An automobile parts supplier quality early warning system and method

Abstract

This invention provides a quality early warning system and method for automotive parts suppliers. The system includes a parts data acquisition module, an indicator information module, an early warning model module, a data analysis and matching module, a risk control measures module, and an early warning information push module. The quality early warning method for automotive parts suppliers of this invention establishes a parts quality early warning indicator system to classify risk types, achieving a shift from post-event result tracing to process and proactive risk warning. Through an XGBoost-Random Forest-Bayesian network hybrid algorithm model, it enhances the ability to identify potential quality fluctuations in suppliers, reducing false alarms and missed alarms. It deeply integrates model early warning with actual business processes, promptly pushing early warning information and improvement suggestions, simultaneously identifying risk batches and guiding the formulation of control measures, forming a closed loop of "early warning-response-control," solving the "lag" problem of traditional post-event early warning, and achieving proactive prevention and control of quality risks.

Description

Technical Field

[0001] This invention relates to the field of information systems, and more specifically to a quality early warning system and method for automotive parts suppliers. Background Technology

[0002] The domestic and international automotive industries are currently entering a stage of high-quality development and fierce competition. On the one hand, the automotive supply chain is characterized by globalization and multiple nodes, with a wide variety of purchased parts and a large number of suppliers. The factors affecting quality are becoming increasingly complex, and a single post-inspection model is insufficient to address potential parts quality risks in the supply chain. Once a quality problem occurs with a purchased part, it can easily trigger a chain reaction such as production stoppages and customer complaints. On the other hand, competition in the automotive market is fierce, and consumers are paying increasing attention to product quality. Quality reputation has become an important part of a company's core competitiveness, and traditional passive quality control is insufficient to meet the market's demand for "zero-defect" products.

[0003] Meanwhile, the application of AI technology in various fields continues to deepen, and its advantages in data mining and risk prediction are gradually becoming more prominent, providing technical support for quality early warning of automotive supply chain parts. However, many automotive companies currently rely solely on post-event early warning mechanisms, lacking the ability to identify quality risks in advance and unable to avoid quality problems of purchased parts in a timely manner. They urgently need to leverage technological means to build a scientific quality early warning system to achieve a shift from "post-event remediation" to "pre-event prevention." Summary of the Invention

[0004] This invention is made to solve the above-mentioned problems, and its purpose is to provide a quality early warning system and method for automotive parts suppliers.

[0005] This invention provides a quality early warning system for automotive parts suppliers, characterized by the following features:

[0006] The parts data acquisition module is used to collect quality data of automotive parts;

[0007] The indicator information storage module stores quality datasets based on the parts quality early warning indicator system;

[0008] The early warning model module contains an early warning model obtained through training and iteration using quality data.

[0009] The three-tiered judgment module is used to determine the risk level of quality data at multiple levels, including the Class I veto judgment unit, the Class II cumulative trigger judgment unit, and the Class III ranking screening judgment unit and ranking calculation submodule.

[0010] The data analysis and matching module is used to call the early warning model to analyze the quality data. When the quality data triggers the high-risk threshold, the risk level is output according to the three-level classification judgment module, and early warning information is generated.

[0011] The risk management measures module is used to store improvement measures and re-inspection plans corresponding to various risk types;

[0012] The early warning information push module is used to push early warning information and corresponding improvement suggestions.

[0013] The automotive parts supplier quality early warning system provided by this invention may also have the following features: the parts quality early warning indicator system divides quality data into three risk levels: the first level indicators are after-sales fault and quality record indicators divided according to trend analysis, including QR and after-sales ECB; the second level indicators are process indicators divided according to process capability analysis, including repeat QR, OSM, and pre-sales batch problem indicators; the third level indicators are other risk indicators, including the first level indicators, such as after-sales batch activity problems, unauthorized changes, and recall indicators.

