Manufacturing intelligent quality management core element extraction method and electronic device

Abstract

The application discloses a manufacturing industry digital quality management core element extraction method and electronic equipment, and belongs to the technical field of data processing. The method comprises the following steps: obtaining raw material purchasing data and usage data from a database according to a first text; obtaining a first text sequence based on the purchasing data and the usage data; extracting a named entity in the first text sequence by using a first large language model to obtain a first named entity sequence; processing the first text sequence by using an implicit Dirichlet distribution to obtain a first element list; performing vectorization on the first named entity sequence to obtain a first vector set corresponding to the first named entity sequence; obtaining a second vector set corresponding to the first named entity sequence according to the correlation degree between words in the first element list and a quality management theme; and determining the correlation degree between elements in the first element list based on the first vector set and the second vector set. The application can automatically discover complex correlations between elements.

Description

Technical Field

[0001] The present invention belongs to the technical field of data processing, and specifically relates to a method for extracting core elements of digital and intelligent quality management in the manufacturing industry and an electronic device. Background Art

[0002] With the development of technology, more and more enterprises have begun to use Internet of Things devices to monitor the data of operating machinery and equipment, and adjust the equipment operation conditions and the upstream and downstream production lines based on the monitored data. The management of product quality has also become more intelligent.

[0003] However, when the existing technology conducts digital and intelligent quality management, the extraction of core elements is relatively general. Almost all the data parameters of Internet of Things devices and operating equipment are counted together. This makes the elements to be processed too many and the calculation amount huge when conducting digital and intelligent quality management. In addition, enterprise practice focuses on the application of single digital technology in a certain vertical quality management scenario, and fails to establish a technical system for digital and intelligent quality management and its supporting management system based on the entire value chain, the entire supply chain, and the entire ecological chain, thus exposing the problem of lagging digital integration and collaboration capabilities in the production, operation, organization and other links of the enterprise. That is, the existing digital and intelligent quality management does not determine the correlation degree between data, and then obtain the correlation degree between core elements, making it impossible to conduct targeted troubleshooting when a problem occurs in one link. Summary of the Invention

[0004] An object of the present invention is to provide a method for extracting core elements of digital and intelligent quality management in the manufacturing industry and an electronic device, which can solve the technical problem that the digital management system in the existing technology needs to process too many elements and there is no correlation between elements.

[0005] According to the first aspect of the present invention, there is provided a method for extracting core elements of digital and intelligent quality management in the manufacturing industry, including:

[0006] Retrieving the procurement data and usage data of raw materials from the database according to the first text, where the procurement data includes procurement process documents and associated business information, and the usage data includes the usage records of raw materials, associated production business information, and associated Internet of Things information data, and the associated Internet of Things information data is generated by associated production business;

[0007] Obtaining a first text sequence based on the procurement data and usage data;

[0008] Using a first large language model to extract named entities in the first text sequence to obtain a first named entity sequence;

[0009] Using the latent Dirichlet allocation to process the first text sequence to obtain a first list of elements;

[0010] Vectorize the first named entity sequence to obtain the first vector set corresponding to the first named entity sequence;

[0011] The second vector set corresponding to the first named entity sequence is obtained based on the relevance between the words in the first element list and the quality management topic;

[0012] The degree of association between elements in the first element list is determined based on the first vector set and the second vector set.

[0013] Optionally, the process of obtaining the first named entity sequence includes:

[0014] The large language model is used to segment and tag the text in the first text sequence, and strings with the part of speech of noun or gerund are added to the first named entity sequence.

[0015] Optionally, the first element list includes words and the topic distribution corresponding to the words;

[0016] The process of obtaining the second vector set includes:

[0017] Generate a topic-word distribution matrix for the text within the first text sequence, sort the words based on word frequency to obtain a first word list, remove words in the first word list whose word frequency is below a first threshold and whose word frequency is above a second threshold, uniformly sample the first word list to obtain a second word list, construct a first probability distribution matrix based on the topic distribution corresponding to the second word list, construct a sparse matrix based on the constructed first probability distribution matrix, and obtain the word vector corresponding to each word based on the sparse matrix;

[0018] Obtain the second vector set based on the words and their corresponding word vectors.

