Product development quality management method, device, equipment and storage medium

Abstract

Embodiments of the present application provide a product development quality control method, device, equipment and storage medium. The method comprises: obtaining product configuration information and historical quality information, determining a configuration feature vector corresponding to the product configuration information, and a quality feature vector corresponding to the historical quality information; determining an initial control list according to the matching relationship between the configuration feature vector and the quality feature vector, and the quality control requirement, and sending the initial control list to a matched target object for audit processing; receiving the list audit result of the target object, and determining a target control list according to the list audit result; wherein the target control list includes quality information that needs to be avoided and controlled in the product development process. The method can improve the quality control efficiency, reduce the time cost and labor cost, and improve the product development efficiency and quality.

Description

Technical Field

[0001] This application relates to the field of computer technology, and in particular to a product development quality control method, apparatus, equipment, and storage medium. Background Technology

[0002] In the product development process, the extraction and utilization of historical quality information is a core element in ensuring the quality stability of new products. For example, in the automotive vehicle development process, as automakers increase their platform-based development and accelerate model iteration, the size of historical quality information databases grows exponentially.

[0003] In related technologies, the horizontal adaptation of historical quality information still generally adopts the traditional model of manual screening + offline meeting review.

[0004] This implementation method is time-consuming, inefficient, costly, and has a high error rate, making it difficult to meet the needs of efficient and accurate quality control. Summary of the Invention

[0005] This application provides a product development quality control method, apparatus, equipment, and storage medium, which can improve quality control efficiency, reduce time and labor costs, and improve product development efficiency and quality.

[0006] In a first aspect, embodiments of this application provide a product development quality control method, including:

[0007] Obtain product configuration information and historical quality information, and determine the configuration feature vector corresponding to the product configuration information and the quality feature vector corresponding to the historical quality information;

[0008] Based on the matching relationship between the configuration feature vector and the quality feature vector, and the quality control requirements, an initial control list is determined, and the initial control list is sent to the matched target object for review and processing.

[0009] Receive the review results of the target object's list, and determine the target control list based on the review results; wherein, the target control list includes quality information that needs to be controlled and avoided during the product development process.

[0010] In one possible implementation, the matching relationship between the configuration feature vector and the quality feature vector is determined according to the following steps:

[0011] Based on the preset configuration tower model, extract the higher-order configuration representation vector of the configuration feature vector;

[0012] Based on the preset mass tower model, extract the higher-order mass representation vector of the mass feature vector;

[0013] The matching relationship between the configuration feature vector and the quality feature vector is determined based on the similarity between the higher-order configuration representation vector and the higher-order quality representation vector.

[0014] In one possible implementation, the number of historical quality information entries is multiple; the matching relationship is used to characterize the matching score between each piece of historical quality information entry and the product configuration information; the quality control requirements are used to indicate information filtering rules and information classification rules.

[0015] Based on the matching relationships and quality control requirements, an initial control list is determined, including:

[0016] Based on the matching score and the information filtering rules, the historical quality information is filtered to obtain an information recommendation list; wherein, the information recommendation list includes quality information to be reviewed;

[0017] Based on the professional domain to which each quality information to be reviewed belongs, the quality information to be reviewed is classified and processed to obtain a classification recommendation list;

[0018] The initial control list is determined based on the classification recommendation list.

[0019] In one possible implementation, the information filtering rules indicate hierarchical filtering rules; the information recommendation list includes at least a first recommendation table and a second recommendation table; the method further includes:

[0020] Determine the quality information to be reviewed and the corresponding risk level information included in the second recommendation table;

[0021] If the risk level information meets the preset risk requirements, the quality information to be reviewed included in the second recommendation table will be updated to the first recommendation table.

[0022] In one possible implementation, the initial control list includes at least one quality information to be reviewed; each of the quality information to be reviewed has corresponding grading information and / or classification information;

[0023] Sending the initial control list to the matched target object for review and processing includes:

[0024] Based on the hierarchical information and / or the classification information, the matching target object is determined;

[0025] The quality information to be reviewed is sent to the matched target object for review processing.

[0026] In one possible implementation, receiving the list review results of the target objects includes:

[0027] In response to the information review instruction initiated by the target object, display the information to be reviewed under the preset review dimensions;

[0028] In response to the target object's editing operation on the information to be reviewed, the checklist review result is determined; wherein, the checklist review result is used to indicate whether the quality information included in the initial control checklist can be used for the quality control of the current product.

[0029] In one possible implementation, the method further includes:

[0030] If the review result of the list indicates that the quality information included in the initial control list cannot be used for the quality control of the current product, then the information processing method is determined according to the information type of the information to be reviewed that does not match the current product;

[0031] Update historical quality information according to the aforementioned information processing method.

[0032] Secondly, embodiments of this application provide a product development quality control device, comprising:

[0033] The acquisition unit is used to acquire product configuration information and historical quality information, and determine the configuration feature vector corresponding to the product configuration information and the quality feature vector corresponding to the historical quality information.

[0034] The first determining unit is used to determine an initial control list based on the matching relationship between the configuration feature vector and the quality feature vector, and the quality control requirements, and to send the initial control list to the matched target object for review processing;

[0035] The second determining unit is used to receive the list review results of the target object and determine the target control list based on the list review results; wherein the target control list includes quality information that needs to be avoided and controlled during the product development process.

[0036] In one possible implementation, the matching relationship between the configuration feature vector and the quality feature vector is determined according to the following steps:

[0037] Based on the preset configuration tower model, extract the higher-order configuration representation vector of the configuration feature vector;

[0038] Based on the preset mass tower model, extract the higher-order mass representation vector of the mass feature vector;

[0039] The matching relationship between the configuration feature vector and the quality feature vector is determined based on the similarity between the higher-order configuration representation vector and the higher-order quality representation vector.

[0040] In one possible implementation, the number of historical quality information items is multiple; the matching relationship is used to characterize the matching score between each piece of historical quality information and the product configuration information; the quality control requirements are used to indicate information filtering rules and information classification rules; at this time, the first determining unit is used to:

[0041] Based on the matching score and the information filtering rules, the historical quality information is filtered to obtain an information recommendation list; wherein, the information recommendation list includes quality information to be reviewed;

[0042] Based on the professional domain to which each quality information to be reviewed belongs, the quality information to be reviewed is classified and processed to obtain a classification recommendation list;

[0043] The initial control list is determined based on the classification recommendation list.

[0044] In one possible implementation, the information filtering rules indicate hierarchical filtering rules; the information recommendation list includes at least a first recommendation table and a second recommendation table; in this case, the device is also used for:

[0045] Determine the quality information to be reviewed and the corresponding risk level information included in the second recommendation table;

[0046] If the risk level information meets the preset risk requirements, the quality information to be reviewed included in the second recommendation table will be updated to the first recommendation table.

[0047] In one possible implementation, the initial control list includes at least one quality information item to be reviewed; each of the quality information items to be reviewed has corresponding hierarchical information and / or classification information; in this case, the first determining unit is used to:

[0048] Based on the hierarchical information and / or the classification information, the matching target object is determined;

[0049] The quality information to be reviewed is sent to the matched target object for review processing.

[0050] In one possible implementation, the second determining unit is configured to:

[0051] In response to the information review instruction initiated by the target object, display the information to be reviewed under the preset review dimensions;

[0052] In response to the target object's editing operation on the information to be reviewed, the checklist review result is determined; wherein, the checklist review result is used to indicate whether the quality information included in the initial control checklist can be used for the quality control of the current product.

[0053] In one possible implementation, the device is also used for:

[0054] If the review result of the list indicates that the quality information included in the initial control list cannot be used for the quality control of the current product, then the information processing method is determined according to the information type of the information to be reviewed that does not match the current product;

[0055] Update historical quality information according to the aforementioned information processing method.

