AI-based Automatic Coding Method for BOM Material Information

Abstract

The present invention discloses an AI-based automatic coding method for BOM material information. First, the data structures of the material lists at different stages of the product are unified and stored as a single set of material list data. Single data with unique part numbers is screened, and some of the data is compared with the standard database. When they do not match, they are first cleaned and then compared. Knowledge graphs, vector conversion, cosine distance, etc. can also be used for further processing to convert the single set of material list data into standard data. For data that cannot be matched, multiple strategies can be used to select matching entries. It also involves the format recognition of the material list and the selection of processing tools to ensure data integrity. When denoising the single set of material list data, key parameters are extracted by combining multiple data sources, and association rules are mined through data mining algorithms and denoising operations are performed. The present invention can improve the accuracy and standardization of material data, greatly improve the coding efficiency, and realize intelligent and refined material management.

Description

Technical Field

[0001] The present invention relates to the field of material informatization. More specifically, the present invention relates to an automatic coding method for BOM material information based on AI. Background Art

[0002] In many industries such as manufacturing and electronic information industries, the Bill of Materials (BOM), as the core document describing the product structure, details all kinds of material information required for the product. Efficient management of material information, especially accurate and unified coding, is the key for enterprises to optimize production processes, control costs, and collaborate in the supply chain. Traditional methods of material information coding rely mostly on manual operations and have many insurmountable problems. The defects that cannot be overcome by manual coding are as follows: First, the efficiency is low, affecting the progress of enterprise procurement, production, inventory and other processes; second, the accuracy cannot be guaranteed, triggering a chain reaction in subsequent procurement, production, inventory management and other links; third, there is a lack of unified standards, making it impossible to share material coding information. Therefore, it is urgent to develop a method that can automatically, accurately and efficiently code the material information in the Bill of Materials. With the rapid development of artificial intelligence (AI) technology, its applications in various industries are becoming increasingly widespread, bringing new opportunities to solve the material coding problem. AI, with its powerful data analysis, pattern recognition and prediction capabilities, can theoretically achieve automatic coding of material information and is expected to break through the limitations of traditional manual coding. However, in the actual process of material management, many factors pose great challenges to the automatic coding of material information based on AI.

[0003] First, in the actual process of material management, the formats of Bills of Materials used by different departments or different business processes within an enterprise vary widely. In the R & D department, to facilitate recording product design ideas, the Bill of Materials may focus on the technical specifications and design parameters of materials, with a relatively flexible format; while in the production department, to facilitate arranging production processes and material distribution, its Bill of Materials pays more attention to the production process and assembly sequence of materials, with a different format. In an electronic manufacturing enterprise, the R & D department may list materials according to functional modules, while the production department may arrange them according to the production line sequence, resulting in a chaotic data structure of the Bill of Materials and making it difficult to directly convert it into a unified standard list. The formats of Bills of Materials provided by suppliers are even more diverse, some presented in Excel spreadsheet form, some in PDF documents, and even some suppliers provide them in custom electronic document formats, which requires enterprises to spend a lot of time and manpower on format conversion and data sorting when integrating Bill of Materials data.

[0004] Secondly, data errors in the bill of materials are extremely common. Problems such as misspelled material names, ambiguous descriptions of material specifications and models, and inaccurate records of material packaging forms are frequent. Take an automobile manufacturing enterprise as an example. In the bill of materials for purchasing engine parts, the name of a key part may be misspelled, leading to serious problems in subsequent communication with suppliers and quality traceability. These incorrect data make the bill of materials unable to be directly used as a standard list because the standard list requires accurate data. Otherwise, it will have a chain of negative impacts on the entire production process.

[0005] Finally, there are also serious information missing problems in the bill of materials. Some bills of materials may lack key material group information, which makes it impossible for enterprises to clearly distinguish the categories of materials, and is not conducive to the classification management and inventory counting of materials. Some bills of materials may not record the item numbers of materials, resulting in the inability to accurately determine the installation positions of materials during the production and assembly process, affecting production efficiency and product quality. In terms of material usage information, there may also be inaccurate or missing data, which will bring great troubles to the formulation of procurement plans and inventory management, making it difficult for the bill of materials to be transformed into a standard list and unable to provide reliable data support for the production and operation of enterprises.

