Wire harness product quality prediction system based on big data

Abstract

The invention discloses a wire harness product quality prediction system based on big data. An initial multi-source data set is acquired; core features in the initial multi-source data set are extracted based on the crimping height, the insulation resistance value and the environment temperature and humidity of the wire harness quality, high-weight features in the core features are screened through a PCA principal component analysis method, and wire harness feature data are obtained; processing time series data based on a long-short-term memory network, performing feature selection by using an extreme gradient boosting tree, and establishing a hybrid prediction model; using an improved IWOA whale optimization algorithm to optimize hyper-parameters of the hybrid prediction model; and inputting the wire harness characteristic data into the target hybrid prediction model for prediction, outputting a quality risk grade index, and if the quality risk grade index exceeds a set threshold, triggering an early warning signal. The limitation of traditional single data or simple model prediction is changed, so that quality prediction better fits an actual production scene, and the accuracy and reliability of prediction are remarkably improved.

Description

Technical Field

[0001] The present invention relates to the technical field of wire harness product quality detection, and in particular to a wire harness product quality prediction system based on big data. Background Art

[0002] Wire harness products (such as automotive, high-voltage, and electronic harnesses) are core components of modern industrial equipment, and their quality directly impacts the safety and reliability of these devices. Traditional wire harness quality inspection relies primarily on manual visual inspection, electrical testing, or mechanical performance testing. This reliance on manual experience leads to low efficiency and accuracy, and is prone to misjudgments due to human oversight. Summary of the Invention

[0003] The purpose of the present invention is to solve the above problems and design a wire harness product quality prediction system based on big data.

[0004] The technical solution of the present invention to achieve the above object is that, further, in the above-mentioned wire harness product quality prediction system based on big data, the wire harness product quality prediction system includes: A wire harness data acquisition module is used to acquire a multi-source data set in a wire harness product production line, and pre-process the multi-source data set to obtain an initial multi-source data set; A core feature extraction module is used to extract core features from the initial multi-source dataset based on the crimping height, insulation resistance value, and ambient temperature and humidity of the wire harness quality, and to screen high-weight features from the core features using a principal component analysis method to obtain wire harness feature data; The prediction model building module is used to process time series data based on the Bi-LSTM long short-term memory network, use the XGBoost extreme gradient boosting tree for feature selection, and establish a hybrid prediction model; use the improved IWOA whale optimization algorithm to optimize the hyperparameters of the hybrid prediction model to obtain the LSTM-XGBoost-IWOA target hybrid prediction model; The wiring harness quality prediction module is used to input the wiring harness feature data into the LSTM-XGBoost-IWOA target hybrid prediction model for prediction, and output a quality risk level index. If the quality risk level index exceeds a set threshold, an early warning signal is triggered.

[0005] Furthermore, in the above-mentioned wire harness product quality prediction system based on big data, the wire harness data acquisition module includes the following units: The data cleaning unit is used to identify and process outliers in the multi-source data set using the Z-score statistical method, delete the data that deviates from the normal range, and obtain a cleaned multi-source data set; The data reduction unit is used to perform dimensionality reduction and format unification on the cleaned multi-source dataset to obtain an initial multi-source dataset.

[0006] Furthermore, in the above-mentioned wire harness product quality prediction system based on big data, the core feature extraction module includes the following units: A data standardization unit is used to extract core features from the initial multi-source data set, and perform standardization processing on the core feature data to obtain standardized core features; A covariance matrix calculation unit, used to calculate the covariance matrix between each feature in the standardized core feature to obtain a core feature matrix; an eigenvector solving unit, configured to perform eigendecomposition on the core eigenmatrix to obtain eigenvalues ​​and corresponding eigenvectors; The high-weight feature screening unit is used to determine the original feature that contributes more to the main component based on the load coefficient between the main component and the original feature, and obtain the wiring harness feature data.

[0007] Furthermore, in the above-mentioned wire harness product quality prediction system based on big data, the prediction model building module includes the following units: The concatenation unit is used to utilize a bidirectional structure, where the forward layer transmits information from past time points to the current time point, and the backward layer transmits information from future time points to the current time point. The output of the bidirectional structure is concatenated in the hidden layer. The conversion unit, used for each LSTM unit, contains a forget gate, an input gate, and an output gate. It selectively retains key information through the gating mechanism and converts the output of the bidirectional hidden layer into a feature vector as the input of the XGBoost model.

[0008] Furthermore, in the above-mentioned wire harness product quality prediction system based on big data, the prediction model building module also includes the following units: The contribution calculation unit is used to calculate the contribution of each feature to reducing the loss function in the tree node splitting based on the splitting characteristics of the decision tree in the XGBoost extreme gradient boosting tree; The threshold setting unit is used to set the importance threshold, filter out irrelevant features that have low contribution to quality prediction, and gradually build multiple decision trees in an iterative manner, with each tree fitting the prediction residual of the previous tree.