[0014] The automotive parts supplier quality early warning system provided by this invention may also have the following features: the early warning model is based on an XGBoost-random Forest hybrid algorithm model, which includes three layers: the first layer is a feature capture layer, which uses the XGBoost algorithm to capture nonlinear feature relationships in the quality data using its gradient boosting framework; the second layer is a generalization fitting layer, which uses the random forest algorithm to reduce the risk of overfitting by constructing multiple decision trees and integrating voting; the third layer is a probabilistic inference layer, which uses a Bayesian network algorithm to perform probabilistic inference on the classification results of the three classification judgment modules and the cumulative ranking results output by the ranking calculation module based on the conditional probability graphical model, and calculates the posterior probability of each supplier being at different risk levels; the outputs of the three layers are weighted and fused to obtain the final risk warning probability.

[0015] The automotive parts supplier quality early warning system provided by this invention may also have the following features: Among the three types of classification judgment modules, the Type I veto judgment unit is used to determine whether a supplier has triggered a veto condition. When a supplier has pre-sales batch problems, unauthorized changes, or recall events, the veto rule is triggered, and the supplier is directly judged as a high-risk supplier; the Type II cumulative trigger judgment unit is used to determine whether a supplier has reached a cumulative trigger threshold. When the cumulative number of pre-sales batch problems, OSM problems, or repeated QR problems of a supplier reaches a preset threshold, the supplier is judged as a high-risk supplier; the Type III ranking and screening judgment unit is used to identify suppliers whose quality is continuously deteriorating but have no obvious violations based on a quantitative ranking model. The Type III ranking and screening judgment unit calls the ranking calculation module to perform a comprehensive ranking and takes the top 10 suppliers with the smallest cumulative ranking value as Type III high-risk suppliers.

[0016] The automotive parts supplier quality early warning system provided by this invention may also have the following feature: the ranking calculation submodule includes:

[0017] The absolute value ranking unit is used to calculate the ranking of each supplier's total ECB and total QR within a rolling time window;

[0018] The trend ranking unit is used to calculate the trend slope of each supplier's after-sales failure and quality records within a rolling time window using linear regression slope, and to rank the trend slope.

[0019] The comprehensive ranking unit is used to sum the rankings across all dimensions to obtain a cumulative sort.

[0020] The high-risk determination unit is used to sort suppliers by cumulative ranking from smallest to largest, and the top 10 are selected as high-risk suppliers.

[0021] This invention also provides a quality early warning method for automotive parts suppliers, comprising the following steps:

[0022] Parts data acquisition steps: Real-time acquisition of quality data for automotive parts;

[0023] The indicator information storage steps involve storing a quality dataset based on the part quality early warning indicator system, with the quality data in the quality dataset categorized according to the part quality early warning indicator system.

[0024] The steps for building the early warning model are as follows: 70% of the quality data in the quality dataset is used as the training dataset, and the remaining 30% of the quality data is used as the validation dataset. Precision, recall, F1 score and ROC curve are used as evaluation indicators to build the early warning model.

[0025] The three-tiered classification process assesses the risk level of quality data at multiple levels and calculates and outputs a multi-dimensional ranking of suppliers.

[0026] The data analysis and matching step calls the early warning model to analyze the quality data and determine whether the quality data triggers the high-risk threshold. If so, it matches the risk level obtained from the three-level judgment step with the corresponding risk type in the part quality early warning indicator system and generates early warning information.

[0027] Risk management measures and steps, and corresponding improvement suggestions based on early warning information;

[0028] The early warning information push process involves sending early warning information and improvement suggestions to engineers and supplier liaisons at each STA, and simultaneously locking down the production and warehousing processes of risky batches of parts.

[0029] Compared with the prior art, the functions and effects of the present invention include:

[0030] (1) The automotive parts supplier quality early warning system of the present invention uses the data analysis and matching module to call the early warning model to process the quality data and generate early warning information. The risk control measures module simultaneously formulates corresponding control measures, and then the early warning information push module pushes emails to STA and suppliers, realizing data-driven and process closed loop. It deeply integrates model early warning with actual business processes, forming a closed loop of "early warning-response-control", transforming "post-event remediation" into "pre-event prevention", solving the problem of the lag of traditional post-event early warning. Through real-time data collection and intelligent early warning, it realizes early detection and early handling of supply chain quality risks.

[0031] (2) In the automotive parts supplier quality early warning system of the present invention, the indicator information storage module divides the quality data into three levels of indicators according to the degree of impact, which corresponds to the three levels in the three-level judgment module. This breaks through the limitations of the traditional single result indicator early warning indicator system and forms a multi-dimensional, all-round, and progressive risk assessment system.

