculate the mean value of each original multi-source data value in the original data matrix X;

[0016] Perform centering processing on the original data matrix X, that is, subtract the mean value of each original multi-source data value from the original multi-source data value to obtain the de-centered value of the original multi-source data,

[0017] According to the de-centered data matrix X ′ Calculate the covariance matrix C, and the covariance matrix C represents the covariance between different de-centered values.

[0018] In a preferred embodiment, calculating the eigenvalues and eigenvectors of the covariance matrix includes the following steps:

[0019] After obtaining the covariance matrix C, use the eigenvalue decomposition method in linear algebra to calculate the eigenvalues and eigenvectors of the covariance matrix C. The equation expression is: Cv = λv, where λ is the eigenvalue and v is the eigenvector;

[0020] Sort the eigenvalues of the covariance matrix in descending order to obtain the sorted eigenvalue list: λ1 ≥ λ2 ≥ … ≥ λ N , where N represents the number of eigenvalues in the covariance matrix, and λ i represents the i-th eigenvalue;

[0021] Sort the eigenvectors of the covariance matrix in descending order to obtain the eigenvector list: v1 ≥ v2 ≥ … ≥ v M , where M represents the number of eigenvectors in the covariance matrix, and v i represents the i-th eigenvector.

[0022] In a preferred embodiment, principal components with a cumulative contribution rate greater than the contribution rate threshold are selected, and the de-centered data matrix is projected onto the principal component space to obtain feature data, including the following steps:

[0023] Calculate the contribution rate of the eigenvalue by the ratio of the eigenvalue to the sum of all eigenvalues, and then accumulate the contribution rates of the eigenvalues to obtain the cumulative contribution rate of each principal component;

[0024] If the cumulative contribution rate of the principal component is greater than the contribution rate threshold, select the eigenvectors corresponding to the first k eigenvalues as the principal components;

[0025] The selected first k eigenvectors form an N×k principal component matrix P, P = [v1, v2, …, v k , where v i represents the i-th eigenvector, k is the number of selected principal components, and project the de-centered data matrix onto the principal component matrix to obtain feature data and obtain a low-dimensional feature data matrix.

[0026] In a preferred embodiment, a quality management strategy is output based on the prediction results of the gated recurrent unit and the analysis results of the SHAP algorithm, including the following steps:

[0027] Obtain the SHAP value of each feature data through the SHAP algorithm, and generate the weight coefficient of each feature data from the SHAP value of the feature data. The generation logic is: sum the SHAP values of all feature data to obtain the total SHAP value, and divide the SHAP value by the total SHAP value to obtain the weight coefficient of the feature data;

[0028] Weightedly calculate all feature data change values to obtain the quality defect coefficient of the future automotive wiring harness. The expression is: In the formula, qsd is the quality defect coefficient, B is the number of feature data change values, ω i is the weight coefficient of the i-th feature data change value, and TS i is the i-th feature data change value;

[0029] Compare the obtained quality defect coefficient with a preset first defect threshold and a second defect threshold. The second defect threshold is used to determine whether the quality of the automotive wiring harness meets the standard, and the first defect threshold is used to determine whether there are latent defects in the automotive wiring harness.

[0030] If the quality defect coefficient is greater than the second defect threshold, it is determined that the production quality of the future automotive wiring harness does not meet the standard;

[0031] If the quality defect coefficient is less than or equal to the second defect threshold and the quality defect coefficient is greater than the first defect threshold, it is determined that the production quality of the future automotive wiring harness meets the standard, but there are latent defects;

[0032] If the quality defect coefficient is less than or equal to the first defect threshold, it is determined that the production quality of the future automotive wiring harness meets the standard.

[0033] In a preferred embodiment, the SHAP algorithm is used to analyze the influence of each feature data on the quality decline, including the following steps:

[0034] Input the feature data predicted by the gated recurrent unit for the future automotive wiring harness production into the SHAP interpretation model, and the SHAP interpretation model calculates the SHAP value of each feature data. The expression is:

[0035] In the formula, φ j is the SHAP value of feature data j, S is a subset of the feature data set, F is the entire feature data set, f(S) represents the prediction result of the gated recurrent unit using only the subset, |S| is the size of the subset, and |F| is the total number of feature data;

[0036] Draw a feature importance ranking diagram through the SHAP-summary-plot tool. In the feature importance ranking diagram, the color represents the size of the feature value. The larger the SHAP value of the feature data, the greater the influence of the feature data on the quality decline.