[0006] With the development of intelligent manufacturing, the degree of automation and informatization of the production process is constantly increasing, and the requirements for the standardization of the bill of materials are also getting higher and higher. Automated production lines require accurate standard lists to achieve automatic material distribution and assembly. The development of the industrial Internet also requires enterprises to be able to carry out efficient data sharing and collaborative work with suppliers and partners based on standard lists. However, the above problems of the current bill of materials pose challenges to the automatic coding of material information, seriously hindering the digital transformation and intelligent development process of enterprises. Summary of the Invention

[0007] An object of the present invention is to solve at least the above problems and provide at least the advantages described hereinafter.

[0008] Another object of the present invention is to provide an AI-based automatic coding method for BOM material information, which can improve the accuracy and standardization of material data, greatly improve the coding efficiency, realize intelligent and refined material management, break the information barriers inside and outside the enterprise, enhance collaboration, optimize processes such as production and procurement, and provide strong support for enterprises to reduce costs and increase efficiency.

[0009] To achieve these objects and other advantages of the present invention, there is provided an AI-based automatic coding method for BOM material information, including:

[0010] S1. Unify the data structure of the bill of materials (BOM) developed at any stage of the product, and store it as a collection of multiple single BOM data. Any single BOM data shall include at least the material name, material group, material specification model, item number, material packaging form, and material quantity.

[0011] S2. Use AI to identify the item number and material quantity in the BOM.

[0012] S3. Screen out each single BOM data with unique item number data, and extract four data, namely the material name, material specification model, item number, and material packaging form, from each single BOM data. Compare the four data of the single BOM data with the data in the pre-constructed standard database. If the four data of the single BOM data exactly match an entry in the standard database, replace the single BOM data with the corresponding entry in the standard database but retain the item number and the corresponding material quantity data in the single BOM data to form single BOM standard data; otherwise, after cleaning the single BOM data, compare it with the data in the standard database again. If the four data of the cleaned single BOM data exactly match an entry in the standard database, replace the single BOM data with the corresponding entry in the standard database but retain the item number and the corresponding material quantity data in the single BOM data to form single BOM standard data.

[0013] S4. Use AI to generate a unique model code based on the material specification model, item number, and material packaging form in the single BOM standard data, and code the material in the form of material group plus unique model code.

[0014] Among them, the method for cleaning the single BOM data is: correct the spelling mistakes of the material name, material specification model, and material packaging form according to the pre-constructed standard vocabulary list in the field of electronic devices.

[0015] Preferably, for the single BOM data that still cannot be exactly matched with the data in the standard database after cleaning in step S3, the following further processing is carried out:

[0016] A1. Utilize the pre-constructed knowledge graph in the field of electronic components to find the key parameter data of the material from the single BOM data. The key parameter data of the material includes component type, packaging form, and performance parameters.

[0017] A2. Unify the expression forms of the same type of key parameters, and convert the key parameter data of the material into a low-dimensional digital vector through a vector conversion model.

[0018] A3. After assigning weights to different key parameter data, combine all the key parameter vector data by weighting to generate a comprehensive vector of the single BOM data.

[0019] A4. Calculate the cosine distance between the comprehensive vector of the single bill of materials data and the comprehensive vectors of each entry in the standard database through the cosine distance algorithm. If the obtained cosine distance exceeds the set matching threshold, replace the single bill of materials data with the corresponding entry in the standard database, but retain the material usage data in the single bill of materials data to form the single bill of materials standard data.

[0020] Preferably, if the number of entries in A4 whose obtained cosine distance exceeds the matching threshold is greater than 1, then:

[0021] Sort the multiple entries whose cosine distance exceeds the matching threshold in ascending order of cosine distance, and select the entry with the smallest cosine distance as the corresponding entry in the standard database.

[0022] Preferably, if the number of entries in A4 whose obtained cosine distance exceeds the matching threshold is greater than 1, then:

[0023] Sort the multiple entries whose cosine distance exceeds the matching threshold in ascending order of cosine distance, select the two entries with the smallest and the second smallest cosine distances, further calculate the difference between the cosine distances corresponding to the two entries. If the difference is not greater than 0.05, select the entry with the smallest cosine distance as the corresponding entry in the standard database; otherwise, further count the number of material key parameter data of the two entries, and select the entry with the larger number of material key parameter data as the corresponding entry in the standard database.

[0024] Preferably, A4 further includes the following steps:

[0025] B1. Establish a historical matching database and calculate the frequency of each entry in the standard database being selected during the historical matching process;

[0026] B2. When the number of entries whose obtained cosine distance exceeds the matching threshold is greater than 1, select the entry with the highest frequency of being selected during the historical matching process as the corresponding entry in the standard database.