[0032] (3) The early warning model in the automotive parts supplier quality early warning system and method of the present invention is based on the XGBoost-random forest-Bayes network hybrid algorithm model. Compared with the existing single algorithm early warning model, the early warning model in the present invention integrates the advantages of multiple algorithms. It adopts XGBoost nonlinear feature capture, random forest to reduce overfitting, and Bayesian network to realize the complete link of probabilistic risk reasoning, thereby improving the ability to identify potential quality fluctuations of suppliers and significantly improving the early warning recall rate. Attached Figure Description

[0033] Figure 1 This is a schematic diagram of the automotive parts supplier quality early warning system of the present invention.

[0034] Figure 2 This is a flowchart of the automotive parts supplier quality early warning method of the present invention.

[0035] In the diagram, 1. Parts data acquisition module; 2. Indicator information module; 3. Early warning model module; 4. Three-category classification judgment module; 41. Ranking calculation sub-module; 5. Data analysis and matching module; 6. Risk control measures module; 7. Early warning information push module; 10. Early warning system. Detailed Implementation

[0036] In the description of this application, it should be noted that, unless otherwise expressly specified and limited, the terms "installation," "connection," and "linking" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection, an electrical connection, or a connection that allows communication between them; they can refer to a direct connection or an indirect connection through an intermediate medium; they can refer to the internal communication between two components or the interaction between two components. Those skilled in the art can understand the specific meaning of the above terms in this application according to the specific circumstances.

[0037] To make the technical means, creative features, objectives and effects of the present invention easy to understand, the following embodiments, in conjunction with the accompanying drawings, specifically illustrate the automotive parts supplier quality early warning system 10 and method of the present invention.

[0038] Figure 1 This is a schematic diagram of the structure of the automotive parts supplier quality early warning system 10 of the present invention.

[0039] like Figure 1 As shown, this embodiment provides a quality early warning system 10 for automotive parts suppliers, including: a parts data acquisition module 1, an indicator information module 2, an early warning model module 3, a three-category classification judgment module 4, a data analysis and matching module 5, a risk control measures module 6, and an early warning information push module 7. The output end of the parts data acquisition module 1 is connected to the early warning model module 3. The parts data acquisition module 1, the indicator information module 2, the early warning model module 3, and the three-category classification judgment module 4 are all connected to the data analysis and matching module 5. The output end of the data analysis and matching module 5 is connected to the early warning information push module 7 through the risk control measures module 6.

[0040] The parts data acquisition module 1 connects to the automotive company's supply chain management system to collect real-time quality data of automotive parts from data sources such as the quality management system and supplier portal. This includes the number of QRs (Quality Rejections), the number of ECBs (Early Concern Binnings), pre-sales batch issues, after-sales batch issues, OSM (OnSite Modification) issues, duplicate QRs, unauthorized modifications, and recall events.

[0041] The indicator information module 2 stores the quality dataset based on the parts quality early warning indicator system, and divides the quality data into three levels of indicators according to the degree of risk impact.

[0042] Specifically, the primary indicators are the result indicators categorized according to trend analysis, including QR, pre-sales PPM value, post-sales ECB, number of pre-sales critical issues, number of post-sales critical issues, and number of recall-related events;

[0043] Secondary indicators are process indicators classified according to process capability analysis, including the PPK value and FTT value of critical characteristics of parts;

[0044] Level 3 indicators are other risk indicators, including supplier personnel turnover rate, quality agreement signing status, IATF16949 certificate validity, automation rate of key processes, and traceability rate of key processes.

[0045] Module 3 of the early warning model constructs a hybrid algorithm model of XGBoost, Random Forest, and Bayesian Network as the early warning model. 70% of the quality data in the quality dataset is used as the training dataset for model parameter tuning and iterative training, optimizing the regularization parameters of XGBoost and the number of trees and node splitting thresholds in the Random Forest. The remaining 30% of the quality data is used as the validation dataset, employing accuracy, recall, F1 score, and ROC curve as evaluation metrics to verify the model's early warning accuracy. Simultaneously, it ensures that the early warning model has good generalization ability across different categories of purchased components to meet the enterprise's need for accurate identification of quality risks.