[0056] Thirdly, embodiments of this application provide a computer device, including: a memory and a processor;

[0057] The memory stores computer-executed instructions;

[0058] The processor executes computer execution instructions stored in the memory, causing the processor to perform the first aspect and / or various possible implementations of the first aspect as described above.

[0059] Fourthly, embodiments of this application provide a computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, are used to implement the first aspect and / or various possible implementations of the first aspect.

[0060] Fifthly, embodiments of this application provide a computer program product, including a computer program that, when executed by a processor, implements the first aspect and / or various possible implementations of the first aspect.

[0061] The product development quality control method, apparatus, equipment, and storage medium provided in this application, after obtaining product configuration information and historical quality information, can first determine the configuration feature vector corresponding to the product configuration information and the quality feature vector corresponding to the historical quality information. Then, based on the matching relationship between the configuration feature vector and the quality feature vector and the quality control requirements, an initial control list is determined. At this point, the initial control list can be obtained through information standardization, feature vector alignment, and intelligent matching, thereby replacing the manual screening process. This reduces the time spent screening single historical quality information from 5 minutes to 2 seconds, improving efficiency by 60 times, thus shortening the product development cycle and improving product development efficiency. Furthermore, the method of obtaining the initial control list based on matching relationships and quality control requirements is more efficient, stable, and accurate than manual screening. Afterwards, the initial control list can be sent to the matched target object for review, allowing for target object verification instead of multi-team centralized meetings. This reduces ineffective cross-departmental collaboration, lowers the redundancy of cross-team collaboration, and shortens the judgment cycle. Finally, the system can receive the review results of the target object's checklist and determine the target control checklist based on these results. This target control checklist includes quality information that needs to be controlled to avoid issues during product development. This implementation method enables the determination of an accurate and reliable target control checklist based on the review results of the target object's checklist, thereby improving the product's quality control capabilities and ultimately enhancing the product's development quality. Attached Figure Description

[0062] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0063] Figure 1 A flowchart illustrating a product development quality control method provided in this application embodiment;

[0064] Figure 2 A flowchart illustrating another product development quality control method provided in this application embodiment;

[0065] Figure 3 A schematic diagram illustrating the implementation process of a product development quality control method provided in this application embodiment;

[0066] Figure 4 A schematic diagram of a product development quality control system provided in this application embodiment;

[0067] Figure 5 A schematic diagram of a product development quality control device provided in this application embodiment;

[0068] Figure 6This is a schematic diagram of the structure of a computer device provided in an embodiment of this application.

[0069] The accompanying drawings illustrate specific embodiments of this application, which will be described in more detail below. These drawings and descriptions are not intended to limit the scope of the concept in any way, but rather to illustrate the concept of this application to those skilled in the art through reference to particular embodiments. Detailed Implementation

[0070] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numbers in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this application. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this application as detailed in the appended claims.

[0071] First, let me explain the terms used in this application:

[0072] PLM: Product Lifecycle Management;

[0073] QMS: Quality Management System;

[0074] MES: Manufacturing Execution System;

[0075] DMS: Dealer Management System;

[0076] API: Application Programming Interface;

[0077] BOM: Bill of Materials;

[0078] EBOM: Engineering Bill of Materials;

[0079] MBOM: Manufacturing BOM (Bill of Materials)

[0080] PBOM: Plan BOM, Planned Bill of Materials;

[0081] FMEA: Failure Mode and Effects Analysis;

[0082] NER: Named Entity Recognition;

[0083] BERT: Bidirectional Encoder Representations from Transformers;

[0084] FAISS: Facebook AI Similarity Search, is a vector search engine.

[0085] APQP: Advanced Product Quality Planning.

[0086] In the product development process, the extraction and utilization of historical quality information is a core element in ensuring the quality stability of new products. For example, in the automotive vehicle development process, as automakers increase their platform-based development and accelerate model iteration, the size of historical quality information databases grows exponentially.

[0087] In related technologies, the horizontal adaptation of historical quality information still generally adopts the traditional model of manual screening + offline meeting review. The core process is as follows: the quality manager needs to manually retrieve massive amounts of historical quality information across systems (such as PLM product lifecycle management system, QMS quality management system, MES manufacturing execution system, DMS dealer management system, and after-sales claims system, etc.). For example, the historical quality information of leading companies can cover more than 10 years and more than 100,000 data points of 50+ mass-produced models. Then, based on personal experience, they check the relevance of each data point to the configuration of the current new model. After completing the initial screening, they organize an offline review meeting with multiple professional teams such as R&D, production, testing, after-sales, and suppliers to jointly determine the applicability of historical quality issues in each piece of historical quality information to the current model, as well as the continuation and optimization of the avoidance strategies corresponding to each historical quality issue in the historical quality information.

[0088] The above-described embodiments have the following technical problems:

[0089] 1. Manual screening is inefficient and prone to errors, which affects the development cycle and quality of products / projects.

[0090] Faced with massive amounts of historical quality information, a quality manager can spend an average of up to 5 minutes filtering a single piece of quality information. A single vehicle model project needs to complete the initial screening of nearly a thousand pieces of historical quality information. In this case, the initial screening stage alone requires the involvement of two quality managers. The full-time working hours in a month affect the development cycle of products / projects.

[0091] At the same time, manual screening relies on personal experience. Therefore, different quality managers may obtain different screening results, which can easily lead to misjudgments (e.g., redundant invalid issues) and omissions (e.g., missing high-risk issues). This can affect the accuracy of the initial screening of historical quality information and further affect the development quality of products / projects.

[0092] 2. Cross-team collaboration has high redundancy, long judgment cycle, and prominent risks of misjudgment and compliance.

[0093] In the traditional model, it is necessary to rely on multiple professional teams to conduct offline centralized reviews to complete the suitability assessment. The complete assessment cycle for a single piece of historical quality information can take as long as 3-5 working days. Furthermore, a single vehicle model project requires more than 5 cross-departmental special review meetings. Engineers from various professional fields need to repeatedly invest a lot of time in ineffective discussions, resulting in extremely high collaboration costs.

[0094] Meanwhile, due to information asymmetry across teams and the lack of a single judgment standard, misjudgments of suitability and omissions of high-risk safety items occurred, leading to the recurrence of similar quality problems after the new models were mass-produced, resulting in huge after-sales claims losses. In addition, the process of offline review, judgment criteria, and modification records are difficult to keep traceable throughout, making it difficult to meet the mandatory requirements of the IATF 16949 system for full-process traceability of quality control, and posing compliance risks.

[0095] 3. There is a waste of resources, and the proportion of ineffective work is high.

[0096] Industry statistics show that approximately 80% of the historical quality information initially screened by manual review for cross-model adaptation is ultimately discarded due to incompatibility with the current vehicle model, duplicate judgments, or lack of necessary control. This renders a significant amount of time invested by quality managers and engineers ineffective, resulting in a double loss of human and time costs. Furthermore, with the compressed development cycle of new vehicle models, engineers are required to complete a large amount of ineffective screening and review work in a short period, making it difficult to dedicate their energy to high-value quality control tasks such as root cause analysis and solution optimization, leading to a decline in core quality control capabilities.

[0097] Research has revealed that while some technologies can perform initial screening using keyword searches based on historical quality information, this approach results in low accuracy and makes it difficult to replace manual screening. Other technologies can perform initial screening through single-stage text similarity matching; however, this approach is ill-suited for scenarios with complex historical quality information, and relying solely on text similarity matching is prone to misjudgments, making it difficult to implement in the actual business processes of vehicle development.

[0098] Therefore, there is an urgent need for an intelligent and automated end-to-end solution to achieve accurate matching, efficient review, and compliance management of historical quality information, thereby adapting to the rapid iteration development needs of the automotive industry.