[0019] Optionally, principal component analysis can be used to reduce the dimensionality of the first or second vector set to make them have consistent dimensions.

[0020] Optionally, a third vector is obtained by normalizing the first and second vectors corresponding to the named entities in the first named entity sequence.

[0021] Optionally, the degree of association between named entities within the first named entity sequence is determined based on the third vector;

[0022] The degree of association between named entities is determined by the cosine approximation of the third vector corresponding to the named entity.

[0023] Optionally, the first named entity sequence includes a first named entity and a second named entity;

[0024] Furthermore, in response to the fact that the correlation between the first named entity and the second named entity is greater than the third threshold and the Euclidean distance is greater than the fourth threshold, the first named entity is updated to the neighbor list of the second named entity, and the second named entity is updated to the neighbor list of the first named entity.

[0025] Alternatively, in response to the fact that the correlation between the first named entity and the second named entity is less than a third threshold and the Euclidean distance is less than a fourth threshold, the neighbor nodes of the first named entity are updated based on the neighbor nodes of the second named entity.

[0026] Optionally, named entities whose neighbor list size is greater than the fifth threshold within the first named entity sequence are obtained and added to the second named entity sequence.

[0027] Optionally, the process of determining the association degree between named entities within the first named entity sequence includes:

[0028] The first associated word sequence corresponding to the named entity in the first named entity sequence is determined based on the first vector set, and in response to the length of the first associated word sequence corresponding to the named entity being greater than the sixth threshold, the named entity is added to the third named entity sequence.

[0029] Based on the second vector set, determine the second associated word sequence that associates words within the third named entity sequence, and in response to the length of the second associated word sequence corresponding to the named entity being greater than the sixth threshold, add the named entity to the fourth named entity sequence.

[0030] According to a second aspect of the present invention, an electronic device is provided, including a memory and a processor, wherein the memory is used to store program data and the processor is used to execute the program data to implement the core element extraction method for intelligent manufacturing quality management described in the first aspect of the present invention.

[0031] The beneficial effects of this invention are as follows: This embodiment integrates multimodal data such as procurement and business data, combining the accurate recognition capabilities of large language models and the unsupervised discovery capabilities of LDA. It can identify both specific entities and abstract elements. By vectorizing some information and transforming qualitative textual descriptions into quantitative mathematical calculations, the system can automatically discover complex relationships between elements and achieve quality traceability and risk prediction. By analyzing the element relationship network, it can predict which production processes might be at risk due to changes in a supplier's qualifications. Simultaneously, it can optimize decision-making, help enterprises identify key connection points in quality management, thereby optimizing processes and achieving intelligent quality management from post-event remediation to pre-event prevention. Attached Figure Description

[0032] Figure 1 This is a flowchart of the method for extracting core elements of digital and intelligent quality management in the manufacturing industry in this embodiment of the invention. Detailed Implementation

[0033] Various exemplary embodiments of the present invention will now be described in detail with reference to the accompanying drawings. It should be noted that, unless otherwise specifically stated, the relative arrangement, numerical expressions, and values ​​of the components and steps set forth in these embodiments do not limit the scope of the invention.

[0034] The following description of at least one exemplary embodiment is merely illustrative and is in no way intended to limit the invention or its application or use.

[0035] Techniques, methods, and apparatus known to those skilled in the art may not be discussed in detail, but where appropriate, such techniques, methods, and apparatus should be considered part of the specification. In all examples shown and discussed herein, any specific values ​​should be interpreted as merely exemplary and not as limitations. Therefore, other examples of exemplary embodiments may have different values.

[0036] It should be noted that similar labels and letters in the following figures indicate similar items; therefore, once an item is defined in one figure, it does not need to be discussed further in subsequent figures.