[0046] Specifically, the early warning model is based on a hybrid algorithm model of XGBoost-random Forest-Bayesian Network, which includes three levels.

[0047] The first layer is the feature capture layer, which uses the XGBoost algorithm to capture nonlinear feature relationships in high-quality data using its gradient boosting framework. The objective function of the XGBoost algorithm is:

[0048]

[0049] in, For loss function, is the regularization term, Tk is the number of leaf nodes, and wk is the leaf weight.

[0050] The second layer is the generalization fitting layer, which uses the random forest algorithm. It reduces the risk of overfitting by constructing multiple decision trees and ensembling them through voting. The prediction formula for this random forest algorithm is:

[0051]

[0052] In the formula, K represents the number of decision trees. Let x be the predicted probability of the k-th decision tree for sample x.

[0053] The third layer is the probabilistic inference layer, which employs a Bayesian network algorithm. Based on a conditional probability graphical model, it performs probabilistic inference on the classification results of the three-category classification judgment module 4 and the cumulative ranking results output by the ranking calculation submodule 41, calculating the posterior probability of each supplier being at different risk levels. The inference formula of this Bayesian network algorithm is as follows:

[0054]

[0055] In the formula, For posterior risk probability, Let be the likelihood function. This represents the prior risk probability.

[0056] The outputs from the three levels are weighted and fused to obtain the final risk warning probability. The weighted fusion formula for the outputs from the three levels is as follows:

[0057]

[0058] The weighting coefficients were determined through grid search optimization on the validation dataset, satisfying the following constraints: , .

[0059] The three-level classification judgment module 4 is used to determine the risk level of quality data at multiple levels, including the Class I veto judgment unit, the Class II cumulative trigger judgment unit, and the Class III ranking and screening judgment unit.

[0060] Specifically, the Class I veto judgment unit is used to determine whether a supplier has triggered the veto condition. When a supplier has a pre-sales batch problem (referring to a quality problem involving a batch of parts in a single event), an unauthorized change problem (referring to a supplier changing the product design or process without approval), or a recall event (referring to a market recall caused by a quality problem), the veto rule is triggered, and the supplier is directly judged as a high-risk supplier.

[0061] The determination formula used for Category I veto units is as follows:

[0062]

[0063] When the judgment result is TypeI(i)=1, the supplier is directly judged as high risk, and no further calculation is required.

[0064] The Type II cumulative trigger determination unit is used to determine whether a supplier has reached the cumulative trigger threshold. When the cumulative number of pre-sales batch issues (batch issues discovered before component delivery), OSM issues (On Site Modification, issues discovered during on-site changes), or repeated QR issues (quality rejections that occur repeatedly for the same part or the same issue) of a supplier reaches a preset threshold, the supplier is determined to be a high-risk supplier.

[0065] The determination formula used for Category II veto units is as follows:

[0066] ,

[0067] N_thresh is the cumulative trigger threshold (default value is 3).

[0068] The assessment result is: suppliers whose cumulative quantity reaches the threshold are judged as high-risk.

[0069] The Category III ranking screening and judgment unit is used to identify suppliers whose quality is continuously deteriorating but who have no obvious violations, based on a quantitative ranking model. The Category III ranking screening and judgment unit judges suppliers that have not been judged by Category I and Category II. Its judgment logic is to call the ranking calculation submodule 41 to perform a comprehensive ranking, and then filter based on the cumulative ranking value, taking the top N_top (e.g., TOP10) suppliers with the smallest cumulative ranking value as Category III high-risk suppliers.

[0070] .

[0071] The combined output of the three categories of judgment is:

[0072] ,

[0073] An alert is triggered if a supplier is classified as high-risk by any category.

[0074] The ranking calculation submodule 41 includes an absolute value ranking unit, a trend ranking unit, a comprehensive ranking unit, and a high-risk judgment unit. It is used to calculate the supplier's ECB absolute value ranking, QR absolute value ranking (including JMC final assembly QR ranking and engine QR ranking), ECB trend SLOPE ranking and QR trend SLOPE ranking based on a preset rolling time window, and calculate the supplier's cumulative ranking value based on the above rankings.