[0099] The product development quality control method provided in this application extracts feature vectors to match new products with historical quality information. Combined with quality control requirements, it automatically performs preliminary screening of historical quality information, thereby improving the efficiency, accuracy, and stability of historical quality information screening. Then, the preliminarily screened quality information is automatically sent to the corresponding target for review, enabling rapid and accurate judgment of the suitability of historical quality information, thus improving the efficiency of historical quality information judgment and shortening the judgment cycle to ensure product development cycle and development quality. Furthermore, the embodiments of this application can also update historical quality information based on the checklist review results, thereby achieving closed-loop management and experience reuse throughout the entire process. Further, it can also meet the requirements of full-process traceability by implementing full-process traceability management for automated processing, thus satisfying the compliance control requirements of the automotive industry.

[0100] The technical solution of this application and how the technical solution of this application solves the above-mentioned technical problems are described in detail below with specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments. The embodiments of this application will now be described with reference to the accompanying drawings.

[0101] Figure 1 This application provides a flowchart illustrating a product development quality control method, as shown in the embodiments below. Figure 1 As shown, the method includes:

[0102] S101. Obtain product configuration information and historical quality information, and determine the configuration feature vector corresponding to the product configuration information and the quality feature vector corresponding to the historical quality information.

[0103] This section uses a vehicle as an example to illustrate the specific implementation of the product development quality control method provided in this application.

[0104] Based on this, product configuration information can indicate vehicle model configuration information. In this case, the acquired product configuration information can be the vehicle model configuration information, and the acquired historical quality information can be the historical quality information of the vehicle recorded by the automaker.

[0105] Optionally, in this embodiment, a standardized API interface can be seamlessly integrated with the vehicle manufacturer's PLM system, BOM management system, and vehicle configuration management system to automatically retrieve the full EBOM and MBOM data of the current new model project, and automatically identify core configuration information according to preset configuration extraction rules to obtain the model configuration information.

[0106] Optionally, in this embodiment, product configuration information can also be obtained through visual manual input. For example, this embodiment can use a provided visual configuration information input interface, categorized into four levels: "vehicle domain - system domain - subsystem domain - component domain," with a pre-defined unified configuration option library for automakers. Then, by responding to drop-down selection operations and batch import operations by staff (e.g., quality managers), the vehicle configuration information can be entered, thereby obtaining the product configuration information. Furthermore, the completeness and standardization of the entered vehicle configuration information can be automatically verified, thereby improving the quality of the vehicle configuration information.

[0107] At this time, the vehicle configuration information obtained may include, but is not limited to: vehicle platform architecture, power type, chassis architecture, electronic and electrical architecture, core electric system configuration, core component models, production process path, welding / painting / final assembly process scheme, target market, application scenarios, regulatory requirements, target customer groups, etc.

[0108] After obtaining the product configuration information, the configuration feature vector corresponding to the product configuration information can be obtained by performing entity linking and feature encoding on the product configuration information.

[0109] In practical implementation, based on a pre-built vehicle configuration ontology library, entity linking and feature encoding can be performed on the cleaned standardized configuration information. The vehicle configuration ontology library is a standardized ontology library built based on the automotive industry standard GB / T 3730.1 and the internal configuration classification system of car companies, covering the entity, attribute, and relationship definitions of the entire vehicle configuration domain. At this point, core configuration features can be extracted according to seven dimensions: "platform-system-subsystem-component-process-scenario-regulation". Based on historical project data, corresponding weight coefficients are set for each dimension (with the platform, component, and process dimensions having a weight of no less than 60%). Then, through a pre-trained BERT encoding model, feature encoding is performed to transform the multi-dimensional configuration features into a fixed-dimensional dense numerical vector, i.e., the configuration feature vector. At the same time, a corresponding structured configuration feature label set is generated, providing a unified vector space foundation for subsequent matching.

[0110] In one possible implementation, after obtaining vehicle configuration information, the obtained vehicle configuration information can be cleaned to ensure that the data is consistent, free of redundancy and error, thereby improving the quality of the obtained vehicle configuration information.

[0111] Specifically, data cleaning can include, but is not limited to: removing redundant information and invalid characters, standardizing naming conventions and terminology, correcting typos and non-standard expressions, completing missing key fields, and removing duplicate configuration information. For example, "800V high-voltage platform" and "800V high-voltage electrical architecture" can be cleaned to "800V high-voltage electrical platform"; or, similar components from different suppliers can be uniformly categorized into corresponding component domains to ensure the standardization and consistency of configuration information.

[0112] In one example, as described in this embodiment, a standardized API interface can be used to connect to the entire range of quality-related systems within an automaker (e.g., quality-related systems include, but are not limited to, QMS quality management system, PLM product lifecycle management system, MES manufacturing execution system, DMS dealer management system, after-sales claims system, supplier management system, and FMEA system, etc.) to automatically acquire historical quality information. The acquired historical quality information can include data from the entire process, including vehicle R&D, component development, manufacturing, after-sales market, supplier management, and compliance certification.

[0113] Optionally, the carriers of historical quality information may include, but are not limited to: 8D problem resolution reports, FMEA analysis documents, fault analysis reports, rectification notices, after-sales work orders, claim forms, customer complaint records, supplier quality rectification reports, and regulatory compliance rectification records.

[0114] Based on this, after obtaining historical quality information, natural language processing technology can be used to analyze the historical quality information, thereby extracting standardized core information from the historical quality information and improving the quality of the historical quality information.

[0115] Specifically, core entity information can be extracted from historical quality information based on pre-trained automotive domain named entity recognition models (e.g., NER models); relationships between entities can be extracted based on relation extraction techniques; and core fault information and root cause information can be extracted based on keyword extraction and topic models.

[0116] At this time, the historical quality information obtained may include, but is not limited to: unique problem identifier, vehicle model / platform, time of occurrence, stage of occurrence, associated BOM level, associated component code / name, failure mode, failure phenomenon, root cause classification, root cause analysis, corrective measures, avoidance strategies, verification results, applicable boundaries, risk level, responsible department, closure status, and compliance attributes.

[0117] After obtaining historical quality information, the corresponding quality feature vector can be obtained by constructing a multi-dimensional feature label system and performing feature vector transformation on the historical quality information.

[0118] In practical implementation, a multi-dimensional and quantifiable feature tag set can be generated for each structured historical problem data based on a pre-built automotive industry quality control knowledge ontology, with each tag assigned a corresponding weight coefficient. Specific tag dimensions include: basic attribute tags: vehicle model, platform architecture, production year, mass production status; configuration matching tags: associated systems, associated parts, process paths, architecture type; scenario tags: R&D stage, manufacturing stage, aftermarket stage, supplier stage; root cause tags: design root cause, manufacturing root cause, supplier root cause, material root cause, human root cause; risk level tags: based on FMEA SOD scores (severity, occurrence, detectability), categorized as low risk, medium risk, high risk, and extremely high risk; and compliance tags: whether safety items, mandatory regulatory items, or mandatory customer requirements are involved.

[0119] Subsequently, the feature tag set of each historical question can be transformed into a quality feature vector with the same dimension and vector space as the configured feature vector through the BERT encoding model, ensuring the effectiveness and accuracy of subsequent similarity calculations. Simultaneously, the FAISS vector database can be used to create a high-performance index for each quality feature vector, constructing a historical quality information feature library that supports millisecond-level retrieval of millions of vectors. This addresses the retrieval performance bottleneck of massive historical question data, ensuring that the response time for full-data matching retrieval is controlled within 2 seconds.