[0009] Furthermore, in the above-mentioned wire harness product quality prediction system based on big data, the prediction model building module also includes the following units: Initialization unit, used to randomly generate 50 sets of hyperparameter combinations as initial whale individuals, each set of parameters corresponds to a candidate model; The data calculation unit is used to substitute each set of parameters into the hybrid model and calculate the prediction accuracy and mean square error using the validation set data; The contraction search unit is used to imitate the whale's spiral contraction search around the optimal solution, and adjusts the step size through adaptive weights to approach the potential optimal parameters; The model acquisition unit is used to select the hyperparameter combination with the highest fitness when the maximum number of iterations is reached, substitute it into the hybrid model for training, and generate the LSTM-XGBoost-IWOA target prediction model.

[0010] Furthermore, in the above-mentioned wire harness product quality prediction system based on big data, the wire harness quality prediction module includes the following units: Vector output unit, used to process time series data using Bi-LSTM and output feature vectors containing long-term dependencies; The result prediction unit is used to input the feature vector output by Bi-LSTM into the XGBoost extreme gradient boosting tree, use the gradient boosting mechanism to model the nonlinear relationship between features, and output the quality risk level prediction result; The monitoring and early warning unit is used to set a quality risk level threshold based on the company's quality standards and historical data, and trigger an early warning signal if the quality risk level index exceeds the set threshold.

[0011] Furthermore, in the method for implementing the above-mentioned wire harness product quality prediction system based on big data, the method includes the following steps: Acquire a multi-source dataset in a wire harness product production line, and pre-process the multi-source dataset to obtain an initial multi-source dataset; Extracting core features from the initial multi-source dataset based on the crimping height, insulation resistance value, and ambient temperature and humidity of the wire harness quality, and using PCA principal component analysis to screen high-weight features from the core features to obtain wire harness feature data; Based on the Bi-LSTM long short-term memory network to process time series data, the XGBoost extreme gradient boosting tree is used for feature selection to establish a hybrid prediction model. The improved IWOA whale optimization algorithm is used to optimize the hyperparameters of the hybrid prediction model to obtain the LSTM-XGBoost-IWOA target hybrid prediction model. The wiring harness feature data is input into the LSTM-XGBoost-IWOA target hybrid prediction model for prediction, and a quality risk level index is output. If the quality risk level index exceeds a set threshold, an early warning signal is triggered.

[0012] Furthermore, in the method for implementing the above-mentioned wire harness product quality prediction system based on big data, the method includes the following steps: Using a bidirectional structure, the forward layer transmits information from the past time point to the current time point, and the backward layer transmits information from the future time point to the current time point. The output of the bidirectional structure is spliced ​​in the hidden layer; Each LSTM unit contains a forget gate, an input gate, and an output gate. It selectively retains key information through the gating mechanism and converts the output of the bidirectional hidden layer into a feature vector as the input of the XGBoost model.

[0013] Furthermore, in the method for implementing the above-mentioned wire harness product quality prediction system based on big data, the method includes the following steps: Based on the splitting characteristics of the decision tree in the XGBoost extreme gradient boosting tree, the contribution of each feature to reducing the loss function in the tree node split is calculated; An importance threshold is set to filter out irrelevant features that have low contribution to quality prediction, and multiple decision trees are gradually constructed in an iterative manner, with each tree fitting the prediction residual of the previous tree.

[0014] Its beneficial effects are as follows: 1. It changes the limitations of traditional single-data or simple model predictions, making quality predictions more relevant to actual production scenarios and significantly improving the accuracy and reliability of predictions. 2. It effectively reduces the dimensionality and complexity of data and reduces data redundancy, which not only improves the efficiency of model training but also avoids prediction bias caused by excessive interference from irrelevant data, allowing the model to focus on key quality influencing factors and improve prediction accuracy. 3. It effectively solves the problem of traditional algorithms easily falling into local optimality and finds the optimal hyperparameter combination for the model. This enables the model to maintain good predictive performance under different production conditions and data distributions, possesses stronger generalization capabilities, and adapts to complex and changing actual production environments. 4. It enables companies to detect quality risks in advance during the production process, changing the previous passive mode of post-detection. Relevant personnel can take timely measures to adjust production parameters, repair equipment, or replace raw materials to avoid the production of defective products, reduce quality losses, and improve production efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Various other advantages and benefits will become apparent to those skilled in the art by reading the following detailed description of the preferred embodiment.The accompanying drawings are only for the purpose of illustrating the preferred embodiment and are not to be considered as limiting the present invention.