[0037] In a preferred embodiment, the contribution rate of the feature value is calculated by the ratio of the feature value to the sum of all feature values. The expression is: In the formula, G(λ i ) represents the contribution rate of the i-th feature value, λ i represents the i-th feature value, and N represents the number of feature values in the covariance matrix;

[0038] Calculate the cumulative contribution rate of each principal component. The expression is: In the formula, G L is the cumulative contribution rate, and k is the current feature value index, representing the cumulative contribution of the first k feature values;

[0039] Project the centered data matrix onto the principal component matrix to obtain the feature data matrix. The expression is: Z = X ′ P, where Z is the feature data matrix, X ′ is the centered data matrix, and P is the principal component matrix.

[0040] In a preferred embodiment, the terminal crimping force, insulation layer thickness, conductor resistance, crimping die wear, equipment operation time, temperature, humidity, vibration amplitude, unqualified rate, and rework rate parameters after normalization processing are used to construct the original data matrix X. The expression is:

[0041] In the formula, m is the number of samples, n is the number of original multi-source data in the sample, and xij represents the j-th original multi-source data value of the i-th sample, and the dimension of the original data matrix X is m×n.

[0042] In a preferred embodiment, in the original data matrix X, the mean value of each original multi-source data value is calculated, and the expression is: In the formula, μ j represents the mean value of the j-th original multi-source data value among all samples, m is the number of samples, and x ij represents the j-th original multi-source data value of the i-th sample;

[0043] The original data matrix X is centered, and the expression is: x′ ij = x ij - μ j , where x′ ij represents the de-centered value of the j-th original multi-source data of the i-th sample;

[0044] The covariance matrix C is calculated based on the de-centered data matrix X′, and the expression is: In the formula, m is the number of samples, X′ T is the transpose matrix of the de-centered data matrix, m - 1 represents the number of samples minus 1, which is used to calculate the unbiased estimate. The dimension of the covariance matrix C is n×n, which represents the covariance between each pair of de-centered values. The element C ij in the covariance matrix C represents the covariance between the i-th de-centered value and the j-th de-centered value.

[0045] A method for traceability management of wire harness processing quality based on big data. The acquisition end acquires multiple sample data from the big database, and each sample data contains original multi-source data related to wire harness anomalies, including the following steps:

[0046] Acquire multiple batches of sample data from the big database. Each sample data contains original multi-source data related to wire harness quality, including process parameters, equipment status data, environmental factors, historical quality record data, and production batch information.

[0047] In the above technical solution, the technical effects and advantages provided by the present invention are:

[0048] After calculating the eigenvalues and eigenvectors of the covariance matrix, the present invention selects the principal components with a cumulative contribution rate greater than the contribution rate threshold, projects the de-centered data matrix into the principal component space to obtain feature data, substitutes the feature data of several historical batches into a gated recurrent unit, and the gated recurrent unit performs time series prediction to predict the quality trend of future wire harnesses. The SHAP algorithm (influence factor analysis algorithm) is used to analyze the influence of each feature data on the quality decline, and a quality management strategy is output based on the prediction result of the gated recurrent unit and the analysis result of the SHAP algorithm. The management method projects high-dimensional data into a low-dimensional principal component space, reduces the computational complexity, uses a gated recurrent unit for time series prediction, early warns of possible quality deterioration, and ensures the production efficiency of automotive wire harnesses. Brief Description of the Drawings

[0049] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings required in the embodiments. Obviously, the drawings described below are only some embodiments recorded in the present invention. For those of ordinary skill in the art, other drawings can also be obtained based on these drawings.

[0050] Figure 1 It is a flowchart of the method of the present invention. Detailed Embodiments

[0051] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the drawings in the embodiments of the present invention. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0052] Embodiment 1

[0053] Please refer to Figure 1 As shown, a wire harness processing quality traceability management method based on big data in this embodiment includes the following steps:

[0054] S1: After cleaning the original multi-source data, construct an original data matrix;

[0055] S2: Generate a covariance matrix based on the original data matrix, and calculate the eigenvalues and eigenvectors of the covariance matrix;

[0056] S3: Select the principal components with a cumulative contribution rate greater than the contribution rate threshold, and project the de-centered data matrix into the principal component space to obtain feature data;

[0057] S4: The gated recurrent unit performs time series prediction and uses the SHAP algorithm to analyze the impact of each feature data on quality degradation;

[0058] S5: Output a quality management strategy based on the prediction results of the gated recurrent unit and the analysis results of the SHAP algorithm.

[0059] The acquisition end collects multiple sample data from the large database. Each sample data contains original multi-source data related to wire harness anomalies. After cleaning the original multi-source data, an original data matrix is constructed, and a covariance matrix is generated based on the original data matrix. After calculating the eigenvalues and eigenvectors of the covariance matrix, the principal components with a cumulative contribution rate greater than the contribution rate threshold are selected. The decentralized data matrix is projected into the principal component space to obtain feature data. The feature data of several historical batches are input into the gated recurrent unit. The gated recurrent unit performs time series prediction to predict the future quality trend of the wire harness. The SHAP algorithm (influence factor analysis algorithm, Shapley-Additive-Explanations) is used to analyze the impact of each feature data on quality degradation. A quality management strategy is output based on the prediction results of the gated recurrent unit and the analysis results of the SHAP algorithm.