[0120] S102. Based on the matching relationship between the configuration feature vector and the quality feature vector, and the quality control requirements, determine the initial control list, and send the initial control list to the matched target object for review and processing.

[0121] In one example, the matching relationship between configuration feature vectors and quality feature vectors can be used to indicate the similarity between configuration feature vectors and quality feature vectors, or it can be used to indicate the overlap between configuration feature vectors and quality feature vectors, etc. Here, the method of determining the matching relationship between configuration feature vectors and quality feature vectors is not limited, and the actual needs shall prevail.

[0122] In one example, quality control requirements can be used to indicate the business rules for quality control in the automotive industry. In this case, historical quality information can be filtered according to the quality control requirements to obtain an initial control list.

[0123] In one example, the target object can be a professional engineer responsible for manual review. In this case, there can be one or more target objects.

[0124] S103. Receive the review results of the target object list and determine the target control list based on the review results; the target control list includes quality information that needs to be controlled during the product development process.

[0125] In one example, after the initial control list is sent to the target, at least one audit item can be displayed through a visual information audit interface, and the list audit result can be obtained by receiving the audit result of the target on each audit item.

[0126] In one example, the audit results of the list can be used to filter historical quality information in the initial control list to obtain the target control list.

[0127] Optionally, the target control list may include, but is not limited to: a unique problem number, failure mode, risk level, reason for adaptation, avoidance strategy, responsible department, completion deadline, and verification node.

[0128] Optionally, the target control list can be directly imported into the APQP control system, FMEA system, project management system, and QMS system to achieve seamless integration with existing R&D quality control processes. No manual secondary entry is required, and it can be directly implemented in the quality control work of new model development.

[0129] As described above, in this embodiment, after obtaining product configuration information and historical quality information, the configuration feature vector corresponding to the product configuration information and the quality feature vector corresponding to the historical quality information can be determined first. Then, based on the matching relationship between the configuration feature vector and the quality feature vector, and the quality control requirements, an initial control list is determined. At this point, the initial control list can be obtained through information standardization, feature vector alignment, and intelligent matching, thereby replacing the manual screening process. This reduces the time spent screening single historical quality information from 5 minutes to 2 seconds, improving efficiency by 60 times, thus shortening the product development cycle and increasing product development efficiency. Furthermore, compared to manual screening, the method of obtaining the initial control list based on matching relationships and quality control requirements is more efficient, stable, and accurate. Afterward, the initial control list can be sent to the matched target object for review, allowing for target object verification instead of multi-team centralized meetings. This reduces ineffective cross-departmental collaboration, lowers the redundancy of cross-team collaboration, and shortens the judgment cycle. Finally, the system can receive the review results of the target object's checklist and determine the target control checklist based on these results. This target control checklist includes quality information that needs to be controlled to avoid issues during product development. This implementation method enables the determination of an accurate and reliable target control checklist based on the review results of the target object's checklist, thereby improving the product's quality control capabilities and ultimately enhancing the product's development quality.

[0130] Figure 2 A flowchart illustrating another product development quality control method provided in this application embodiment is shown below. Figure 2 As shown, in this embodiment... Figure 1 Based on the examples, the product development quality control method is described in detail, which includes:

[0131] S201. Obtain product configuration information and historical quality information, and determine the configuration feature vector corresponding to the product configuration information and the quality feature vector corresponding to the historical quality information.

[0132] In one example, this step can be referred to the content described in S101 above, and will not be repeated in detail here.

[0133] In this embodiment of the application, the matching relationship between the configuration feature vector and the quality feature vector can be determined based on the similarity between them. Specifically, the matching relationship between the configuration feature vector and the quality feature vector can be determined based on a pre-trained dual-tower deep matching model, as described in the process below.

[0134] S202. Based on the preset configuration tower model, extract the higher-order configuration representation vector of the configuration feature vector.

[0135] S203. Based on the preset quality tower model, extract the higher-order quality representation vector of the quality feature vector.

[0136] S204. Based on the similarity between the higher-order configuration representation vector and the higher-order quality representation vector, determine the matching relationship between the configuration feature vector and the quality feature vector.

[0137] In one example, a pre-trained dual-tower deep matching model may include a pre-defined configuration tower model, a pre-defined quality tower model, and a top-level model. The pre-defined configuration tower model and the pre-defined quality tower model can be two sub-networks sharing a common underlying encoding layer; for example, they can be convolutional neural networks or multi-layer deep neural networks. The top-level model is used to calculate the similarity between high-order configuration representation vectors and high-order quality representation vectors; exemplarily, this similarity can be a pre-defined similarity.

[0138] In one example, when the top layer of the model calculates the similarity between the higher-order configuration representation vector and the higher-order quality representation vector, it can use the FAISS vector database to perform a full vector retrieval of the higher-order quality representation vector, thereby calculating the cosine similarity between the higher-order configuration representation vector and all historical quality information.

[0139] In one example, the matching relationship between the configuration feature vector and the quality feature vector can indicate a matching relationship or a non-matching relationship; or, it can indicate the matching score between the configuration feature vector and the quality feature vector, in which case the matching score can be between 0 and 100.

[0140] In the above implementation, the configuration feature vector and the quality feature vector can be aligned in a unified vector space based on the pre-trained dual-tower deep matching model with a shared coding layer design, thereby reducing the deviation of heterogeneous data matching and providing reliable data support for the calculation of matching relationships. Then, the matching relationship between the configuration feature vector and the quality feature vector is determined according to the similarity calculation method, so that the matching relationship has continuity and interpretability.

[0141] In one possible implementation, the pre-training and optimization process of the dual-tower deep matching model specifically includes the following steps:

[0142] Step 1: Constructing the training dataset.

[0143] A training dataset is constructed based on historical problem-solving annotation data from mass-produced vehicle projects (e.g., projects from the past 3 years). This training dataset includes complete data on vehicle configuration information and historical quality information, as well as manually annotated suitability results (e.g., suitable / unsuitable). Optionally, the amount of annotated data should be no less than 500,000 records to meet model training requirements. The training dataset can then be divided into training, validation, and test sets in an 8:1:1 ratio.

[0144] Step 2: Data augmentation.

[0145] Data augmentation techniques such as synonym replacement, feature masking, label smoothing, and random sampling of negative samples are used to expand the training set, improve the model's generalization ability, and reduce overfitting.

[0146] Step 3: Model training and parameter optimization.

[0147] The Adam optimizer is used, the cross-entropy loss function is adopted, the batch size is set to 64, the initial learning rate is 1e-5, the training epochs are 30, and an early stopping mechanism (training stops if the validation set loss does not decrease for 3 consecutive epochs) is used to prevent overfitting.

[0148] Step 4: Verify the model's effectiveness.

[0149] After training, the model's performance is verified using a test set. The model is required to achieve a Top100 matching accuracy of ≥85% and a Top500 matching accuracy of ≥90% to meet business implementation requirements.

[0150] In one possible implementation, the dual-tower deep matching model provided in this application supports incremental training, that is, the model parameters can be continuously optimized based on newly added project data, thereby continuously improving the matching effect of the model.

[0151] In one possible implementation, there are multiple historical quality information entries; the matching relationship is used to characterize the matching score between each historical quality information entry and the product configuration information.

[0152] At this point, when determining the initial control list based on quality control requirements and matching relationships, historical quality information that meets accuracy requirements can be identified through these requirements. Simultaneously, this historical quality information is categorized according to the quality control requirements to ensure that the determined initial control list can be accurately sent to the matched target objects for review and processing. In this context, the quality control requirements serve as the guidelines for information filtering and classification rules.

[0153] Based on this, according to the matching relationship between the configuration feature vector and the quality feature vector, and the quality control requirements, an initial control list is determined, which specifically includes the following process:

[0154] S205. Based on the matching score and information filtering rules, the historical quality information is filtered to obtain an information recommendation list; the information recommendation list includes quality information to be reviewed.