[0016] Figure 1 Schematic diagram of a first embodiment of a wire harness product quality prediction system based on big data in an embodiment of the present invention; Figure 2 Schematic diagram of a second embodiment of a wire harness product quality prediction system based on big data in an embodiment of the present invention; Figure 3 Schematic diagram of a third embodiment of a wire harness product quality prediction system based on big data in an embodiment of the present invention. DETAILED DESCRIPTION

[0017] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0018] Those skilled in the art will appreciate that, unless otherwise stated, the singular forms "a", "an", "" and "the" used herein may also include plural forms. It should be further understood that the term "comprising" used in the specification of the present invention refers to the presence of features, integers, steps, operations, elements and / or components, but does not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components and / or groups thereof.

[0019] The present invention will be described in detail below with reference to the accompanying drawings. Figure 1 As shown, a wire harness product quality prediction system based on big data includes: A wire harness data acquisition module is used to acquire a multi-source data set from a wire harness product production line, pre-process the multi-source data set, and obtain an initial multi-source data set; Specifically, this embodiment further includes a data cleaning unit for identifying and processing outliers in the multi-source data set using a Z-score statistical method, deleting data that deviates from a normal range, and obtaining a cleaned multi-source data set; The data reduction unit is used to reduce the dimensions and unify the format of the cleaned multi-source data set to obtain the initial multi-source data set.

[0020] Specifically, (i) multi-source dataset acquisition; Production data: This covers the real-time operating parameters of each piece of equipment on the production line, such as the crimping machine's key parameters like pressure, temperature, and speed, and the terminal machine's die model and crimping times. It also includes process parameters like wire cutting length, wire stripping length, and terminal crimping depth. It also records information like production time, shifts, and operators, allowing for subsequent analysis of how different factors impact quality.

[0021] Test data: This includes crimp height test data, which uses a high-precision sensor to collect the crimp height of the terminal in real time after the crimping process is completed; insulation resistance test data, which uses a professional insulation resistance tester to measure the insulation resistance between each conductor and between the conductor and the shield layer of the finished wire harness; in addition, there is also continuity test data and voltage drop test data, which comprehensively reflect the electrical performance of the wire harness.

[0022] Environmental data: Mainly refers to the ambient temperature and humidity data in the production workshop, which is collected in real time and at multiple points through temperature and humidity sensors deployed in different locations in the workshop; at the same time, data on environmental factors such as air pressure and air quality (dust concentration) that may affect the production process and wire harness quality are collected.

[0023] Historical data: including records of quality defects in previously produced wire harness products, including detailed records of defect types (poor crimping, insulation damage, wire short circuit, etc.), location, and time of occurrence; historical inspection reports, covering various inspection indicators and results of each batch of wire harnesses; quality statistics, such as yield rate and defect rate in different time periods; and quality problem data reported by customers, including failure phenomena and causes during use.

[0024] (2) Data preprocessing; Data cleaning: Remove noise from the acquired multi-source data, use statistical methods (Z-score method) to identify and process outliers, and correct or delete data that significantly deviates from the normal range; for missing data, use mean filling, median filling, or regression filling methods based on the missing ratio and data characteristics; at the same time, detect and remove duplicate data to ensure the uniqueness and accuracy of the data.

[0025] Data integration: Integrate data from different data sources (production equipment systems, test equipment systems, environmental monitoring systems, quality record systems, etc.) to resolve data conflicts, such as unifying data formats (time format, units, etc.), and performing consistency processing on the same entities (equipment numbers, product batch numbers) in different systems to ensure data integrity and consistency.

[0026] Data conversion: Qualitative data (operator name, shift name, terminal type, etc.) is quantified and converted into numerical data using methods such as one-hot encoding and label encoding. Numerical data is normalized or standardized, such as minimum-maximum normalization to scale the data to the [0,1] range, or Z-score standardization to give the data zero mean and unit variance for subsequent data analysis and model training.

[0027] Data reduction: Dimensionality reduction techniques such as principal component analysis (PCA) are used to simplify the dimensions of preprocessed multi-source datasets, reducing the amount of data while retaining the main information. At the same time, feature selection methods such as correlation analysis and variance analysis are combined to screen out features that have a greater impact on quality prediction and reduce data complexity.

[0028] The core feature extraction module is used to extract the core features from the initial multi-source dataset based on the crimp height, insulation resistance value, and ambient temperature and humidity of the wire harness quality. The PCA principal component analysis method is used to screen the high-weight features in the core features to obtain the wire harness feature data. Specifically, this embodiment further includes a data standardization unit for extracting core features from the initial multi-source data set, and performing standardization processing on the core feature data to obtain standardized core features; A covariance matrix calculation unit is used to calculate the covariance matrix between each feature in the standardized core feature to obtain the core feature matrix; The eigenvector solving unit is used to perform eigendecomposition on the core eigenmatrix to obtain eigenvalues ​​and corresponding eigenvectors; The high-weight feature screening unit is used to determine the original feature that contributes more to the main component based on the load coefficient between the main component and the original feature, and obtain the wiring harness feature data.