[0060] In this application, after calculating the eigenvalues and eigenvectors of the covariance matrix, the principal components with a cumulative contribution rate greater than the contribution rate threshold are selected. The decentralized data matrix is projected into the principal component space to obtain feature data. The feature data of several historical batches are input into the gated recurrent unit. The gated recurrent unit performs time series prediction to predict the future quality trend of the wire harness. The SHAP algorithm (influence factor analysis algorithm) is used to analyze the impact of each feature data on quality degradation. A quality management strategy is output based on the prediction results of the gated recurrent unit and the analysis results of the SHAP algorithm. The management method projects high-dimensional data into a low-dimensional principal component space to reduce the computational complexity, uses a gated recurrent unit for time series prediction, and gives early warnings of possible quality deterioration to ensure the production efficiency of automotive wire harnesses.

[0061] Embodiment 2

[0062] The acquisition end collects multiple sample data from the large database. Each sample data contains original multi-source data related to wire harness anomalies, including the following steps:

[0063] Collect multi-source quality data from the production process and perform data cleaning to prepare for subsequent analysis. Collect multiple batches of samples from the large database. Each sample contains original data related to wire harness quality, including: process parameters (terminal crimping force, insulation layer thickness, conductor resistance), equipment status (crimping die wear, equipment operation time), environmental factors (temperature and humidity, vibration), historical quality records (unqualified rate, rework rate), production batch information (timestamp, operator, raw material batch).

[0064] After cleaning the original multi-source data, construct the original data matrix, and generate the covariance matrix based on the original data matrix, including the following steps:

[0065] Clean the original multi-source data, including removing outliers (using the IQR method and Z-score method to eliminate extreme abnormal data), handling missing values (using mean filling and interpolation method to complete missing data), and normalizing data (normalization processing to make data with different features have the same scale). Among them, common methods for filling missing values include using the mean, median, or interpolation method to fill missing data. For example, for numerical features, the mean of the feature can be used to fill missing values; for time series data, linear interpolation can be used to fill missing data.

[0066] Delete missing values: In some cases, if the proportion of missing data is very high, it may be necessary to directly delete the feature or sample.

[0067] Construct the original data matrix X with the parameters such as terminal crimping force, insulation layer thickness, conductor resistance, crimping die wear, equipment operation time, temperature, humidity, vibration amplitude, unqualified rate, and rework rate after the normalization process. The expression is:

[0068] In the formula, m is the number of samples, n is the number of original multi-source data in the sample, and x ij represents the j-th original multi-source data value of the i-th sample. The dimension of the original data matrix X is m×n, where each row represents a sample and each column represents an original multi-source data value.

[0069] In the original data matrix X, calculate the mean of each original multi-source data value. The expression is:

[0070] In the formula, μ j represents the mean of the j-th original multi-source data value in all samples, m is the number of samples, and x ij represents the j-th original multi-source data value of the i-th sample.

[0071] And perform centering processing on the original data matrix X, that is, subtract the mean of each original multi-source data value from the original multi-source data value. The expression is: x′ ij =x ij -μ j , where x′ ij represents the decentralized value of the j-th original multi-source data of the i-th sample, x ij represents the j-th original multi-source data value of the i-th sample, and μ jdenotes the mean of the j-th original multi-source data value among all samples. The de-centered data matrix X′ changes the mean of each column to 0.

[0072] Calculate the covariance matrix C based on the de-centered data matrix X′. The covariance matrix C represents the covariance between different de-centered values, and the calculation formula is as follows: In the formula, m is the number of samples, X ′T is the transpose matrix of the de-centered data matrix. m - 1 represents the number of samples minus 1, which is used to calculate the unbiased estimate. The dimension of the covariance matrix C is n×n, representing the covariance between each pair of de-centered values. The element C ij in the covariance matrix C represents the covariance between the i-th de-centered value and the j-th de-centered value. The covariance value represents the linear relationship between two features. A positive value indicates a positive correlation between the two features, a negative value indicates a negative correlation, and a zero value indicates no correlation.