[0037] In the specification of this invention, the terms "first" and "second" may explicitly or implicitly include one or more of the same feature. In the description of this invention, unless otherwise stated, "multiple" means two or more. Furthermore, in the specification, "and / or" indicates at least one of the connected objects, and the character " / " generally indicates that the preceding and following objects are in an "or" relationship.

[0038] like Figure 1 As shown in the figure, this embodiment introduces a method for extracting core elements of digital and intelligent quality management in the manufacturing industry, including steps 1100-1700.

[0039] Step 1100: Based on the first text, retrieve the raw material procurement data and usage data from the database. The procurement data includes procurement process documents and related business information. The usage data includes raw material usage records, related production business information, and related IoT information data, wherein the related IoT information data is generated by related production business.

[0040] The first text is related to digital quality management, such as a report or a text describing a quality issue.

[0041] The procurement process documents include procurement contracts, supplier qualification certificates, purchase orders, and inspection reports. Business information includes the procurement officer, procurement time, and supplier information. Usage data includes which specific production batch used the raw material (usage records), the production plan for that batch of products, process parameters (related to production business information), and sensor readings such as temperature, pressure, and vibration during the production process (related IoT information data).

[0042] Step 1200: Obtain the first text sequence based on the procurement data and usage data.

[0043] Structured data (such as database tables) and semi-structured / unstructured data (such as reports and data logs) obtained from the database are merged and preprocessed. Finally, they are transformed into one or more coherent, machine-readable plain text segments. For example, a sentence like this might be generated: "Supplier A provided raw material E for product D of model C in batch B." This first text sequence forms the basis for subsequent analysis.

[0044] Step 1300: Use the first large language model to extract named entities from the first text sequence to obtain the first named entity sequence.

[0045] Using a specially trained and optimized first-class language model, such as GPT or BERT, the system identifies entity names with specific meanings from the first text sequence. For example, from the example text above, the first-class language model identifies "supplier A", "batch B", "model C", "product D", and "raw material E", which together form the first named entity sequence.

[0046] Step 1400: Process the first text sequence using the hidden Dirichlet distribution to obtain the first list of elements.

[0047] Latent Dirichlet Allocation (LDA) is a classic unsupervised topic model. LDA does not concern itself with specific entities, but rather analyzes the co-occurrence patterns of words to uncover hidden topics behind the text. Each topic consists of a set of keywords with high probabilities.

[0048] The first element list is not a list of entities, but rather a list of topics or concepts. For example, LDA extracts "supplier evaluation," "qualification certification," and "contract terms" from the text, which correspond to procurement compliance. Meanwhile, "temperature control," "temperature monitoring," and "process parameters" correspond to production process control. This method uncovers implicit quality management elements within the text, which are often processes, standards, or areas of focus.

[0049] Step 1500: Vectorize the first named entity sequence to obtain the first vector set corresponding to the first named entity sequence.

[0050] Using word embedding techniques or sentence embedding models, each named entity (such as "supplier A" or "product D") is transformed into a high-dimensional mathematical vector. This vector captures the semantic information of the entity. Entities with similar semantics are also closer in distance within the vector space.

[0051] Step 1600: Obtain the second vector set corresponding to the first named entity sequence based on the relevance between the words in the first element list and the quality management topic.

[0052] For each element in the first element list, the correlation between that element and each entity in the first named entity sequence is calculated. For example, "Supplier A" has a high correlation with the procurement compliance element, while "Product D" has a low correlation with the procurement compliance element and a high correlation with the production process control element.

[0053] Each element is represented as a weight vector, the dimensions of which are determined by the entity sequence. The value of each dimension is the correlation score between the corresponding entity and the element, forming a second vector set.

[0054] Step 1700: Determine the correlation between elements in the first element list based on the first vector set and the second vector set.