[0075] Specifically, the absolute value ranking unit is used to calculate the ranking of each supplier's total ECB count and total QR count based on a 6MOP rolling window. The formula for calculating the absolute value ranking is as follows:

[0076] ,

[0077] In the formula, ECB_total(i) is the total number of ECBs of supplier i within the 6MOP window. The RANK function sorts them in descending order, with the larger the number, the higher the ranking.

[0078] The trend ranking unit is used to calculate the trend slope of each supplier's ECB and QR within a 6MOP window (6 Months Observation Period, which is a rolling time window) using linear regression slope, and to rank the trend slopes.

[0079] Furthermore, the trend slope ranking calculation process for this trend ranking unit is as follows:

[0080]

[0081] Among them, QR JMC,total (i) represents the total number of JMC final assembly QRs for supplier i within the 6MOP window;

[0082]

[0083] Among them, QR ENG,total (i) represents the total number of engine QRs for supplier i within the 6MOP window.

[0084] The formula for calculating the SLOPE function for trend ranking is:

[0085] ,

[0086] in, The ECB average over 6 months, where k is the time index;

[0087] The formula for calculating the ECB trend ranking is:

[0088] ,

[0089] Similarly, the formulas for calculating the QR trend slope and ranking are as follows:

[0090] ,

[0091] .

[0092] The overall ranking unit is used to sum the rankings across all dimensions to obtain a cumulative ranking. The formula for calculating the overall ranking is:

[0093]

[0094] Where C(i) is the cumulative ranking value of supplier i, and the smaller the cumulative ranking value, the higher the overall risk.

[0095] The high-risk determination unit is used to sort suppliers by cumulative ranking from smallest to largest, and the top 10 are selected as high-risk suppliers.

[0096] The data analysis and matching module 5 is used to call the early warning model to analyze the quality data collected by the parts data acquisition module 1. When the quality data triggers the high-risk threshold, it generates early warning information based on the risk level output by the three-category classification judgment module 4, including: risky parts category, related suppliers, risk indicators and risk level.

[0097] The Risk Management Measures Module 6 is used to store improvement measures and re-inspection plans corresponding to various risk types. Combining the supply chain collaborative management concept in the literature, it outputs corresponding supplier rectification requirements, parts re-inspection plans and other improvement suggestions based on early warning information, thereby improving the efficiency of enterprise quality risk handling.

[0098] The early warning information push module 7 sends early warning information and corresponding improvement suggestions to the corresponding STA engineers and supplier liaisons via email, and simultaneously locks down the production and warehousing process of risky batches of parts.

[0099] Figure 2 This is a flowchart of the automotive parts supplier quality early warning method of the present invention.

[0100] like Figure 2 As shown, this embodiment also provides a quality early warning method for automotive parts suppliers, including the following steps S1 to S5.

[0101] S1, Parts Data Acquisition, collects various quality data of automotive parts and suppliers in real time from the automotive company's supply chain management system.

[0102] S2, Indicator information storage, establish a parts quality early warning indicator system, and classify past quality data according to the parts quality early warning indicator system.

[0103] S3. Establish an early warning model. Use 70% of the quality data in the quality dataset as the training dataset and the remaining 30% as the validation dataset. Use accuracy, recall, F1 score and ROC curve as evaluation metrics to establish the early warning model.

[0104] The establishment process of this early warning model is divided into five stages:

[0105] Phase 1: Data Preprocessing and Feature Engineering

[0106] Time window aggregation is performed on the primary indicators to generate 6MOP rolling features: calculate the cumulative value, mean, and month-on-month change rate of each supplier's indicators in the most recent 6 months;

[0107] Standardize the secondary indicators: Use the Z-score standardization method to convert indicators with different dimensions to the same scale;

[0108] The three-level indicators are coded and normalized: one-hot coding is used for categorical indicators, and min-max normalization is used for continuous indicators.

[0109] Phase Two: XGBoost Feature Capture

[0110] Input: Time series characteristics of primary indicators;

[0111] Process: XGBoost builds decision tree ensembles iteratively through gradient boosting, with each tree fitting the residual of the previous tree to capture the non-linear interaction between indicators;

[0112] Output: Ranking of feature importance for each primary indicator and initial risk score;

[0113] Parameter optimization: L1 regularization α controls feature sparsity, L2 regularization λ controls model complexity, and optimization is performed on the training set through grid search.