[0155] S206. Based on the professional domain to which each quality information to be reviewed belongs, classify the quality information to be reviewed to obtain a classification recommendation list.

[0156] S207. Determine the initial control list based on the classification recommendation list.

[0157] In one example, information filtering rules are used to indicate the rules for filtering matching scores.

[0158] Optionally, the information filtering rules can indicate a matching score filtering threshold. In this case, the matching scores can be filtered according to the information filtering rules to obtain historical quality information that meets the matching score filtering threshold, which is the quality information to be reviewed, and thus obtain the information recommendation list.

[0159] For example, the number of matching score filtering thresholds indicated by the information filtering rules can be one, two, or three, etc. There is no limit to the number of matching score filtering thresholds here, and it is based on actual needs.

[0160] In one example, the professional domain to which the quality information to be reviewed belongs can indicate the product's supply chain. For example, the professional domain can include, but is not limited to: body, chassis, electric drive system, electronics, interior and exterior trim, manufacturing, after-sales service, and suppliers.

[0161] As described above, the embodiments of this application can filter out historical quality information with a high degree of matching to product configuration information, i.e., quality information to be reviewed, according to information filtering rules. This filters out invalid historical quality information and reduces the workload of manual review. Simultaneously, the filtered quality information to be reviewed can be categorized according to information classification rules, which helps to process the quality information to be reviewed by specific professionals, thereby achieving targeted review of the quality information to be reviewed.

[0162] In one possible implementation, if there are multiple matching score filtering thresholds indicated by the information filtering rules, then multiple information recommendation tables can be determined.

[0163] Based on this, if the number of matching score filtering thresholds indicated by the information filtering rule is two, then the information filtering rule indicates a hierarchical filtering rule. In this case, the information recommendation list includes at least a first recommendation table and a second recommendation table.

[0164] Specifically, if the matching score screening thresholds are 80 and 60, then historical quality information with a matching score greater than or equal to 80 can be used to determine the first recommendation list, and historical quality information with a matching score greater than or equal to 60 and less than 80 can be used to determine the second recommendation list. At this time, the quality information to be reviewed in the initial control list can be graded through the first recommendation list and the second recommendation list. For example, the review intensity of the first recommendation list is stronger than that of the second recommendation list, so as to improve the review efficiency and accuracy of the quality information to be reviewed.

[0165] Furthermore, in order to enhance the control intensity of high-risk control items (that is, high-risk quality information to be reviewed), in the embodiments of the present application, it is also possible to determine the risk level information corresponding to the quality information to be reviewed included in the second recommendation list. If the risk level information meets the preset risk requirements, the quality information to be reviewed included in the second recommendation list is updated to the first recommendation list.

[0166] In one example, the preset risk requirements may indicate but are not limited to: safety items, legally mandatory items, and quality information to be reviewed with an extremely high risk level.

[0167] At this time, by reclassifying the high-risk quality information to be reviewed in the second recommendation list into the first recommendation list, the review intensity of the high-risk quality information to be reviewed can be improved, thereby reducing the development risk of the current product.

[0168] In a possible implementation manner, after determining the initial control list, a multi-dimensional matching result visualization report can also be generated according to the initial control list. At this time, the report may include but is not limited to: the total number of recommended quality information, distribution in each professional domain, risk level distribution, matching score distribution, special list of high-risk information, details of the quality information to be reviewed, etc. At this time, the visualization report supports one-click export in Excel / PDF format, which is convenient for the quality manager to quickly preview and control.

[0169] In one example, the special list of high-risk information indicates issues related to safety items, legally mandatory items, and extremely high risk levels. At this time, the special list of high-risk information may include but is not limited to risk conduction analysis, control requirements, verification plans, and emergency response plans, etc. At this time, the special list of high-risk information can be used for project milestone reviews and high-level risk control, so as to ensure the quality of product development.

[0170] In a possible implementation manner, the initial control list includes at least one quality information to be reviewed; at this time, each quality information to be reviewed has corresponding grading information and / or classification information.

[0171] In one example, the grading information can be determined based on the type of recommendation table, and the classification information can indicate the professional domain to which the quality information to be reviewed belongs.

[0172] At this point, when sending the initial control list to the matching target object for review, the matching target object can be determined first based on the hierarchical information and / or classification information. Then, the quality information to be reviewed can be sent to the matching target object for review.

[0173] For example, if the quality information to be reviewed is a first recommendation table, then the classification information can be level one, and the target objects to be matched can include responsible engineers and supervising engineers; if the quality information to be reviewed is a second recommendation table, then the classification information can be level two, and the target objects to be matched can include senior engineers, responsible engineers and supervising engineers.

[0174] For example, the target object matched with the classification information can indicate the engineer within the professional domain to which the quality information to be reviewed belongs.

[0175] Based on this, if the quality information to be reviewed is the first recommended table, and the professional domain indicated by the classification information is the vehicle body domain, then the target objects for matching can include the responsible engineer of the vehicle body domain and the supervising engineer of the vehicle body domain.

[0176] In this embodiment, strict fine-grained access control can be set up, so that each engineer can only view and edit quality information in their own professional domain, and cannot modify content across professional domains. They can only submit collaborative opinions on quality information outside their own professional domain, so as to ensure data security, clear responsibilities, and well-defined boundaries.

[0177] In the above implementation, the target objects can be determined by hierarchical information and / or classification information, thereby enabling the targeted distribution of quality information to be reviewed in the initial control list. This not only improves the efficiency and professionalism of the review process, but also reduces ineffective cross-departmental collaboration, shortens the judgment cycle, and reduces the misjudgment rate.

[0178] In one possible implementation, before sending the initial control list to the matching target object, this embodiment of the application may further perform deduplication and merging processing on the quality information to be reviewed included in the initial control list. For example, for duplicate quality information to be reviewed for the same failure mode, the same root cause, and the same component, only the latest quality information with the most complete rectification plan and the best verification effect is retained, thereby reducing the problems of duplicate sending / recommendation and duplicate review.

[0179] In one possible implementation, after sending the initial control list to the matched target, the target can receive the list review results. To ensure consistency in judgment standards across different professional domains and among different engineers, thereby reducing the misjudgment rate due to differences in individual experience, the target can be assisted in the review process by pre-setting multiple review dimensions (i.e., preset review dimensions) for the information to be reviewed.

[0180] Based on this, the results of the review of the list of target objects are received, which may specifically include the process described in S208 to S209 below.

[0181] S208. In response to the information review instruction initiated by the target object, display the information to be reviewed under the preset review dimensions.

[0182] S209. In response to the editing operation of the target object on the information to be reviewed, determine the list review result; wherein, the list review result is used to indicate whether the quality information included in the initial control list can be used for the quality control of the current product.

[0183] In one example, the preset review dimensions may include, but are not limited to: configuration adaptability dimension, strategy applicability dimension, risk necessity dimension, boundary integrity dimension, and compliance dimension.

[0184] Among them, the information to be reviewed under the configuration adaptability dimension is used to indicate whether the platform, architecture, parts, and process of the current vehicle model are consistent with the carrier of the quality information to be reviewed, and whether there are the same fault causes.

[0185] The information to be reviewed under the strategy applicability dimension is used to indicate whether the avoidance and rectification strategy of the quality information to be reviewed is applicable to the current model and whether it needs to be adjusted and optimized.

[0186] The information to be reviewed under the risk necessity dimension is used to indicate whether the quality information issue to be reviewed needs to be pre-controlled in the current vehicle model project, and whether there is a clear quality risk.

[0187] The information to be reviewed under the boundary integrity dimension is used to indicate whether the applicable boundary definition of the quality information to be reviewed is accurate and whether it needs to be supplemented or corrected.