[0055] The first vector set represents the semantic vectors of entities. The second vector set represents the weight vectors of elements, which essentially represent the strength of the association between elements and each entity. By calculating the similarity between the element vectors in the second vector set, the degree of association between these elements can be determined; for example, cosine similarity can be calculated.

[0056] Ultimately, a network of interconnected elements is obtained. For example, calculations reveal a certain similarity between the vectors of the elements of procurement compliance and production process control, indicating that at the data level, problems in the procurement process are related to certain parameters in the production process. This provides data-driven insights for quality traceability and root cause analysis.

[0057] This embodiment integrates multimodal data from procurement and operations, combining the precise recognition capabilities of large language models with the unsupervised discovery capabilities of LDA. It can identify both specific entities and abstract elements. By vectorizing some information and transforming qualitative textual descriptions into quantitative mathematical calculations, the system can automatically discover complex relationships between elements and achieve quality traceability and risk prediction. By analyzing the element relationship network, it can predict which production stages might be at risk due to changes in a supplier's qualifications. Simultaneously, it can optimize decision-making, helping enterprises identify key connection points in quality management, thereby optimizing processes and achieving intelligent quality management from post-event remediation to pre-event prevention.

[0058] In this embodiment, the process of obtaining the first named entity sequence includes: using a large language model to segment and tag the text in the first text sequence, and adding strings with the part of speech of noun or gerund to the first named entity sequence.

[0059] First, the text within the first text sequence is segmented, breaking the coherent sequence into individual meaningful words. For example, the sentence "Supplier A provided high-quality raw material E" is segmented into "supplier", "A", "provided", "high-quality", and "raw material E". Large language models can use their built-in word segmenters to perform this segmentation.

[0060] Then, each segmented word is labeled with its grammatical role, such as noun, verb, adjective, etc. Based on its massive training data, the large language model has a strong ability to determine the part of speech of words in specific contexts. For example, it can accurately distinguish the different parts of speech of "management" in "quality management" and "management equipment".

[0061] In the field of quality management, most of the core entities are nouns and gerunds.

[0062] This method is simple to implement, computationally efficient, and can capture most key entities. It also directly leverages the powerful underlying language understanding capabilities of large language models, avoiding the cost of training complex models.

[0063] In this embodiment, the first element list includes words and the topic distribution corresponding to the words.

[0064] The process of obtaining the second vector set includes: generating a topic-word distribution matrix of the text in the first text sequence; sorting the words based on word frequency to obtain a first word list; removing words in the first word list whose word frequency is lower than a first threshold and whose word frequency is higher than a second threshold; uniformly sampling the first word list to obtain a second word list; constructing a first probability distribution matrix based on the topic distribution corresponding to the second word list; constructing a sparse matrix based on the constructed first probability distribution matrix; obtaining the word vector corresponding to each word based on the sparse matrix; and obtaining the second vector set based on the words and their corresponding word vectors.

[0065] The first element list is not merely a list of words, but rather a distribution of words and their corresponding topics. This means that each word is represented as a probability distribution vector across different topics. For example, the word "temperature" has a high probability on the topic of "production process control" but a low probability on the topic of "procurement compliance."

[0066] After processing the first text sequence using LDA, a topic-word distribution matrix will be obtained. Each row of this matrix represents a topic, each column represents a word, and each value in the matrix indicates the probability of the word appearing in the topic. For example, in the row corresponding to the "production process control" topic, the probability value corresponding to "temperature" will be relatively high.

[0067] Statistically analyze the occurrence frequencies of all words in the entire corpus and sort them, removing low-frequency words and high-frequency words. Words with too low word frequencies may be typos, rare words, or unimportant accidental occurrences, and they lack statistical significance and will be filtered as noise. Words with too high word frequencies do not help in differentiating topics and instead interfere with the model. Removing them can enhance the representativeness of features. For example, words such as "of", "is", "problem", or overly general words in the field. This step is standard text preprocessing, which can retain those words with sufficient representativeness and distinctiveness.