[0029] Specifically, (i) core feature extraction; From the pre-processed initial multi-source dataset, the following core features are extracted based on the key factors affecting the quality of the wiring harness: Crimp height: directly reflects the tightness of the terminal crimping and is a key parameter affecting the crimping quality. The crimping height of each terminal is accurately measured by a sensor.

[0030] Insulation resistance value: An important indicator that reflects the insulation performance of the wiring harness. The insulation resistance test is used to obtain the resistance values ​​between each wire and between the wire and the shielding layer.

[0031] Ambient temperature and humidity: The temperature and humidity in the production workshop have an impact on the wire material, terminal crimping process, etc. Ambient temperature and humidity data are collected and recorded in real time.

[0032] Equipment operating parameters: such as the pressure, temperature, speed of the crimping machine, the mold model of the terminal machine, etc. These parameters directly affect the stability of the production process and the quality of the wire harness.

[0033] Process parameters: including wire cutting length, wire stripping length, terminal crimping depth, etc. The accuracy and consistency of process parameters are the basis for ensuring the quality of wire harnesses.

[0034] Wire material and specifications: Wire material (purity of copper wire), cross-sectional area, insulation material, etc. Different materials and specifications will affect the electrical and mechanical properties of the wiring harness.

[0035] Terminal type and quality: The terminal type (ring terminal, fork terminal), material, surface treatment, etc. The quality of the terminal is directly related to the reliability of crimping.

[0036] (2) High-weight feature screening; The PCA principal component analysis method is used to screen the extracted core features. The specific steps are as follows: Data standardization: Standardize the core feature data to eliminate the impact of different feature dimensions and numerical ranges, making each feature comparable.

[0037] Calculate the covariance matrix: Based on the standardized data, calculate the covariance matrix between each feature to reflect the linear correlation between the features.

[0038] Solve eigenvalues ​​and eigenvectors: Perform eigendecomposition on the covariance matrix to obtain eigenvalues ​​and corresponding eigenvectors. The size of the eigenvalue represents the amount of information contained in the principal component.

[0039] Determine the number of principal components: According to the principle that the cumulative variance contribution rate reaches more than 85%, select the first k principal components, each of which is a linear combination of the original features.

[0040] Screening high-weight features: According to the load coefficient between the principal component and the original feature, determine the original features that contribute more to the principal component, that is, high-weight features. These features retain the main information of the original data and are used for subsequent model training.

[0041] The prediction model building module is used to process time series data based on the Bi-LSTM long short-term memory network, use the XGBoost extreme gradient boosting tree for feature selection, and build a hybrid prediction model. The improved IWOA whale optimization algorithm is used to optimize the hyperparameters of the hybrid prediction model to obtain the LSTM-XGBoost-IWOA target hybrid prediction model. Specifically, this embodiment further includes a splicing unit for utilizing a bidirectional structure, wherein the forward layer transmits information from the past time point to the current time point, and the backward layer transmits information from the future time point to the current time point, and the output of the bidirectional structure is spliced ​​in the hidden layer; The conversion unit, used for each LSTM unit, contains a forget gate, an input gate, and an output gate. It selectively retains key information through the gating mechanism and converts the output of the bidirectional hidden layer into a feature vector as the input of the XGBoost model.

[0042] The contribution calculation unit is used to calculate the contribution of each feature to reducing the loss function in the tree node splitting based on the splitting characteristics of the decision tree in the XGBoost extreme gradient boosting tree; The threshold setting unit is used to set the importance threshold, filter out irrelevant features that have low contribution to quality prediction, and gradually build multiple decision trees in an iterative manner, with each tree fitting the prediction residual of the previous tree.

[0043] Initialization unit, used to randomly generate 50 sets of hyperparameter combinations as initial whale individuals, each set of parameters corresponds to a candidate model; The data calculation unit is used to substitute each set of parameters into the hybrid model and calculate the prediction accuracy and mean square error using the validation set data; The contraction search unit is used to imitate the whale's spiral contraction search around the optimal solution, and adjusts the step size through adaptive weights to approach the potential optimal parameters; The model acquisition unit is used to select the hyperparameter combination with the highest fitness when the maximum number of iterations is reached, substitute it into the hybrid model for training, and generate the LSTM-XGBoost-IWOA target prediction model.

[0044] Specifically, (i) hybrid prediction model construction; 1. Bi-LSTM long short-term memory network design; The Bi-LSTM network is specifically designed to process production data with time series characteristics. Its core function is to capture the long-term dependencies between data at different time points. The specific implementation logic is as follows: Input layer design: Receive preprocessed time series feature data, including production parameters (crimping machine pressure, temperature), test data (crimping height, insulation resistance), environmental data (temperature and humidity), etc., arranged in chronological order. The data format is [time step, feature dimension]. For example, a 10-minute time window is used, and each window contains 20 features, forming an input matrix of [60, 20] (assuming the time step is 60).