[0188] The information to be reviewed under the compliance dimension is used to indicate whether the control of the quality information to be reviewed meets regulatory requirements and mandatory customer requirements.

[0189] Optionally, the editing operations for the information to be reviewed may include: selecting whether it is compatible or not, and confirming the basis for review. For example, the selection of whether it is compatible or not can be used to select "consistent" or "incompatible"; the confirmation of the basis for review can be used to fill in or confirm the basis for review.

[0190] In one possible implementation, the embodiments of this application may also set corresponding review time requirements for quality information to be reviewed at different risk levels, so as to further improve review efficiency.

[0191] Specifically, the review period for quality information pending review that is of extremely high risk is 1 working day, the review period for quality information pending review that is of high risk is 2 working days, and the review period for quality information pending review that is of medium to low risk is 3 working days.

[0192] In one possible implementation, the embodiments of this application may also set corresponding review time requirements for each target object, thereby urging the target object to complete the review as soon as possible to improve review efficiency. For example, for senior engineers, the corresponding review time can be set to 1 working day; for responsible engineers, the corresponding review time can be set to 1 working day; for supervisory engineers, the corresponding review time can be set to 2 working days, etc.

[0193] Optionally, for review tasks that are not completed within the time limit, a reminder message can be sent to the corresponding target object through at least one communication method (e.g., email, SMS, in-system message, etc.) to ensure that the review process is closed on time.

[0194] Optionally, in order to meet the IATF16949 system requirements of the automotive industry, this application embodiment can design a full-process traceability mechanism to meet the compliance control requirements of car companies.

[0195] Specifically, it can record the entire process of all review operations, automatically recording the operator, operation time, content before and after modification, review opinions, and judgment basis, generating a full-process traceability ledger that cannot be tampered with or deleted, thereby meeting the mandatory requirements of the IATF 16949 system for full-process traceability of quality control.

[0196] S210. Based on the review results of the list, determine the target control list; the target control list includes quality information that needs to be controlled and avoided during the product development process.

[0197] At this point, the initial control list can be filtered again based on the list review results to obtain quality information applicable to the current vehicle model / product under each preset review dimension, thus obtaining the target control list.

[0198] At the same time, the accuracy and reliability of the target control list can be ensured based on the review criteria.

[0199] In one possible implementation, if the audit results indicate that the quality information included in the initial control list cannot be used for the quality control of the current product, then the information processing method is determined based on the information type of the information to be reviewed that does not match the current product, and then the historical quality information is updated according to the information processing method.

[0200] In one example, the information types for the current product mismatch to be reviewed can include incompatibility and optimization needs. For incompatibility, the corresponding information processing method is reason labeling, such as filling in and recording the reason for the incompatibility and the basis for the judgment. For optimization needs, the corresponding information processing method is online modification, such as modifying the content to be optimized online and filling in detailed optimization instructions and justification.

[0201] At this point, a three-branch refined processing mechanism of "consistent / incompatible / requires optimization" can be used to execute differentiated processing procedures for different review results, thereby improving the comprehensiveness and refinement of the processing and ensuring the quality of the historical quality information database.

[0202] Optionally, under the traditional model, historical quality information lacks standardized accumulation, resulting in low experience reuse rates and no economies of scale. Specifically, in the traditional model, the experience in managing historical quality information is scattered across documents, reports, and engineers' personal experiences in various projects, lacking standardized, structured accumulation and correlation. It is difficult to reuse experience across different models and projects, and each new model project needs to repeatedly execute the entire process of screening, review, and judgment, resulting in a large amount of repetitive work. At the same time, because the existing model cannot achieve large-scale reuse of experience, there is no room for reducing the marginal cost of new model projects. At this point, even if the level of platform development by car companies increases, it is difficult to achieve economies of scale in quality control, resulting in a disconnect from the industry trend of platform development.

[0203] Therefore, in this embodiment of the application, after determining the target control list, knowledge can be accumulated based on the target control list to facilitate experience reuse. Simultaneously, after processing the quality information of the type requiring optimization, knowledge can also be accumulated to facilitate experience reuse.

[0204] In one possible implementation, during knowledge accumulation, the historical quality information database can be dynamically updated in real time based on the full review results corresponding to historical quality information. This achieves dynamic improvement of historical quality information and ensures the accuracy and completeness of the historical quality information database. Specific update rules include the following:

[0205] For quality information that is deemed "consistent," the list of compatible vehicle models, application project records, and verification information can be updated, and the compatibility scenarios for the current vehicle model can be supplemented. For quality information that is deemed "incompatible," the applicable boundary labels, incompatible scenario information, and reasons for incompatibility can be updated, and the negative sample feature library can be improved. For quality information that is deemed "needs optimization," the avoidance strategy, applicable boundaries, feature label set, and risk level information can be updated, and version records before and after modification can be retained to achieve full version traceability.

[0206] Furthermore, in this embodiment, an automotive quality control knowledge graph can be constructed / updated based on the updated historical quality information database, thereby enabling visualization and convenient searching of the historical quality information database.

[0207] In practice, a dynamically updated automotive quality control knowledge graph can be constructed based on the updated historical quality information database and a graph database to achieve visualization, correlation, and reusability of quality information.

[0208] Optionally, the specific build rules can be as follows:

[0209] Entity node definition: It includes nine major categories of core entity nodes, including vehicle model nodes, platform nodes, system nodes, component nodes, failure mode nodes, root cause nodes, corrective action nodes, scenario nodes, and regulatory nodes.

[0210] The definition of relationships includes six core categories: attribution, association, causation, adaptation, mutual exclusion, and inclusion, which clarify the relationship logic between entities.

[0211] Dynamic update mechanism: After the completion of the full-process management of each vehicle model project, the entity nodes and relationships of the knowledge graph are automatically updated based on the project data to ensure the real-time and completeness of the knowledge graph.

[0212] At this point, the knowledge graph for automotive quality control, which is built / updated, can intuitively display the transmission chain, scope of application, and solutions of quality information. It also supports engineers to quickly retrieve relevant quality knowledge through dimensions such as parts, failure modes, and root causes, enabling efficient reuse of quality knowledge.

[0213] In one possible embodiment, this application embodiment can also be based on a quality control knowledge graph and use the K-Means clustering algorithm to perform cluster analysis on rectification and avoidance strategies for historical problems with the same root cause, the same failure mode, and the same system, extract standardized and reusable quality risk avoidance solutions, and build a whole vehicle quality risk solution library; each solution corresponds to a clear applicable scenario, configuration boundary, implementation steps, verification standards, and responsible department, thereby supporting engineers to call it with one click in the process of new model development, FMEA analysis, and problem rectification, without having to formulate control solutions from scratch, and greatly improving the cross-project experience reuse rate (for example, the experience reuse rate is increased by 40%).

[0214] In one possible embodiment, the embodiments of this application may also perform incremental training on the dual-tower deep matching model based on the full review results of historical quality information, thereby improving the accuracy and reliability of determining the matching relationship between configuration feature vectors and quality feature vectors.

[0215] Specifically, the incremental training dataset can be determined based on the configuration feature vector, quality feature vector, matching results, and engineer review results used in this study. The results of "consistent" judgments are considered positive samples, while the results of "unsuitable" judgments are considered negative samples. Subsequently, the dual-tower deep matching model can be incrementally trained according to a preset iteration cycle (e.g., after a single project ends / monthly fixed iterations) to update the model's network parameters and continuously improve the model's matching accuracy.

[0216] This implementation method enables the model's generalization ability and accuracy to continuously improve as project data accumulates, achieving self-learning and self-optimization of the intelligent agent.