[0068] The first list of words after word frequency screening may still be very large. By using uniform sampling, on the one hand, it can reduce the number of words to be processed in subsequent steps, and on the other hand, it can ensure the balance of topic coverage, avoiding words of certain high-frequency topics dominating the entire list and ensuring that representative words can be extracted from all topics. The second list of words is the high-quality core vocabulary set that finally remains after screening and optimization and can best represent the topic characteristics of the entire corpus.

[0069] The first probability distribution matrix is a new matrix formed by only extracting the elements corresponding to those words in the second list of words from the topic-word distribution matrix.

[0070] Based on the first probability distribution matrix, a sparse matrix is constructed, which will set the probability values below a certain threshold in the first probability distribution matrix to zero. Because although a word has a very small probability in all topics, only one or two topics with the highest probabilities truly define its meaning. Through sparsification, strong signals can be highlighted, making the strong associations between each word and a few core topics more obvious, reducing noise, weakening the weak interference of irrelevant topics, and also improving computational efficiency.

[0071] Each row of the matrix after sparsification is the word vector of a word. The dimension of this vector is equal to the number of topics, and the value of each dimension represents the association strength between the word and the corresponding topic. After sparsification, most of the values of the dimensions are zero.

[0072] We now have a dictionary mapping the second list of words to their topic word vectors. Then, we iterate through each entity in the first named entity sequence. If the entity exists in the constructed word vector dictionary, we use its vector directly. If the entity is not in the dictionary, it is split, and the vectors of the split words are weighted and summed to synthesize a vector representing the entity. Finally, we assign such a topic-based vector to each entity in the first named entity sequence; this set is the second vector set.

[0073] In this embodiment, principal factor analysis is used to reduce the dimensionality of the first vector set or the second vector set so that they have a consistent dimension.

[0074] The first vector set is generated based on a general semantic model and has a high original dimensionality. The second vector set is generated based on an LDA topic model, and its dimensionality equals the number of topics. Because their dimensions differ, subsequent calculations are impossible. Dimensionality reduction is used to establish a common, comparable benchmark space for these two vector sets from different sources and with different semantic spaces.

[0075] Principal component analysis, also known as principal factor analysis, is a classic linear dimensionality reduction technique. Its goal is to find the most important features in the data and retain as much information as possible from the original data using fewer dimensions.

[0076] In this embodiment, a third vector is obtained by normalizing the first and second vectors corresponding to the named entities within the first named entity sequence. The correlation between named entities within the first named entity sequence is determined based on the third vector; wherein, the correlation between named entities is determined by the cosine approximation of the third vector corresponding to the named entity.

[0077] The first vector represents the features of the named entity in the general semantic space, and the second vector represents the features of the named entity in the specific domain topic space. For each entity, its corresponding first and second vectors are concatenated. For example, after dimensionality reduction, both the first and second vectors are 50-dimensional. Therefore, for a given entity, its third vector is formed by concatenating its 50-dimensional first vector and 50-dimensional second vector to create a 100-dimensional third vector. Furthermore, normalization is used to eliminate the influence of dimensions, preparing for subsequent calculations of the cosine approximation.

[0078] For any two entities in the sequence, calculate the cosine similarity between their normalized third vectors.

[0079] The first named entity sequence includes a first named entity and a second named entity.

[0080] Furthermore, in response to the fact that the correlation between the first named entity and the second named entity is greater than the third threshold and the Euclidean distance is greater than the fourth threshold, the first named entity is updated to the neighbor list of the second named entity, and the second named entity is updated to the neighbor list of the first named entity.

[0081] Alternatively, in response to the fact that the correlation between the first and second named entities is less than a third threshold and the Euclidean distance is less than a fourth threshold, the neighboring nodes of the first named entity are updated based on the neighboring nodes of the second named entity.

[0082] Retrieve named entities whose neighbor list size is greater than the fifth threshold within the first named entity sequence and add them to the second named entity sequence.