[0045] Bidirectional hidden layer mechanism: Using a bidirectional structure (forward LSTM and backward LSTM), the forward layer transmits information from past time points to the current time point, while the backward layer transmits information from future time points to the current time point. The outputs of these two layers are concatenated in the hidden layer, enabling the model to leverage contextual information from both historical and future data. For example, the current quality of a crimping process may be related to temperature fluctuations in the equipment over the previous few hours, or it may be affected by insulation resistance anomalies in subsequent testing. The bidirectional structure can comprehensively capture these cross-period correlations.

[0046] Memory unit and gating mechanism: Each LSTM unit contains a forget gate, an input gate, and an output gate. The gating mechanism selectively retains or forgets key information. For example, when a device parameter experiences a brief fluctuation (such as transient voltage instability), the forget gate determines whether the noise needs to be filtered to prevent short-term interference from affecting long-term trend analysis. The input gate, on the other hand, incorporates stable process parameters (consistently stable crimp depth) into long-term memory.

[0047] The output layer converts the output of the bidirectional hidden layer into a feature vector, which serves as the input to the XGBoost model. The output layer does not use an activation function and directly outputs continuous-valued features, preserving the changing trend of the original time series data.

[0048] 2. XGBoost feature selection and model training; XGBoost undertakes two core tasks in the hybrid model: feature screening and strong prediction model construction. The specific process is as follows: Feature Importance Assessment: Based on the splitting properties of decision trees, each feature's contribution to reducing the loss function during a tree node split is calculated. For example, if crimp height has the highest feature importance score, it indicates that this parameter plays a key role in determining quality risk. If ambient humidity has a lower score, its weight may be reduced or even eliminated in subsequent modeling.

[0049] Dynamic feature screening: By setting an importance threshold (retaining the top 30% of features), the model automatically filters out irrelevant features (minor operating parameters of non-critical equipment) that contribute little to quality prediction, reducing data redundancy. For example, if the importance score of wire cut length falls below the threshold, the model ignores this feature and focuses on core parameters such as crimp height and insulation resistance.

[0050] Gradient boosting training: Multiple decision trees are iteratively constructed, with each tree fitting the prediction residuals of the previous tree. For example, the first tree learns the underlying pattern in the data (the linear relationship between crimp height and yield), while the second tree specifically learns examples where the first tree mispredicted (abnormal crimping in high humidity environments). This weighted combination of multiple trees improves the model's ability to fit complex nonlinear relationships.

[0051] Regularization prevents overfitting: L1 / L2 regularization terms are added to the objective function to constrain the complexity of the decision tree. For example, the maximum depth of a single tree is limited (set to 6 layers) to prevent the model from overfitting to noise in the training data and improve generalization ability.

[0052] 3. Hybrid model fusion logic; Bi-LSTM and XGBoost are combined through "feature-level fusion": Time series feature extraction: Bi-LSTM first processes the time series data and outputs a feature vector containing long-term dependencies (a 50-dimensional feature vector is generated for each time window), converting the dynamic change pattern of the original time series data into static features.

[0053] Nonlinear modeling: The feature vector output by Bi-LSTM is input into XGBoost, and its gradient boosting mechanism is used to model the nonlinear relationship between features (the interaction between crimp height and insulation resistance), ultimately outputting the quality risk level prediction result.

[0054] (2) Improve the parameters of the IWOA whale optimization algorithm; 1. Traditional algorithm improvement strategy; To address the problems of traditional whale optimization algorithms that are prone to falling into local optimality and slow convergence, two improvement mechanisms are introduced: Adaptive weight adjustment: Initially (for the first 30% of iterations), a large global search weight is set to allow individual whales to explore the hyperparameter space extensively, avoiding missing the optimal solution. For example, the search step factor is linearly decayed from 1.5 to 1.0, expanding the search range to cover a wide range of parameters such as the number of Bi-LSTM hidden layer neurons (20-200) and the XGBoost learning rate (0.01-0.3).

[0055] In the later stages (after 70% of iterations), the search step size is reduced to focus on a localized, refined search near the current optimal solution, improving the accuracy of parameter fine-tuning. For example, the step size factor is decayed from 1.0 to 0.2, and dense sampling is performed within the initially determined optimal parameter range (100-120 hidden layer neurons).

[0056] Elite retention mechanism: After each iteration, the top 5% of individuals with the highest fitness are retained as "elite solutions" and directly enter the next generation of the population to prevent the loss of high-quality parameter combinations due to random perturbations. For example, if a set of hyperparameters (120 hidden layers, a learning rate of 0.1, and a tree depth of 8) achieves a model accuracy of 92%, this solution is retained for subsequent iterations to prevent performance regression due to over-exploration.