[0217] Figure 3 This is a schematic diagram illustrating the implementation process of a product development quality control method provided in an embodiment of this application, such as... Figure 3 As shown, in this embodiment of the application, data can be acquired first, such as product configuration information and historical quality information, and the acquired data can be characterized. After the product configuration information is characterized, a configuration feature vector can be obtained, and after the historical quality information is characterized, a quality feature vector can be obtained.

[0218] Next, based on the matching relationship between the configuration feature vector and the quality feature vector, as well as the quality control requirements, the historical quality information can be automatically preliminarily screened to obtain an initial control list.

[0219] Next, review tasks (e.g., review tasks for quality information to be reviewed in the initial control list) can be automatically distributed to matching target objects so that the target objects can perform review processing.

[0220] At this point, the review results of the target object's list can be received, and the quality information to be reviewed can be processed according to the matching information processing method based on the review results.

[0221] Specifically, if the audit result of the list is "consistent", then it can be directly included in the control list, that is, the target control list; if the audit result of the list is "inconsistent", then it can be included in the negative sample; if the audit result of the list is "needs optimization", then the quality information can be optimized and included in the control list, that is, the target control list.

[0222] Subsequently, this application embodiment can also update the historical quality information database based on the full review results of historical quality information, and incrementally optimize the model used to determine the matching relationship between configuration feature vectors and quality feature vectors based on the updated historical quality information database, thereby realizing the self-optimization of the model.

[0223] Figure 4 A schematic diagram of a product development quality control system provided in this application embodiment is shown below. Figure 4 As shown, the product development quality control system includes: a perception layer, a decision-making layer, an execution layer, and a feedback iteration layer.

[0224] The perception layer is responsible for collecting, standardizing, cleaning, and characterizing product configuration information and multi-source historical quality information to achieve unified representation of heterogeneous data and provide a standardized data foundation for subsequent intelligent matching.

[0225] Decision-making level: Based on a pre-trained dual-tower deep matching model and a high-performance vector retrieval engine, intelligent and accurate matching of product configuration information and historical quality information is achieved, completing preliminary screening and hierarchical recommendation, replacing the manual initial screening process.

[0226] Execution layer: Construct a standardized hierarchical review process to achieve professional and targeted distribution of matching results, branched judgment processing, and full-process traceability control, and complete adaptability judgment and strategy optimization.

[0227] Feedback Iteration Layer: Based on the review results, the layer dynamically updates the historical quality information database, constructs a quality control knowledge graph, incrementally iterates the AI ​​model, and accumulates a reusable solution library, thereby achieving closed-loop self-optimization of the intelligent agent and large-scale reuse of experience.

[0228] As described above, this application provides a method for refining an intelligent agent that avoids historical quality problems, adapts to product (e.g., automotive) development processes, possesses high matching accuracy, full-process closed-loop control, standardized knowledge accumulation, and self-iterative capabilities. It addresses the following core technical problems existing in the prior art:

[0229] 1. It addresses the core pain points of traditional manual screening methods, such as extremely low efficiency, unstable accuracy, and easy omission of high-risk issues in massive historical problem data.

[0230] This application replaces the initial manual screening process with an intelligent matching and screening method, reducing the screening time for a single piece of historical quality information from 5 minutes to 2 seconds, increasing screening efficiency by 60 times; for a single vehicle model project (assuming 1,000 pieces of historical quality information need to be screened), a single screening can save approximately 83 hours of work time.

[0231] 2. To address the issues of high redundancy in collaboration, long judgment cycles, inconsistent judgment standards, high risk of misjudgment, and poor traceability of the entire process in the existing cross-team offline review model, and to meet the compliance management requirements of the automotive industry.

[0232] This application, through a tiered review process, can reduce the number of cross-departmental special review meetings by 5 within a single project cycle, cumulatively reducing ineffective cross-departmental collaboration time by 332 hours, and saving a total of 415 hours of work time per project.

[0233] If an automaker needs to develop 20 new products / new models / facelifted models annually, it can save up to 8,300 man-hours per year, shortening the quality control cycle for new products / new models and thus adapting to the industry's rapid iteration development needs.

[0234] Meanwhile, the matching model provided in this application has a stable matching accuracy of ≥85%, which reduces the manual review rate to below 15%, thereby reducing resource waste caused by invalid issues by up to 60%. Furthermore, the built-in high-risk priority rule achieves 100% coverage of safety and regulatory mandatory issues, significantly reducing the omission of high-risk issues due to manual screening. This lowers the probability of similar quality problems recurring after the mass production of new models, further reducing after-sales claims losses. In addition, this application enables full-process operation to be fully traceable and tamper-proof, thus meeting the mandatory requirements of IATF 16949 for full-process traceability of quality control, thereby reducing compliance risks.

[0235] 3. Address industry pain points such as the lack of standardized accumulation of quality control knowledge, low cross-project experience reuse rate, large amount of repetitive work, high marginal cost of new projects, and difficulty in achieving economies of scale in existing technologies.

[0236] This application, through standardized knowledge accumulation, increases the reusability of cross-team quality control experience by 40%, realizes the pre-emptive avoidance of quality risks in the entire vehicle development process, and thus improves the company's overall vehicle quality control capabilities.

[0237] 4. Fill the gap in existing technologies for closed-loop self-iterative intelligent agent solutions, construct a full-link intelligent management and control architecture of "perception-decision-execution-feedback-iteration", realize end-to-end automated and intelligent management and control to avoid historical quality problems, and adapt to the development needs of rapid iteration of new models in the automotive industry.

[0238] Figure 5 A schematic diagram of a product development quality control device provided in this application embodiment is shown below. Figure 5 As shown, the product development quality control device 50 provided in this embodiment includes:

[0239] The acquisition unit 501 is used to acquire product configuration information and historical quality information, and to determine the configuration feature vector corresponding to the product configuration information and the quality feature vector corresponding to the historical quality information.

[0240] The first determining unit 502 is used to determine an initial control list based on the matching relationship between the configuration feature vector and the quality feature vector, and the quality control requirements, and to send the initial control list to the matched target object for review and processing.

[0241] The second determining unit 503 is used to receive the list review results of the target objects and determine the target control list based on the list review results; wherein, the target control list includes quality information that needs to be avoided and controlled during the product development process.

[0242] In one possible implementation, the matching relationship between the configuration feature vector and the quality feature vector is determined according to the following steps:

[0243] Based on the preset configuration tower model, extract the higher-order configuration representation vector of the configuration feature vector;

[0244] Based on the pre-defined quality tower model, extract the higher-order quality representation vector of the quality feature vector;

[0245] The matching relationship between configuration feature vectors and quality feature vectors is determined based on the similarity between the higher-order configuration representation vector and the higher-order quality representation vector.

[0246] In one possible implementation, the number of historical quality information entries is multiple; the matching relationship is used to characterize the matching score between each piece of historical quality information and the product configuration information; the quality control requirements are used to indicate the information filtering rules and information classification rules; at this time, the first determining unit 502 is used to:

[0247] Based on the matching score and information filtering rules, historical quality information is filtered to obtain an information recommendation list; the information recommendation list includes quality information that needs to be reviewed.

[0248] Based on the professional domain to which each quality information to be reviewed belongs, the quality information to be reviewed is classified and processed to obtain a classification recommendation list;

[0249] Based on the categorized recommendation list, determine the initial control list.

[0250] In one possible implementation, the information filtering rules indicate hierarchical filtering rules; the information recommendation list includes at least a first recommendation table and a second recommendation table; in this case, the device is also used for:

[0251] Determine the quality information to be reviewed and the corresponding risk level information included in the second recommendation table;

[0252] If the risk level information meets the preset risk requirements, the quality information to be reviewed included in the second recommendation table will be updated to the first recommendation table.