[0114] Phase 3: Random Forest Generalization Fitting

[0115] Input: Secondary and tertiary indicators, and the initial risk score output by XGBoost as supplementary features;

[0116] Process: Random forest constructs multiple decision trees through bootstrap sampling and random feature selection, with each tree voting independently to reduce variance;

[0117] Output: Overall risk probability for each supplier;

[0118] Parameter optimization: The number of trees (nestimators) controls the ensemble size, and the maximum depth (maxdepth) controls the complexity of a single tree.

[0119] Phase Four: Bayesian Network Probabilistic Inference

[0120] Input: the classification results of the three-category classification judgment module 4, the cumulative ranking results of the ranking calculation submodule 41, the feature importance of XGBoost, and the risk probability of random forest;

[0121] Process: Construct a conditional probability graphical model with each risk indicator as a node and the causal dependencies between indicators as edges. Learn the conditional probability table based on the training data and calculate the posterior probability of each supplier being at different risk levels through Bayesian inference.

[0122] Output: Quantified risk level probability distribution.

[0123] Phase 5, Weighted Fusion:

[0124] The outputs of the three levels are weighted and fused to obtain the final warning probability, and the weight coefficients are optimized by grid search on the validation dataset.

[0125] S4, a three-tiered classification system, assesses the risk level of quality data at multiple levels and calculates and outputs multi-dimensional supplier rankings.

[0126] This embodiment provides a ranking calculation example, which is based on the actual quality data of 104 suppliers of a certain automobile company from June to November 2025 (6MOP window), and shows the specific calculation process of the three-category classification judgment in step S4.

[0127] I. Examples of Class I and Class II determinations are shown in Tables 1 and 2 below:

[0128] Table 1: Criteria for Veto in Category I:

[0129] supplier After-sales bulk Unauthorized alteration recall Class I determination S_A 0 0 0 Not triggered S_B 1 0 0 trigger S_C 0 0 0 Not triggered

[0130] Table 2: Type II Cumulative Trigger Determination:

[0131] supplier Pre-sales bulk OSM Repeat QR Grand total Class II determination S_A 1 1 1 3 trigger S_B - - - - Triggered (Type I) S_C 0 1 0 1 Not triggered

[0132] II. Category III Ranking Screening Determination:

[0133] The following table shows the data of the top 10 suppliers with the smallest cumulative ranking value C(i) among those suppliers who did not trigger Category I and Category II, as shown in Tables 3 and 4 below:

[0134] Table 3: Summary of TOP10 Supplier Rankings by Dimension:

[0135] supplier R_ECB RQRJMC RQRENG RSECB RSQR C(i) S1 5 24 14 2 2 47 S2 5 12 14 19 7 57 S3 4 2 14 29 15 64 S4 15 5 14 16 15 65 S5 8 12 14 25 14 73 S6 15 12 14 20 19 80 S7 15 44 14 2 18 93 S8 2 44 14 33 1 94 S9 29 24 14 7 29 103 S10 32 12 14 10 45 113

[0136] Note: R_ECB = Total ECB Rank (Descending Order), R_QR_JMC = JMC Assembly QR Rank, R_QR_ENG = Engine QR Rank, R_S_ECB = ECB Trend SLOPE Rank, R_S_QR = QR Trend SLOPE Rank.

[0137] Table 4: Key Indicators of Top 10 Suppliers

[0138] supplier ECB total ECB_SLOPE JMC_QR QR_SLOPE Engine QR S1 21 +6.00 4 +3.66 0 S2 21 +22.00 5 +1.97 0 S3 22 +29.00 19 +1.49 0 S4 13 +13.00 8 +1.49 0 S5 19 +20.00 5 +1.54 0 S6 13 +11.00 5 +1.17 0 S7 13 +3.00 3 +1.20 0 S8 43 +24.00 3 +6.40 0 S9 9 +5.00 4 +0.86 0 S10 8 +9.00 5 +0.34 0

[0139] III. Verification of Cumulative Sorting Calculation:

[0140] Let's take S1 as an example to verify the cumulative sorting calculation:

[0141]

[0142] The supplier's total ECB (21) and JMC_QR (4) are at a medium level, but the ECB trend SLOPE=+6.00 (ranked RSECB=2, with the strongest deterioration trend) and QR trend SLOPE=+3.66 (ranked RS_QR=2, with the strongest deterioration trend) result in a cumulative ranking value of 47, which is the lowest among the TOP10 and has the highest overall risk.