[0073] After calculating the eigenvalues and eigenvectors of the covariance matrix, select the principal components with a cumulative contribution rate greater than the contribution rate threshold, and project the de-centered data matrix onto the principal component space to obtain the feature data, including the following steps:

[0074] After obtaining the covariance matrix C, calculate the eigenvalues and eigenvectors of the covariance matrix C. The equation expression is: Cv = λv, where λ is the eigenvalue and v is the eigenvector. In this application, the eigenvalue decomposition method in linear algebra (such as through functions like numpy.linalg.eig) is used to calculate the eigenvalues and eigenvectors of the covariance matrix. The relevant code example is as follows:

[0075] import numpy as np

[0076] # Example covariance matrix C

[0077] C = np.array([[2.0, 0.8], [0.8, 1.0]]);

[0078] # Calculate eigenvalues and eigenvectors

[0079] eigenvalues, eigenvectors = np.linalg.eig(C);

[0080] # Output results

[0081] print("Eigenvalues:", eigenvalues);

[0082] print("Eigenvectors:\n", eigenvectors)。

[0083] Sort the eigenvalues of the covariance matrix in descending order to obtain a sorted list of eigenvalues: λ1≥λ2≥…≥λ N , where N represents the number of eigenvalues in the covariance matrix, and λ i represents the i-th eigenvalue. Sort the eigenvectors of the covariance matrix in descending order to obtain a sorted list of eigenvectors: v1≥v2≥…≥v M , where M represents the number of eigenvectors in the covariance matrix, and v i represents the i-th eigenvector.

[0084] Calculate the contribution rate of eigenvalues by the ratio of an eigenvalue to the sum of all eigenvalues. The expression is: where G(λ i ) represents the contribution rate of the i-th eigenvalue, λ i represents the i-th eigenvalue, and N represents the number of eigenvalues in the covariance matrix. Then accumulate the contribution rates of eigenvalues to obtain the cumulative contribution rate of each principal component. The expression is: where G L is the cumulative contribution rate, k is the current eigenvalue index, representing the cumulative contribution of the first k eigenvalues, λ i represents the i-th eigenvalue, and N represents the number of eigenvalues in the covariance matrix.

[0085] Usually, a contribution rate threshold (such as 90%, 95%, etc.) is set to determine how many principal components to select. Assume the set contribution rate threshold is 95%, that is, the principal components with a cumulative contribution rate greater than 95% are selected. If the cumulative contribution rate is greater than the contribution rate threshold, select the eigenvectors corresponding to the first k eigenvalues as the principal components. For example, when the cumulative contribution rate exceeds 95%, select the principal components corresponding to the first k eigenvectors.

[0086] Combine the selected principal component eigenvectors into a principal component matrix for data projection. The selected first k eigenvectors form an N×k principal component matrix P, where each column is an eigenvector. The expression is: P = [v1, v2, …, v k , and v i represents the i-th eigenvector, and k is the number of selected principal components.

[0087] Project the centered data matrix onto the principal component matrix to obtain the feature data and get the low-dimensional feature data matrix. The expression is: Z = X ′ P, where Z is the feature data matrix, X ′ is the centered data matrix, and P is the principal component matrix. Through this step, the original high-dimensional data is converted into low-dimensional data, retaining the main features and trends of the data.

[0088] Thereby obtaining low-dimensional feature data, which includes low-dimensional terminal crimping force, low-dimensional insulation layer thickness, low-dimensional conductor resistance, low-dimensional crimping die wear, low-dimensional equipment operation time, low-dimensional temperature, low-dimensional humidity, low-dimensional vibration amplitude, low-dimensional unqualified rate, and low-dimensional rework rate.

[0089] Substitute the feature data of several historical batches into the gated recurrent unit (GRU). The gated recurrent unit performs time series prediction to predict the quality trend of future wire harnesses, including the following steps:

[0090] Obtain the feature data matrix of several historical batches from the production process, [Z1, Z2, …, Z A , where A is the number of historical production batches, and Z i is the low-dimensional feature data of the i-th batch of wire harnesses. The low-dimensional feature data includes low-dimensional terminal crimping force, low-dimensional insulation layer thickness, low-dimensional conductor resistance, low-dimensional crimping die wear, low-dimensional equipment operation time, low-dimensional temperature, low-dimensional humidity, low-dimensional vibration amplitude, low-dimensional unqualified rate, and low-dimensional rework rate.

[0091] Input the feature data matrix of several historical batches into the pre-trained gated recurrent unit. The gated recurrent unit outputs the quality trend of future E batches of automotive wire harnesses, that is, the change status of each future feature data.

[0092] The pre-training of the gated recurrent unit includes the following steps:

[0093] The basic structure of the gated recurrent unit includes:

[0094] Input layer: Receive the historical feature data Z A .

[0095] GRU layer: Process time series data and extract the temporal features of historical data. Multiple GRU layers and appropriate numbers of neurons can be set.

[0096] Output layer: Predict the quality trend of future E batches of automotive wire harnesses. That is, the prediction results for each future batch.

[0097] Use the feature data of several historical batches as input, train through historical data, optimize the weights and parameters of the gated recurrent unit, use a loss function (such as mean squared error MSE) to measure the difference between the prediction results and the actual data, and update the parameters of the gated recurrent unit through the backpropagation algorithm. The pre-training of the gated recurrent unit belongs to the prior art and will not be elaborated in this application.