[0057] 2. Hyperparameter optimization range; The core hyperparameters that need to be optimized are divided into two categories: Bi-LSTM network parameters: The number of neurons in the hidden layer (affects the model capacity, such as 50-200, with a step size of 10); Training batch size (affects training stability, such as 32, 64, 128, 256); Learning rate (controls the gradient descent step size, such as 0.001, 0.01, 0.1); Dropout rate (to prevent overfitting, such as 0.1, 0.2, 0.3); XGBoost model parameters: Maximum tree depth (controls the complexity of a single tree, such as 3-10); Subsampling rate (random sampling ratio of samples, such as 0.6, 0.8, 1.0); Regularization parameters (L1 / L2 weights, such as 0.01, 0.1, 1.0); Learning rate (increase step size, such as 0.01, 0.05, 0.1); 3. Optimization process implementation steps; Initialize the population: Randomly generate 50 sets of hyperparameter combinations as initial whale individuals, and each set of parameters corresponds to a candidate model.

[0058] Fitness evaluation: Each set of parameters was substituted into the hybrid model, and the prediction accuracy (the proportion of samples correctly classified into quality risk levels) and mean square error (MSE, reflecting the prediction accuracy of continuous values) were calculated using the validation set data.

[0059] The fitness function is constructed using a weighted summation method (accuracy accounts for 70% of the weight and MSE accounts for 30% of the weight) to comprehensively evaluate the model performance.

[0060] Simulating predation behavior: Surrounding predation: Imitate the whale's spiral contraction search around the optimal solution (the individual with the highest current fitness), adjust the step size through adaptive weights, and gradually approach the potential optimal parameters.

[0061] Bubble Net Attack: Randomly select local areas for deep search to prevent the algorithm from falling into the local neighborhood of the elite solution and balance global exploration and local development.

[0062] Iteration termination condition: When the maximum number of iterations (200 generations) is reached or the fitness value fluctuation is less than 0.5% for 20 consecutive generations, it is considered convergence and the search is stopped.

[0063] Optimal parameter determination: Select the hyperparameter combination with the highest fitness, substitute it into the hybrid model for final training, and generate the LSTM-XGBoost-IWOA target prediction model.

[0064] Through the above-mentioned refined design, the hybrid model can not only use Bi-LSTM to capture the dynamic correlation of time series data, but also process nonlinear feature interactions through XGBoost. Combined with the efficient parameter adjustment of the improved whale algorithm, it can ultimately achieve accurate prediction of wire harness quality risks.

[0065] The wire harness quality prediction module is used to input the wire harness feature data into the LSTM-XGBoost-IWOA target hybrid prediction model for prediction and output the quality risk level index. If the quality risk level index exceeds the set threshold, an early warning signal is triggered.

[0066] Specifically, this embodiment further includes a vector output unit for processing time series data using Bi-LSTM and outputting a feature vector containing long-term dependencies; The result prediction unit is used to input the feature vector output by Bi-LSTM into the XGBoost extreme gradient boosting tree, use the gradient boosting mechanism to model the nonlinear relationship between features, and output the quality risk level prediction result; The monitoring and early warning unit is used to set the quality risk level threshold according to the company's quality standards and historical data. If the quality risk level index exceeds the set threshold, an early warning signal will be triggered.

[0067] Specifically, (a) quality prediction; The filtered wire harness feature data is fed into the trained LSTM-XGBoost-IWOA target hybrid prediction model. The model then calculates the quality risk level index through forward propagation. The quality risk level index is a numerical value that comprehensively reflects the quality risk of the wire harness. The value range is [0, 10], with larger values ​​indicating higher quality risk.

[0068] (2) early warning mechanism; Threshold setting: Set the quality risk level threshold based on the company's quality standards and historical data. For example, if the threshold is set to 6, when the quality risk level index exceeds 6, it is considered that there is a high quality risk.

[0069] Early warning signal triggering: When the quality risk level index output by the model exceeds the set threshold, the system automatically triggers an early warning signal and informs relevant personnel (production management personnel, quality inspection personnel) through sound and light alarms, SMS notifications, email reminders, etc.

[0070] Early warning and handling measures: After receiving the early warning signal, relevant personnel will immediately check the production process and analyze the possible causes of quality risks, such as equipment failure, abnormal process parameters, raw material quality problems, etc., and take corresponding measures to adjust and improve, such as shutting down equipment for maintenance, resetting process parameters, replacing raw materials, etc., to eliminate quality risks in a timely manner and ensure the stable quality of wire harness products.