[0143] S5, Data Analysis and Matching: The early warning model is called to analyze the quality data and determine whether the quality data triggers a high-risk threshold. If so, the risk level obtained in step S4 is matched with the corresponding risk type in the part quality early warning indicator system to generate early warning information.

[0144] In the example above, although the top 10 suppliers did not trigger Category I and Category II, they had the lowest cumulative ranking value in the Category III ranking screening, and were all judged as Category III high-risk suppliers, triggering the generation of warning information.

[0145] S6, Risk Management Measures: Based on the early warning information, output corresponding improvement suggestions such as supplier rectification requirements and parts re-inspection plans.

[0146] S7 sends early warning information and improvement suggestions to engineers and supplier contacts at each STA, and simultaneously locks down the production and warehousing processes of risky batches of parts.

[0147] The role and effects of the embodiments:

[0148] (1) The automotive parts supplier quality early warning system of the present invention uses the data analysis and matching module 5 to call the early warning model to process the data of the parts data acquisition module 1, and generates early warning information based on the output results of the three-class classification judgment module 4. The risk control measures module 6 simultaneously formulates corresponding control measures, and the early warning information push module 7 pushes emails to STA and suppliers, realizing the implementation mechanism of "data-driven + process closed loop". It deeply integrates model early warning with actual business processes, forming a closed loop of "early warning-response-control", solving the "lag" problem of traditional post-event early warning, and realizing proactive prevention and control of quality risks.

[0149] (2) In the automotive parts supplier quality early warning system of the present invention, the indicator information storage module divides the quality data into three levels of indicators according to the degree of impact, establishes a parts quality early warning indicator system of "three-level hierarchical and result and process dual-dimensional analysis", breaks through the early warning logic of traditional single result indicator, and adapts the corresponding risk type for different level indicators. The early warning information is generated through the data analysis matching module, realizing the transformation from "post-event result traceability" to "process and risk pre-warning".

[0150] (3) The early warning model in the automotive parts supplier quality early warning system and method of the present invention is based on the XGBoost-random forest-Bayes network hybrid algorithm model. Compared with the early warning model with a single algorithm in the industry, the model in the present invention integrates the advantages of multiple algorithms, namely: XGBoost captures nonlinear features, random forest reduces overfitting, and Bayesian network realizes probabilistic risk reasoning, thereby improving the ability to identify potential quality fluctuations of suppliers and reducing the problems of "false alarms" and "missed alarms".

[0151] (4) The automotive parts supplier quality early warning system of the present invention has a built-in risk control measures module 6. Combined with the supply chain collaborative management concept in the literature, it provides suggestions such as supplier rectification requirements and parts re-inspection plan, which can improve the efficiency of enterprise quality risk handling.

[0152] Those skilled in the art should understand that this invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of the invention. Various changes and modifications can be made to this invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the invention as claimed. The scope of protection of this invention is defined by the appended claims and their equivalents.

Claims

1. An automotive parts supplier quality early warning system characterized by, include: The parts data acquisition module is used to collect quality data of automotive parts; The indicator information storage module stores quality datasets based on the parts quality early warning indicator system; The early warning model module includes an early warning model obtained through training and iteration using the aforementioned quality data; The three-tiered judgment module is used to determine the risk level of quality data at multiple levels, including the Class I veto judgment unit, the Class II cumulative trigger judgment unit, and the Class III ranking screening judgment unit and ranking calculation submodule. The data analysis and matching module is used to call the early warning model to analyze the quality data. When the quality data triggers a high-risk threshold, the module outputs the risk level according to the three-category classification judgment module and generates early warning information. The risk management measures module is used to store improvement measures and re-inspection plans corresponding to various risk types; The early warning information push module is used to push the early warning information and corresponding improvement suggestions.