[0098] Use the SHAP algorithm (Shapley-Additive-Explanations, an algorithm for analyzing influencing factors) to analyze the impact of each feature data on the quality decline, including the following steps:

[0099] Since the gated recurrent unit belongs to the deep learning model and is suitable for time series analysis, Deep-SHAP (an extension of SHAP based on DeepLIFT) or Kernel-SHAP (a sampling-based approximation method) is used to calculate the SHAP values.

[0100] Define the SHAP interpreter: In this application, the SHAP library is used to load the gated recurrent unit, and the input sample range for calculating the SHAP values is defined, and a benchmark data set is obtained for comparison with historical data to interpret future prediction batches.

[0101] The characteristic data of the gated recurrent unit predicting the future production of automotive wire harnesses is input into the SHAP interpretation model. The SHAP interpretation model calculates the SHAP values (i.e., contribution values) of each characteristic data. The SHAP value represents the degree of contribution of this characteristic to the prediction of quality decline, and the expression is:

[0102] In the formula, φ j is the SHAP value of the characteristic data j, S is a subset of the characteristic data set, F is the entire characteristic data set, f(S) represents the prediction result of the gated recurrent unit using only the subset, |S| is the size of the subset, representing the number of characteristic data of the currently selected characteristic subset, |F| is the total number of characteristic data. The SHAP algorithm traverses all possible characteristic combinations, calculates the marginal contribution of each characteristic data under different combinations, and finally obtains the SHAP value of each characteristic data.

[0103] During the SHAP calculation process, we will traverse all subsets S of the characteristic set F to analyze the influence of each characteristic under different combinations. The entire characteristic data set contains all characteristics that affect quality prediction, such as "terminal pressing force", "conductor resistance", "insulation layer thickness", etc. f(S∪{j}) represents the model prediction result after adding the characteristic data j on the basis of the subset S.

[0104] Suppose there are three characteristics affecting the quality of the wire harness:

[0105] Terminal pressing force (Feature1);

[0106] Conductor resistance (Feature2);

[0107] Insulation layer thickness (Feature3).

[0108] The goal is to analyze the contribution of the terminal pressing force (Feature1) to the prediction of quality decline.

[0109] First, for example, calculate the model prediction values for different feature subsets S. When only using "conductor resistance", the model predicts that the probability of the harness quality decreasing is 0.6. When only using "insulation layer thickness", the model predicts that the probability of quality decreasing is 0.5. When only using "terminal pressing force", the model predicts that the probability of quality decreasing is 0.7. When using "conductor resistance + terminal pressing force", the model predicts that the probability of quality decreasing is 0.8. When using "all features", the model predicts that the probability of quality decreasing is 0.9.

[0110] Calculate the marginal contribution of the terminal pressing force f(S ∪ {terminal pressing force}) - f(S):

[0111] For S = {conductor resistance}:

[0112] f(S ∪ {terminal pressing force}) = f({conductor resistance, terminal pressing force}) = 0.8;

[0113] f(S) = f({conductor resistance}) = 0.6;

[0114] The contribution value of the terminal pressing force = 0.8 - 0.6 = 0.2.

[0115] Calculate in turn to get the contribution value of the insulation layer thickness = 0.25, and the contribution value of the conductor resistance + insulation layer thickness = 0.2.

[0116] Then calculate the SHAP value of the terminal pressing force:

[0117] The contribution of the terminal pressing force (SHAP value = 0.217) to the quality decrease is relatively large, indicating that the abnormality in the terminal pressing process may be a key factor affecting the quality. If the SHAP value of the conductor resistance is higher than that of the terminal pressing force, it may indicate that material aging or wire quality problems are the main influencing factors. Combining these results, production managers can give priority to checking the terminal pressing process and optimizing the equipment maintenance strategy.

[0118] Use the SHAP-summary-plot tool to draw a feature importance ranking graph to see which feature data has the most significant impact on the quality decrease. In the feature importance ranking graph, the color represents the size of the feature value, and the positive or negative of the SHAP value indicates the direction of the impact on the quality trend. A higher SHAP value of the feature data indicates that this feature has a greater impact on the quality decrease.

[0119] Output the quality management strategy based on the prediction results of the gated recurrent unit and the analysis results of the SHAP algorithm, including the following steps:

[0120] The prediction result of the gated recurrent unit is the change status of the feature data, that is, the change value of the feature data is obtained. The change value of the feature data includes the terminal crimping force deviation, the insulation layer thickness deviation, the conductor resistance growth rate, the crimping die wear degree, the equipment operation duration deviation, the temperature deviation, the humidity deviation, the vibration amplitude, the unqualified rate, and the rework rate. The above feature data are all feature data after dimensionality reduction processing (including normalization processing to map the value ranges between feature data to [0, 1]).