[0253] In one possible implementation, the initial control list includes at least one quality information item to be reviewed; each quality information item to be reviewed has corresponding hierarchical information and / or classification information; at this time, the first determining unit 502 is used to:

[0254] Based on hierarchical information and / or classification information, determine the matching target object;

[0255] The quality information to be reviewed is sent to the matched target for review and processing.

[0256] In one possible implementation, the second determining unit 503 is configured to:

[0257] In response to an information review instruction initiated by the target object, display the information to be reviewed under the preset review dimensions;

[0258] In response to the editing operation of the target object on the information to be reviewed, the checklist review result is determined; wherein, the checklist review result is used to indicate whether the quality information included in the initial control checklist can be used for the quality control of the current product.

[0259] In one possible implementation, the device is also used for:

[0260] If the review results indicate that the quality information included in the initial control list cannot be used for the quality control of the current product, then the information processing method shall be determined according to the information type of the information to be reviewed that does not match the current product.

[0261] Update historical quality information according to the information processing method.

[0262] The product development quality control device provided in this embodiment can execute the method provided in the above method embodiment. Its implementation principle and technical effect are similar, and will not be described in detail here.

[0263] Figure 6 This is a schematic diagram of the structure of a computer device provided in an embodiment of this application. Figure 6 As shown, the computer device 60 provided in this embodiment includes at least one processor 601 and a memory 602. Optionally, the computer device 60 further includes a communication component 603. The processor 601, memory 602, and communication component 603 are connected via a bus 604.

[0264] In a specific implementation, at least one processor 601 executes computer execution instructions stored in memory 602, causing at least one processor 601 to perform the above-described method.

[0265] The specific implementation process of processor 601 can be found in the above method embodiments, and its implementation principle and technical effect are similar. It will not be repeated here.

[0266] In the above embodiments, it should be understood that the processor can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), etc. The general-purpose processor can be a microprocessor or any conventional processor. The steps of the method disclosed in this invention can be directly implemented by a hardware processor, or implemented by a combination of hardware and software modules within the processor.

[0267] The memory may include random access memory (RAM) and may also include non-volatile memory (NVM), such as at least one disk storage device.

[0268] The bus can be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, or an Extended Industry Standard Architecture (EISA) bus, etc. Buses can be categorized as address buses, data buses, control buses, etc. For ease of illustration, the buses shown in the accompanying drawings are not limited to a single bus or a single type of bus.

[0269] This application also provides a computer program product, including a computer program that, when executed by a processor, implements the above-described method.

[0270] This application also provides a computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, implement the above-described method.

[0271] The aforementioned readable storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk. The readable storage medium can be any available medium accessible to a general-purpose or special-purpose computer.

[0272] An exemplary readable storage medium is coupled to a processor, enabling the processor to read information from and write information to the readable storage medium. Of course, the readable storage medium can also be a component of the processor. The processor and the readable storage medium can reside in an Application Specific Integrated Circuit (ASIC). Alternatively, the processor and the readable storage medium can exist as discrete components in the device.

[0273] The division of units is merely a logical functional division; in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be indirect coupling or communication connection through some interfaces, devices, or units, and may be electrical, mechanical, or other forms.

[0274] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0275] In addition, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0276] If a function is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0277] Those skilled in the art will understand that all or part of the steps of the above-described method embodiments can be implemented by hardware related to program instructions. The aforementioned program can be stored in a computer-readable storage medium. When executed, the program performs the steps of the above-described method embodiments; and the aforementioned storage medium includes various media capable of storing program code, such as ROM, RAM, magnetic disks, or optical disks.

[0278] Finally, it should be noted that other embodiments of the invention will readily occur to those skilled in the art upon consideration of the specification and practice of the invention disclosed herein. This invention is intended to cover any variations, uses, or adaptations of the invention that follow the general principles of the invention and include common knowledge or customary techniques in the art not disclosed herein, and is not limited to the precise structures described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The scope of the invention is limited only by the appended claims.

Claims

1. A product development quality control method, characterized in that, include: Obtain product configuration information and historical quality information, and determine the configuration feature vector corresponding to the product configuration information and the quality feature vector corresponding to the historical quality information; Based on the matching relationship between the configuration feature vector and the quality feature vector, and the quality control requirements, an initial control list is determined, and the initial control list is sent to the matched target object for review and processing. Receive the review results of the target object's list, and determine the target control list based on the review results; wherein, the target control list includes quality information that needs to be controlled and avoided during the product development process.

2. The method according to claim 1, characterized in that, The matching relationship between the configuration feature vector and the quality feature vector is determined according to the following steps: Based on the preset configuration tower model, extract the higher-order configuration representation vector of the configuration feature vector; Based on the preset mass tower model, extract the higher-order mass representation vector of the mass feature vector; The matching relationship between the configuration feature vector and the quality feature vector is determined based on the similarity between the higher-order configuration representation vector and the higher-order quality representation vector.

3. The method according to claim 1, characterized in that, The number of historical quality information items is multiple; the matching relationship is used to characterize the matching score between each piece of historical quality information and the product configuration information. The quality control requirements are used to indicate information filtering rules and information classification rules; Based on the matching relationships and quality control requirements, an initial control list is determined, including: Based on the matching score and the information filtering rules, the historical quality information is filtered to obtain an information recommendation list; wherein, the information recommendation list includes quality information to be reviewed; Based on the professional domain to which each quality information to be reviewed belongs, the quality information to be reviewed is classified and processed to obtain a classification recommendation list; The initial control list is determined based on the classification recommendation list.

4. The method according to claim 3, characterized in that, The information filtering rules indicate hierarchical filtering rules; the information recommendation list includes at least a first recommendation table and a second recommendation table; the method further includes: Determine the quality information to be reviewed and the corresponding risk level information included in the second recommendation table; If the risk level information meets the preset risk requirements, the quality information to be reviewed included in the second recommendation table will be updated to the first recommendation table.

5. The method according to claim 1, characterized in that, The initial control list includes at least one quality information to be reviewed; each quality information to be reviewed has corresponding hierarchical information and / or classification information; Sending the initial control list to the matched target object for review and processing includes: Based on the hierarchical information and / or the classification information, the matching target object is determined; The quality information to be reviewed is sent to the matched target object for review processing.

6. The method according to any one of claims 1-5, characterized in that, Receive the audit results of the target object list, including: In response to the information review instruction initiated by the target object, display the information to be reviewed under the preset review dimensions; In response to the target object's editing operation on the information to be reviewed, the checklist review result is determined; wherein, the checklist review result is used to indicate whether the quality information included in the initial control checklist can be used for the quality control of the current product.

7. The method according to claim 6, characterized in that, The method further includes: If the review result of the list indicates that the quality information included in the initial control list cannot be used for the quality control of the current product, then the information processing method is determined according to the information type of the information to be reviewed that does not match the current product; Update historical quality information according to the aforementioned information processing method.

8. A product development quality control device, characterized in that, include: The acquisition unit is used to acquire product configuration information and historical quality information, and determine the configuration feature vector corresponding to the product configuration information and the quality feature vector corresponding to the historical quality information. The first determining unit is used to determine an initial control list based on the matching relationship between the configuration feature vector and the quality feature vector, and the quality control requirements, and to send the initial control list to the matched target object for review processing; The second determining unit is used to receive the list review results of the target object and determine the target control list based on the list review results; wherein the target control list includes quality information that needs to be avoided and controlled during the product development process.

9. A computer device, characterized in that, include: Memory, processor; The memory stores computer-executed instructions; The processor executes computer execution instructions stored in the memory, causing the processor to perform the method as described in any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer-executable instructions, which, when executed by a processor, are used to implement the method as described in any one of claims 1-7.

Images