[0071] Its beneficial effects are as follows: by obtaining a multi-source data set from a wire harness product production line, the multi-source data set is pre-processed to obtain an initial multi-source data set; the core features in the initial multi-source data set are extracted based on the crimping height, insulation resistance value, and ambient temperature and humidity of the wire harness quality, and the high-weight features in the core features are screened using the PCA principal component analysis method to obtain the wire harness feature data; the time series data is processed based on the Bi-LSTM long short-term memory network, and the feature selection is performed using the XGBoost extreme gradient boosting tree to establish a hybrid prediction model; the hyperparameters of the hybrid prediction model are optimized using the improved IWOA whale optimization algorithm to obtain the LSTM-XGBoost-IWOA target hybrid prediction model; the wire harness feature data is input into the LSTM-XGBoost-IWOA target hybrid prediction model for prediction, and a quality risk level index is output. If the quality risk level index exceeds the set threshold, an early warning signal is triggered. 1. It changes the limitations of traditional single data or simple model predictions, making quality predictions more in line with actual production scenarios and significantly improving the accuracy and reliability of predictions. 2. It effectively reduces the dimensionality and complexity of the data and reduces data redundancy, which not only improves the efficiency of model training but also avoids prediction bias caused by excessive interference from irrelevant data, allowing the model to focus on key quality influencing factors and improve prediction accuracy. 3. It effectively solves the problem of traditional algorithms easily falling into local optimality and finds the optimal hyperparameter combination for the model. This enables the model to maintain good predictive performance under different production conditions and data distributions, possesses stronger generalization capabilities, and adapts to complex and changing actual production environments. 4. It enables companies to detect quality risks in advance during the production process, changing the previous passive mode of post-detection. Relevant personnel can take timely measures to adjust production parameters, repair equipment, or replace raw materials to avoid the production of defective products, reduce quality losses, and improve production efficiency.

[0072] See also Figure 2 In the wire harness product quality prediction system based on big data, the core feature extraction modules include: The data standardization unit is used to extract the core features from the initial multi-source data set, standardize the core feature data, and obtain standardized core features; A covariance matrix calculation unit is used to calculate the covariance matrix between each feature in the standardized core feature to obtain the core feature matrix; The eigenvector solving unit is used to perform eigendecomposition on the core eigenmatrix to obtain eigenvalues ​​and corresponding eigenvectors; The high-weight feature screening unit is used to determine the original feature that contributes more to the main component based on the load coefficient between the main component and the original feature, and obtain the wiring harness feature data.

[0073] See also Figure 3In the wire harness product quality prediction system based on big data, the wire harness quality prediction module includes: Vector output unit, used to process time series data using Bi-LSTM and output feature vectors containing long-term dependencies; The result prediction unit is used to input the feature vector output by Bi-LSTM into the XGBoost extreme gradient boosting tree, use the gradient boosting mechanism to model the nonlinear relationship between features, and output the quality risk level prediction result; The monitoring and early warning unit is used to set the quality risk level threshold according to the company's quality standards and historical data. If the quality risk level index exceeds the set threshold, an early warning signal will be triggered.

[0074] The above shows and describes the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The above embodiments and descriptions are merely preferred examples of the present invention and are not intended to limit the present invention. Various changes and improvements may be made to the present invention without departing from the spirit and scope of the present invention, and all such changes and improvements fall within the scope of the present invention. The scope of protection claimed in the present invention is defined by the appended claims and their equivalents.

Claims

1. A wire harness product quality prediction system based on big data, characterized by: The wire harness product quality prediction system includes: A wire harness data acquisition module is used to acquire a multi-source data set in a wire harness product production line, and pre-process the multi-source data set to obtain an initial multi-source data set; A core feature extraction module is used to extract core features from the initial multi-source dataset based on the crimping height, insulation resistance value, and ambient temperature and humidity of the wire harness quality, and to screen high-weight features from the core features using a principal component analysis method to obtain wire harness feature data; The prediction model building module is used to process time series data based on the Bi-LSTM long short-term memory network, use the XGBoost extreme gradient boosting tree for feature selection, and establish a hybrid prediction model; use the improved IWOA whale optimization algorithm to optimize the hyperparameters of the hybrid prediction model to obtain the LSTM-XGBoost-IWOA target hybrid prediction model; The wiring harness quality prediction module is used to input the wiring harness feature data into the LSTM-XGBoost-IWOA target hybrid prediction model for prediction, and output a quality risk level index. If the quality risk level index exceeds a set threshold, an early warning signal is triggered.

2. The wire harness product quality prediction system based on big data according to claim 1, characterized in that: The wiring harness data acquisition module includes the following units: The data cleaning unit is used to identify and process outliers in the multi-source data set using the Z-score statistical method, delete the data that deviates from the normal range, and obtain a cleaned multi-source data set; The data reduction unit is used to perform dimensionality reduction and format unification on the cleaned multi-source dataset to obtain an initial multi-source dataset.