[0121] And the SHAP value of each feature data is obtained through the SHAP algorithm, and the weight coefficient of each feature data is generated based on the SHAP value of the feature data. The generation logic is: the SHAP total value is obtained by summing the SHAP values of all feature data, and the weight coefficient of the feature data is obtained by dividing the SHAP value by the SHAP total value;

[0122] The weighted calculation of all feature data change values is used to obtain the quality defect coefficient of the future automotive wire harness. The expression is: In the formula, qsd is the quality defect coefficient, B is the number of feature data change values, ω i is the weight coefficient of the i-th feature data change value, and TS i is the i-th feature data change value.

[0123] In this application, since the feature data change values include the terminal crimping force deviation, the insulation layer thickness deviation, the conductor resistance growth rate, the crimping die wear degree, the equipment operation duration deviation, the temperature deviation, the humidity deviation, the vibration amplitude, the unqualified rate, and the rework rate, therefore, B = 10, that is, the low-dimensional terminal crimping force, the low-dimensional insulation layer thickness, the low-dimensional conductor resistance, the low-dimensional crimping die wear, the low-dimensional equipment operation time, the low-dimensional temperature, the low-dimensional humidity, the low-dimensional vibration amplitude, the low-dimensional unqualified rate, and the low-dimensional rework rate are weighted and calculated to obtain the quality defect coefficient.

[0124] The larger the quality defect coefficient, the more unqualified the predicted production quality of the automotive wire harness is. The obtained quality defect coefficient is compared with the preset first defect threshold and the second defect threshold. The second defect threshold is used to judge whether the quality of the automotive wire harness meets the standard, and the first defect threshold is used to judge whether there are latent defects in the automotive wire harness.

[0125] If the quality defect coefficient is greater than the second defect threshold, it is judged that the production quality of the future automotive wire harness does not meet the standard;

[0126] If the quality defect coefficient is less than or equal to the second defect threshold and greater than the first defect threshold, it is determined that the production quality of the future automotive wiring harness meets the standard, but there are latent defects (latent defects are a type of hidden defect, that is, after the production of the automotive wiring harness is completed, the quality of the automotive wiring harness is detected as meeting the standard in the short term, but it will affect the later service life of the automotive wiring harness and reduce the bearing capacity of the automotive wiring harness under severe working conditions, etc.);

[0127] If the quality defect coefficient is less than or equal to the first defect threshold, it is determined that the production quality of the future automotive wiring harness meets the standard.

[0128] After determining the production quality of the future automotive wiring harness, the generated management strategies are as follows:

[0129] 1) The quality defect coefficient is greater than the second defect threshold (production quality does not meet the standard): Immediately stop production of the current production batch and conduct quality inspections to identify the key factors causing the defects. Combine SHAP influence factor analysis to determine the main defect sources (such as insufficient crimping force, die wear, high humidity, etc.). Analyze the similarity with previous unqualified batches through historical data comparison to determine whether it is a repetitive problem. Reset the equipment parameters (such as crimping force, temperature and humidity, conductor resistance, etc.). If it is caused by equipment aging, arrange for equipment maintenance or replacement. Strengthen online monitoring and conduct 100% quality inspection on the current production batch to ensure the quality stability of subsequent batches. Use enhanced detection methods (such as microscopic metallographic analysis, electrical testing, pull-off force testing, etc.) to detect potential hazards. Provide technical training to production personnel for the problems found to enhance quality awareness and operation specifications. If it is found that the quality defect involves wiring harness design problems, the design department needs to be notified for optimization, such as adjusting the wiring harness specifications, terminal materials, etc.

[0130] 2) The quality defect coefficient is between the first defect threshold and the second defect threshold (production quality meets the standard, but there are latent defects): Issue a quality warning to the production management team, indicating that the current batch may have latent defects and special attention is required. Based on SHAP analysis, find the main factors affecting quality and appropriately adjust the production parameters (such as optimizing the terminal crimping force, reducing the impact of die wear). Increase the sampling frequency of key quality parameters, such as improving the monitoring accuracy of temperature and humidity and adjusting the environmental conditions in real time. Use intelligent sensors to monitor the equipment operation status to prevent latent defects caused by equipment fluctuations. Conduct additional aging tests, vibration tests, and long-term reliability tests on this batch of wiring harnesses to ensure their stability under severe working conditions. Conduct additional durability tests before leaving the factory and establish a traceability mechanism to monitor the long-term performance of this batch of products. If this batch of wiring harnesses is supplied to key application scenarios (such as high-temperature, high-vibration environments), it is recommended to prioritize the investigation of high-risk applications to ensure that the product quality meets the requirements.