[0083] If the correlation between the first and second named entities is greater than the third threshold, it indicates that the two entities are very similar in semantics and topic patterns. However, if the Euclidean distance is greater than the fourth threshold, it indicates that although the two entities are similar in pattern, there are significant differences in feature strength. This may mean that they are participants of different types and strengths in the same process. For example, a core supplier and an important but non-core component.

[0084] The system updates the neighbor list of the first named entity to the neighbor list of the second named entity, and then updates the neighbor list of the second named entity to the neighbor list of the first named entity. This is a process of establishing a bidirectional strong connection. The system considers that although the two entities have different characteristics, their relationship is strong enough that they should be directly linked. This is similar to two frequently interacting people following each other on a social network.

[0085] If the correlation between the first and second named entities is less than the third threshold, it indicates that the two entities do not have direct and obvious pattern similarity. Meanwhile, if the Euclidean distance is less than the fourth threshold, it indicates that their overall feature strengths are very similar.

[0086] The neighbor nodes of the first named entity are updated based on the neighbor nodes of the second named entity. This is an indirect connection inference process. The system considers that the two entities are not directly similar, but they belong to the same category because their feature strengths are close. Using this rule, the system can recommend friends of entity A to entity B. For example, entities A and B may not be highly related, but their feature strengths are very similar. If entity A and entity C are strong neighbors, the system might also add entity C to entity B's neighbor list. This helps to discover hidden community structures.

[0087] In graph theory, the size of the neighbor list is called the degree of a node. A node with a higher degree has more connections. The system traverses all entities, selecting only those with degree centrality higher than a fifth threshold, ultimately obtaining a second named entity sequence. This sequence is no longer the original set of all entities, but rather a set of key entities selected after network analysis.

[0088] By constructing a network, we can analyze which entities act as hubs and which as bridges. Entities in the second named entity sequence represent key participants in the entire digital quality management system of the manufacturing industry; these may include core suppliers, critical equipment, and bottleneck process parameters. This can guide resource allocation and risk control. Quality management resources and efforts should be prioritized for these key entities in the second named entity sequence, as problems with these key entities will rapidly spread through the network. Analyzing the neighbors of these key entities allows for a better understanding of business process dependencies, providing data support for process optimization and reengineering.

[0089] In this embodiment, the process of determining the association degree between named entities within the first named entity sequence includes:

[0090] The first associated word sequence corresponding to the named entity in the first named entity sequence is determined based on the first vector set, and in response to the length of the first associated word sequence corresponding to the named entity being greater than the sixth threshold, the named entity is added to the third named entity sequence.

[0091] Based on the second vector set, determine the second associated word sequence that associates words within the third named entity sequence, and in response to the length of the second associated word sequence corresponding to the named entity being greater than the sixth threshold, add the named entity to the fourth named entity sequence.

[0092] This is a keyword sequence-based filtering mechanism that adds a new filtering layer based on local context and global topic to determine the relevance between entities, making the final analysis results more accurate and reliable. The purpose of this process is to perform a pre-screening before calculating the relevance of all entities, ensuring that subsequent complex calculations focus only on entities with sufficiently rich information and clear context. Two levels of filtering are employed: the first layer is based on general semantics, and the second layer is based on domain topic.

[0093] For a given named entity, the system searches within the semantic space comprised of the entire first set of vectors for the few other words or entities most similar to its vector. This list of most similar words constitutes the first sequence of related words for that entity. It represents the most relevant context for this entity from the perspective of a general language model.

[0094] The system checks the length of the first associated word sequence for each entity. If the length exceeds a sixth threshold, the entity is retained. An entity with many similar neighbors in the semantic space is considered information-rich and semantically clear. It appears frequently in the training corpus, has diverse contexts, and is not an isolated or obscure item. It has sufficient data to support reliable association analysis. Conversely, if an entity's associated word sequence is short, it may be an isolated point, a new word, or noisy data; excluding it can improve the overall analysis quality.