2. The automotive parts supplier quality early warning system according to claim 1, characterized in that: The component quality early warning index system divides quality data into three risk levels: The primary indicators are after-sales fault and quality record indicators categorized according to trend analysis, including QR and after-sales ECB; Secondary indicators are process indicators categorized according to process capability analysis, including repeat QRs, OSM, and pre-sales batch problem indicators; Level 3 indicators are other risk indicators, including issues related to bulk after-sales activities, unauthorized changes, and recall indicators.

3. The automotive parts supplier quality early warning system according to claim 1, Its features are: The early warning model is based on the XGBoost-Random Forest hybrid algorithm and includes three levels: The first layer is the feature capture layer, which uses the XGBoost algorithm to capture the nonlinear feature relationships in the quality data using its gradient boosting framework. The second layer is the generalization fitting layer, which uses the random forest algorithm to reduce the risk of overfitting by constructing multiple decision trees and integrating voting. The third layer is the probabilistic reasoning layer, which uses a Bayesian network algorithm and a conditional probability graphical model to perform probabilistic reasoning on the classification results of the three classification judgment modules and the cumulative ranking results output by the ranking calculation module, and calculates the posterior probability of each supplier being at different risk levels. The outputs from the three levels are weighted and fused to obtain the final risk warning probability.

4. The automotive parts supplier quality early warning system according to claim 1, characterized in that: Among the three types of classification judgment modules, the Type I veto judgment unit is used to determine whether a supplier has triggered a veto condition. When a supplier has pre-sales batch problems, unauthorized changes, or recall events, the veto rule is triggered, and the supplier is directly judged as a high-risk supplier. The Type II cumulative trigger determination unit is used to determine whether a supplier has reached the cumulative trigger threshold. When the cumulative number of pre-sales batch issues, OSM issues, or duplicate QR issues of a supplier reaches the preset threshold, the supplier is determined to be a high-risk supplier. The Category III ranking screening and determination unit is used to identify suppliers whose quality is continuously deteriorating but who have no obvious violations based on a quantitative ranking model. The Category III ranking screening and determination unit calls the ranking calculation submodule to perform a comprehensive ranking and takes the top 10 suppliers with the smallest cumulative ranking value as Category III high-risk suppliers.

5. The automotive parts supplier quality early warning system according to claim 1, Its features are: The ranking calculation submodule includes: The absolute value ranking unit is used to calculate the total ranking of ECB and QR for each supplier within a rolling time window. The trend ranking unit is used to calculate the trend slope of each supplier's after-sales failures and quality records within the rolling time window using linear regression slope, and to rank the trend slope. The comprehensive ranking unit is used to sum the rankings across all dimensions to obtain a cumulative sort. The high-risk determination unit is used to sort suppliers by cumulative ranking from smallest to largest, and the top 10 are selected as high-risk suppliers.

6. A method for quality early warning of automotive parts suppliers, using the automotive parts supplier quality early warning system as described in any one of claims 1-5, characterized in that: Includes the following steps: Parts data acquisition steps: Real-time acquisition of quality data for automotive parts; The indicator information storage step involves storing a quality dataset based on a parts quality early warning indicator system, wherein the quality data in the quality dataset is classified according to the parts quality early warning indicator system. The steps for building the early warning model are as follows: 70% of the quality data in the quality dataset is used as the training dataset, and the remaining 30% of the quality data is used as the validation dataset. Precision, recall, F1 score and ROC curve are used as evaluation indicators to build the early warning model. The three-tiered classification process assesses the risk level of quality data at multiple levels and calculates and outputs a multi-dimensional ranking of suppliers. The data analysis and matching step involves calling the early warning model to analyze the quality data and determine whether the quality data triggers a high-risk threshold. If so, the risk level obtained from the three-category classification judgment step is matched with the corresponding risk type in the part quality early warning index system to generate early warning information. Risk management measures and steps, and output corresponding improvement suggestions based on the early warning information; The early warning information push process involves sending the warning information and improvement suggestions to engineers at each STA and supplier liaisons, and simultaneously locking down the production and warehousing processes of parts in risky batches.

Images