[0131] 3) The quality defect coefficient is less than or equal to the first defect threshold (production quality meets the standard): Record the current process parameters as the reference for the standard process to ensure that subsequent batches are produced according to the same parameters. Combine historical quality data analysis to optimize production efficiency and increase the production speed on the premise of ensuring quality. Plan for the maintenance of crimping equipment, molds, etc. to prevent equipment aging caused by long-term use from affecting quality. Build a data model for the production process and optimize the AI prediction algorithm to improve the accuracy of future quality prediction.

[0132] The above formulas are all calculated by taking the numerical values without dimensions. The formulas are obtained by collecting a large amount of data for software simulation to get a formula closest to the actual situation. The preset parameters in the formulas are set by those skilled in the art according to the actual situation.

[0133] In the description of this specification, the descriptions referring to terms such as "one embodiment", "example", "specific example", etc. mean that the specific features, structures, materials or characteristics described in connection with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in a suitable manner in any one or more embodiments or examples.

[0134] The preferred embodiments of the present invention disclosed above are only used to help explain the present invention. The preferred embodiments do not describe all the details in detail, nor do they limit the present invention to only the specific implementation manners. Obviously, many modifications and variations can be made according to the content of this specification. These embodiments are selected and specifically described in this specification to better explain the principles and practical applications of the present invention, so that those skilled in the art can understand and utilize the present invention well. The present invention is only limited by the claims and their full scope and equivalents.

Claims

1. A method for tracing and managing the quality of wire harness processing based on big data, characterized in that: The management method includes the following steps: S1: Construct an original data matrix after cleaning the original multi-source data; S2: Generate a covariance matrix based on the original data matrix, and calculate the eigenvalues and eigenvectors of the covariance matrix; S3: Select the principal components with a cumulative contribution rate greater than the contribution rate threshold, and project the de-centered data matrix into the principal component space to obtain feature data; S4: Use a gated recurrent unit for time series prediction, and analyze the impact of each feature data on quality degradation using the SHAP algorithm; S5: Output a quality management strategy based on the prediction results of the gated recurrent unit and the analysis results of the SHAP algorithm.

2. The quality traceability management method for harness processing based on big data according to claim 1, characterized in that: Generating a covariance matrix based on the original data matrix includes the following steps: Construct an original data matrix X with the parameters of terminal crimping force, insulation layer thickness, conductor resistance, crimping die wear, equipment operation time, temperature, humidity, vibration amplitude, unqualified rate, and rework rate after normalization processing, and calculate the mean value of each original multi-source data value in the original data matrix X; Perform centering processing on the original data matrix X, that is, subtract the mean value of each original multi-source data value from the original multi-source data value to obtain the de-centered value of the original multi-source data; Calculate the covariance matrix C based on the de-centered data matrix X′, and the covariance matrix C represents the covariance between different de-centered values.

3. The method for tracing and managing the quality of wire harness processing based on big data according to claim 2, wherein: Calculating the eigenvalues and eigenvectors of the covariance matrix includes the following steps: After obtaining the covariance matrix C, use the eigenvalue decomposition method in linear algebra to calculate the eigenvalues and eigenvectors of the covariance matrix C. The equation expression is: Cv = λv, where λ is the eigenvalue and v is the eigenvector; Sort the eigenvalues of the covariance matrix in descending order to obtain the sorted eigenvalue list: λ1≥λ2≥…≥λ N , where N represents the number of eigenvalues in the covariance matrix, and λ i represents the i-th eigenvalue; Sort the eigenvectors of the covariance matrix in descending order to obtain a list of eigenvectors: v1≥v2≥…≥v M , where M represents the number of eigenvectors in the covariance matrix, and v i represents the i-th eigenvector.

4. The method for traceability management of wire harness processing quality based on big data according to claim 3, characterized in that: Selecting the principal components with a cumulative contribution rate greater than the contribution rate threshold and projecting the de-centered data matrix into the principal component space to obtain feature data includes the following steps: Calculate the contribution rate of the eigenvalue by the ratio of the eigenvalue to the sum of all eigenvalues, and then accumulate the contribution rate of the eigenvalue to obtain the cumulative contribution rate of each principal component; If the cumulative contribution rate of the principal component is greater than the contribution rate threshold, select the eigenvectors corresponding to the first k eigenvalues as the principal components; The selected top k eigenvectors form an N×k principal component matrix P, P = [v1, v2, …, v k , where v i represents the i-th eigenvector, k is the number of principal components selected, and the de-centered data matrix is projected onto the principal component matrix to obtain the feature data and acquire the low-dimensional feature data matrix.