[0095] For entities already selected from the third named entity sequence in the first step, the system performs the same operation in the topic space formed by the second vector set. It searches for several other words or entities most similar to its topic vector. This list constitutes the second sequence of related terms for that entity. It represents the other elements most relevant to this entity from the perspective of the topic of manufacturing quality management.

[0096] Similarly, the system checks if the length of the second associated word sequence for each entity is greater than the sixth threshold. If so, the entity successfully enters the final fourth named entity sequence. Some entities may be common in general language and thus pass the first layer of filtering, but they may not be core elements in a specific quality management domain. This layer of filtering ensures that the entities that remain are not only linguistically explicit but also highly relevant and important in the domain's business context.

[0097] This keyword sequence filtering mechanism significantly enhances the robustness and practicality of the entire method, automatically filtering out entities with low information content and weak relevance to prevent them from contaminating subsequent complex calculations and improving the signal-to-noise ratio of the final result. Subsequent dimensionality reduction, fusion, and network computation are all computationally intensive operations, and this mechanism can improve computational efficiency. Pre-reducing the number of entities can significantly shorten processing time. Through a dual filtering mechanism, it strongly ensures that the final set of entities analyzed in depth closely revolves around the core theme of digital and intelligent quality management in manufacturing. The first and second keyword sequences of each entity can themselves serve as part of the analysis report, intuitively explaining why a particular entity is considered important.

[0098] This embodiment introduces an electronic device, including a memory and a processor. The memory is used to store program data, and the processor is used to execute the program data to implement the core element extraction method for intelligent manufacturing quality management described in any embodiment of the present invention.

[0099] While specific embodiments of the present invention have been described in detail by way of examples, those skilled in the art should understand that the above examples are for illustrative purposes only and are not intended to limit the scope of the invention. Those skilled in the art should understand that modifications can be made to the above embodiments without departing from the scope and spirit of the invention.

[0100] Those skilled in the art will recognize that the modules and algorithm steps described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0101] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working process of the above-described apparatus and equipment can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0102] In the embodiments provided in this application, it should be understood that the disclosed apparatus and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of modules is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple modules 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 through some interfaces; the indirect coupling or communication connection between apparatuses or modules may be electrical, mechanical, or other forms.

[0103] The modules described as separate components may or may not be physically separate. The components shown as modules may or may not be physical modules; that is, they may be located in one place or distributed across multiple network modules. Some or all of the modules can be selected to achieve the purpose of the embodiments of the present invention, depending on actual needs.

[0104] In addition, the functional modules in the embodiments of the present invention can be integrated into one processing module, or each module can exist physically separately, or two or more modules can be integrated into one module.

[0105] If the aforementioned functions are implemented as software functional modules and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, essentially, or the part that contributes to the prior art, or a portion 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, ROM, RAM, magnetic disks, or optical disks.

[0106] The above description is merely a preferred embodiment of this application and an explanation of the technical principles employed. Those skilled in the art should understand that the scope of the invention involved in this application is not limited to technical solutions formed by specific combinations of the above-described technical features, but should also cover other technical solutions formed by arbitrary combinations of the above-described technical features or their equivalents without departing from the inventive concept. For example, technical solutions formed by substituting the above features with (but not limited to) technical features with similar functions disclosed in this application.

[0107] It should be understood that the sequence numbers of the steps in the invention's content and embodiments do not absolutely imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention. The foregoing description of embodiments of this disclosure has been provided for illustrative and descriptive purposes. The foregoing description is not exhaustive and is not intended to limit this disclosure to the exact form disclosed. Various modifications and variations may exist based on the foregoing teachings, or various modifications and variations may be derived from the practice of this disclosure. These embodiments were chosen and described to illustrate the principles of this disclosure and its practical application, so that those skilled in the art can utilize this disclosure in various implementations and modifications suitable for the specific purpose of the concept.