3. The wire harness product quality prediction system based on big data according to claim 1, characterized in that: The core feature extraction module includes the following units: A data standardization unit is used to extract core features from the initial multi-source data set, and perform standardization processing on the core feature data to obtain standardized core features; A covariance matrix calculation unit, used to calculate the covariance matrix between each feature in the standardized core feature to obtain a core feature matrix; an eigenvector solving unit, configured to perform eigendecomposition on the core eigenmatrix to obtain eigenvalues ​​and corresponding eigenvectors; The high-weight feature screening unit is used to determine the original feature that contributes more to the main component based on the load coefficient between the main component and the original feature, and obtain the wiring harness feature data.

4. The wire harness product quality prediction system based on big data according to claim 1, characterized in that: The prediction model building module includes the following units: The concatenation unit is used to utilize a bidirectional structure, where the forward layer transmits information from past time points to the current time point, and the backward layer transmits information from future time points to the current time point. The output of the bidirectional structure is concatenated in the hidden layer. The conversion unit, used for each LSTM unit, contains a forget gate, an input gate, and an output gate. It selectively retains key information through the gating mechanism and converts the output of the bidirectional hidden layer into a feature vector as the input of the XGBoost model.

5. The wire harness product quality prediction system based on big data according to claim 4, characterized in that: The prediction model building module also includes the following units: The contribution calculation unit is used to calculate the contribution of each feature to reducing the loss function in the tree node splitting based on the splitting characteristics of the decision tree in the XGBoost extreme gradient boosting tree; The threshold setting unit is used to set the importance threshold, filter out irrelevant features that have low contribution to quality prediction, and gradually build multiple decision trees in an iterative manner, with each tree fitting the prediction residual of the previous tree.

6. The wire harness product quality prediction system based on big data according to claim 4, characterized in that: The prediction model building module also includes the following units: Initialization unit, used to randomly generate 50 sets of hyperparameter combinations as initial whale individuals, each set of parameters corresponds to a candidate model; The data calculation unit is used to substitute each set of parameters into the hybrid model and calculate the prediction accuracy and mean square error using the validation set data; The contraction search unit is used to imitate the whale's spiral contraction search around the optimal solution, and adjusts the step size through adaptive weights to approach the potential optimal parameters; The model acquisition unit is used to select the hyperparameter combination with the highest fitness when the maximum number of iterations is reached, substitute it into the hybrid model for training, and generate the LSTM-XGBoost-IWOA target prediction model.

7. The wire harness product quality prediction system based on big data according to claim 1, characterized in that: The harness quality prediction module includes the following units: Vector output unit, used to process time series data using Bi-LSTM and output feature vectors containing long-term dependencies; The result prediction unit is used to input the feature vector output by Bi-LSTM into the XGBoost extreme gradient boosting tree, use the gradient boosting mechanism to model the nonlinear relationship between features, and output the quality risk level prediction result; The monitoring and early warning unit is used to set a quality risk level threshold based on the company's quality standards and historical data, and trigger an early warning signal if the quality risk level index exceeds the set threshold.

8. The method for implementing the wire harness product quality prediction system based on big data as claimed in claim 1, characterized in that: The method comprises the following steps: Acquire a multi-source dataset in a wire harness product production line, and pre-process the multi-source dataset to obtain an initial multi-source dataset; Extracting core features from the initial multi-source dataset based on the crimping height, insulation resistance value, and ambient temperature and humidity of the wire harness quality, and using PCA principal component analysis to screen high-weight features from the core features to obtain wire harness feature data; Based on the Bi-LSTM long short-term memory network to process time series data, the XGBoost extreme gradient boosting tree is used for feature selection to establish a hybrid prediction model. The improved IWOA whale optimization algorithm is used to optimize the hyperparameters of the hybrid prediction model to obtain the LSTM-XGBoost-IWOA target hybrid prediction model. The wiring harness feature data is input into the LSTM-XGBoost-IWOA target hybrid prediction model for prediction, and a quality risk level index is output. If the quality risk level index exceeds a set threshold, an early warning signal is triggered.

9. The method for implementing the wire harness product quality prediction system based on big data as claimed in claim 1, characterized in that: The method comprises the following steps: Using a bidirectional structure, the forward layer transmits information from the past time point to the current time point, and the backward layer transmits information from the future time point to the current time point. The output of the bidirectional structure is spliced ​​in the hidden layer; Each LSTM unit contains a forget gate, an input gate, and an output gate. It selectively retains key information through the gating mechanism and converts the output of the bidirectional hidden layer into a feature vector as the input of the XGBoost model.

10. The method for implementing the wire harness product quality prediction system based on big data as claimed in claim 1, characterized in that: The method comprises the following steps: Based on the splitting characteristics of the decision tree in the XGBoost extreme gradient boosting tree, the contribution of each feature to reducing the loss function in the tree node split is calculated; An importance threshold is set to filter out irrelevant features that have low contribution to quality prediction, and multiple decision trees are gradually constructed in an iterative manner, with each tree fitting the prediction residual of the previous tree.

Images