5. The method for traceability management of harness processing quality based on big data according to claim 4, characterized in that: Outputting a quality management strategy based on the prediction results of the gated recurrent unit and the analysis results of the SHAP algorithm includes the following steps: Obtain the SHAP value of each feature data through the SHAP algorithm, and generate the weight coefficient of each feature data based on the SHAP value of the feature data. The generation logic is: sum the SHAP values of all feature data to obtain the total SHAP value, and divide the SHAP value by the total SHAP value to obtain the weight coefficient of the feature data; The weighted calculation of all the change values of the characteristic data is performed to obtain the quality defect coefficient of the future automotive wiring harness, and the expression is: In the formula, qsd is the quality defect coefficient, B is the number of change values of the characteristic data, ω i is the weight coefficient of the i-th change value of the characteristic data, TS i is the i-th change value of the characteristic data; Compare the obtained quality defect coefficient with a preset first defect threshold and a second defect threshold. The second defect threshold is used to judge whether the quality of the automotive wire harness meets the standard, and the first defect threshold is used to judge whether there are latent defects in the automotive wire harness; If the quality defect coefficient is greater than the second defect threshold, judge that the production quality of the future automotive wire harness does not meet the standard; If the quality defect coefficient is less than or equal to the second defect threshold and the quality defect coefficient is greater than the first defect threshold, judge that the production quality of the future automotive wire harness meets the standard but there are latent defects; If the quality defect coefficient is less than or equal to the first defect threshold, it is determined that the production quality of the future automotive wire harness meets the standard.

6. The method for tracing and managing the quality of wire harness processing based on big data according to claim 5, characterized in that: Use the SHAP algorithm to analyze the impact of each feature data on the quality decline, including the following steps: Input the feature data predicted by the gated recurrent unit for the future production of automotive wire harnesses into the SHAP interpretation model. The SHAP interpretation model calculates the SHAP value of each feature data. The expression is: where φ j is the SHAP value of the feature data j, S is a subset of the feature data set, F is the entire feature data set, f(S) represents the prediction result of the gated recurrent unit using only the subset, |S| is the size of the subset, and |F| is the total number of feature data; Draw a feature importance ranking diagram through the SHAP-summary-plot tool. In the feature importance ranking diagram, the color represents the size of the feature value. The larger the SHAP value of the feature data, the greater the impact of the feature data on the quality decline.

7. The method for traceability management of harness processing quality based on big data according to claim 3, characterized in that: The contribution rate of the eigenvalue is calculated by the ratio of the eigenvalue to the sum of all eigenvalues, and the expression is: In the formula, G(λ i ) represents the contribution rate of the i-th eigenvalue, λ i represents the i-th eigenvalue, and N represents the number of eigenvalues in the covariance matrix; Calculate the cumulative contribution rate of each principal component, and the expression is as follows: In the formula, G L is the cumulative contribution rate, and k is the current eigenvalue index, representing the cumulative contribution of the first k eigenvalues; Project the de-centered data matrix onto the principal component matrix to obtain the feature data matrix. The expression is: Z = X'P, where Z is the feature data matrix, X' is the de-centered data matrix, and P is the principal component matrix.

8. The method for traceability management of harness processing quality based on big data according to claim 2, characterized in that: Construct the original data matrix X with the terminal crimping force, insulation layer thickness, conductor resistance, crimping die wear, equipment operation time, temperature, humidity, vibration amplitude, unqualified rate, and rework rate parameters after normalization processing. The expression is: where m is the number of samples, n is the number of original multi-source data in the samples, and x ij represents the j-th original multi-source data value of the i-th sample, and the dimension of the original data matrix X is m×n.

9. The method for tracing and managing the quality of wire harness processing based on big data according to claim 8, characterized in that: In the original data matrix X, calculate the mean value of each original multi-source data value, and the expression is: In the formula, μ j represents the mean value of the j-th original multi-source data value in all samples, m is the number of samples, and x ij represents the j-th original multi-source data value of the i-th sample; The original data matrix X is centered, and the expression is: x′ ij = x ij - μ j , where x′ ij represents the decentralized value of the j-th original multi-source data of the i-th sample; Calculate the covariance matrix C based on the decentralized data matrix X′, and the expression is: In the formula, m is the number of samples, and X′ T is the transpose matrix of the decentralized data matrix. m - 1 represents the number of samples minus 1, which is used to calculate the unbiased estimate. The dimension of the covariance matrix C is n×n, indicating the covariance between each pair of decentralized values. The element C ij in the covariance matrix C represents the covariance between the i-th decentralized value and the j-th decentralized value.

10. The method for traceability management of harness processing quality based on big data according to claim 9, characterized in that: The acquisition end collects multiple sample data from the large database. Each sample data contains the original multi-source data related to the wire harness anomaly, including the following steps: Collect multiple batches of sample data from the large database. Each sample data contains the original multi-source data related to the wire harness quality, including process parameters, equipment status data, environmental factors, historical quality record data, and production batch information.

Images