Claims

1. A method for extracting core elements of digital and intelligent quality management in manufacturing, characterized in that, include: The procurement data and usage data of raw materials are retrieved from the database based on the first text. The procurement data includes procurement process documents and related business information. The usage data includes raw material usage records, related production business information and related Internet of Things (IoT) information data, wherein the related IoT information data is generated by related production business. The first text sequence was obtained based on procurement and usage data; The first named entity sequence is obtained by using the first language model to extract named entities from the first text sequence. The first text sequence is processed using a hidden Dirichlet distribution to obtain the first list of elements. Vectorize the first named entity sequence to obtain the first vector set corresponding to the first named entity sequence; The second vector set corresponding to the first named entity sequence is obtained based on the relevance between the words in the first element list and the quality management topic; The degree of association between elements in the first element list is determined based on the first vector set and the second vector set. The first element list includes words and the topic distribution corresponding to those words; The process of obtaining the second vector set includes: Generate a topic-word distribution matrix for the text within the first text sequence, sort the words based on word frequency to obtain a first word list, remove words in the first word list whose word frequency is below a first threshold and whose word frequency is above a second threshold, uniformly sample the first word list to obtain a second word list, construct a first probability distribution matrix based on the topic distribution corresponding to the second word list, construct a sparse matrix based on the constructed first probability distribution matrix, and obtain the word vector corresponding to each word based on the sparse matrix; Obtain the second vector set based on the words and their corresponding word vectors; The process of determining the association between named entities within the first named entity sequence includes: The first associated word sequence corresponding to the named entity in the first named entity sequence is determined based on the first vector set, and in response to the length of the first associated word sequence corresponding to the named entity being greater than the sixth threshold, the named entity is added to the third named entity sequence. Based on the second vector set, determine the second associated word sequence that associates words within the third named entity sequence, and in response to the length of the second associated word sequence corresponding to the named entity being greater than the sixth threshold, add the named entity to the fourth named entity sequence.

2. The method for extracting core elements of digital and intelligent quality management in manufacturing as described in claim 1, characterized in that, The process of obtaining the first named entity sequence includes: The large language model is used to segment and tag the text in the first text sequence, and strings with the part of speech of noun or gerund are added to the first named entity sequence.

3. The method for extracting core elements of digital and intelligent quality management in manufacturing as described in claim 1, characterized in that, Principal analysis is used to reduce the dimensionality of either the first or second vector set, making them have a consistent dimension.

4. The method for extracting core elements of digital and intelligent quality management in manufacturing as described in claim 3, characterized in that, The third vector is obtained by normalizing the first and second vectors corresponding to the named entities in the first named entity sequence.

5. The method for extracting core elements of digital and intelligent quality management in manufacturing as described in claim 4, characterized in that, The degree of association between named entities within the first named entity sequence is determined based on the third vector. The degree of association between named entities is determined by the cosine approximation of the third vector corresponding to the named entity.

6. The method for extracting core elements of digital and intelligent quality management in manufacturing as described in claim 5, characterized in that, The first named entity sequence includes a first named entity and a second named entity; Furthermore, in response to the fact that the correlation between the first named entity and the second named entity is greater than the third threshold and the Euclidean distance is greater than the fourth threshold, the first named entity is updated to the neighbor list of the second named entity, and the second named entity is updated to the neighbor list of the first named entity. Alternatively, in response to the fact that the correlation between the first named entity and the second named entity is less than a third threshold and the Euclidean distance is less than a fourth threshold, the neighbor nodes of the first named entity are updated based on the neighbor nodes of the second named entity.

7. The method for extracting core elements of digital and intelligent quality management in manufacturing as described in claim 6, characterized in that, Retrieve named entities whose neighbor list size is greater than the fifth threshold within the first named entity sequence and add them to the second named entity sequence.

8. An electronic device, characterized in that, The device includes a memory and a processor, wherein the memory is used to store program data and the processor is used to execute the program data to implement the core element extraction method for intelligent manufacturing quality management as described in any one of claims 1-7.

Images