[0027] Preferably, step S1 further includes format recognition of the bill of materials formulated at any stage of the product and selection of the corresponding processing tool to preliminarily detect the data in the bill of materials, and return the bill of materials lacking any data such as material name, material group, material specification model, item number, material packaging form, and material usage to the bill of materials provider to supplement the missing information.

[0028] Preferably, in the step S1, the steps for unifying the data structure of the bill of materials formulated at any stage of the product include:

[0029] Perform format recognition on the bill of materials formulated at any stage of the product, which includes:

[0030] C1. Identify file format A through the file extension and identify file format B through the file header;

[0031] C2. Use multiple rules in the file format judgment rule library to respectively judge file format A and file format B, quantify the judgment results of various rules into scores, assign weights to each score, and form the comprehensive score of file format A and the comprehensive score of file format B according to the weighted combination of scores and weights. Select the file format with the larger comprehensive score as the file format of the bill of materials;

[0032] According to the identified file format, select the corresponding processing tool to extract various data in the bill of materials and generate a single bill of materials data with a unified data structure.

[0033] Preferably, the unified data structure in step S1 includes at least the following fields:

[0034] The material name field is used to clearly identify the specific name of the material, and its data type is a string;

[0035] The material group field is used to group materials according to the classification criteria of function, source or production stage, and its data type is a string;

[0036] The material specification and model field is used to record the technical parameters and model specifications of the material, and its data type is a string;

[0037] The item number field is used to identify the specific position number of the material in the product structure, and its data type is a string;

[0038] The material packaging form field is used to record the packaging type of the material, and its data type is a string; and

[0039] The material usage information field is used to record the quantity of the material, and its data type is an integer.

[0040] Preferably, the method for screening the item number data in the bill of materials in step S3 to have uniqueness includes: reading a bill of materials data, comparing its item number with the recorded item numbers, and updating the frequency information. If the frequency information of this item number is greater than 1, it is determined as a non-unique item number.

[0041] Preferably, noise data needs to be removed before cleaning the single bill of materials data. The method for removing the noise data includes:

[0042] D1. Take the single bill of materials data as the core data source, determine its material name information, and extract the key parameter data of this material;

[0043] D2. Then, using the design drawings of the material product, production process documents, and material descriptions provided by suppliers as data sources, and aiming at the differences in data formats of different data sources, after unified processing, extract the key parameter data of materials from one or more sources.

[0044] D3. Combine the data extracted in steps D1 and D2 to synthesize a comprehensive feature vector to represent the material.

[0045] D4. Use data mining algorithms to mine the frequent item sets and association rules between different features in the feature vector.

[0046] D5. When reading a single material list data to be denoised, first determine its material name information, match the single material list data with the same material name in the single material list data, then extract the frequent item sets and association rules between the corresponding different features, and perform denoising operations on this single material list data to be denoised according to these frequent item sets and association rules.

[0047] The present invention has at least the following beneficial effects:

[0048] First, the AI-based automatic coding method for BOM material information provided by the present invention realizes the unification of the data structure of the material list, clarifies the recording specifications of key information such as material names and material groups, not only ensures the integrity and consistency of data from the source, but also simplifies the data processing volume for subsequent coding work. Through the comparison and cleaning mechanism with the standard database, correct spelling mistakes based on the standard vocabulary list in the field of electronic devices, making the material data conform to industry standards, and effectively avoiding information misunderstandings and incorrect transmissions caused by non-standard data.

[0049] Second, the AI-based automatic coding method for BOM material information provided by the present invention runs through the entire process of enterprise material management from data collection, processing, standardization to coding of the material list; in the production link, accurate material coding helps to accurately formulate production plans and material distribution; in the procurement link, it is convenient for efficient communication with suppliers; in the inventory management link, it is conducive to inventory counting and control, comprehensively improving the refinement level of enterprise material management, optimizing the enterprise operation process, reducing operation costs, and enhancing the competitiveness of the enterprise.

[0050] Thirdly, for the material data that cannot be directly matched with the standard database, the AI-based automatic BOM material information coding method provided by the present invention uses the knowledge graph in the field of electronic components, the vector conversion model and the cosine distance algorithm technology. By finding the key parameter data of the material, converting it into a low-dimensional digital vector, and calculating the cosine distance, the intelligent matching of complex material data is realized. When facing new types of electronic components, even if their data formats are special, the corresponding entries in the standard database can be efficiently found, significantly improving the applicability and flexibility of material coding;

[0051] Fourthly, when the cosine distance matching result is not unique in the AI-based automatic BOM material information coding method provided by the present invention, multiple strategies are adopted for processing, such as sorting by cosine distance and selecting the smallest entry, or further calculating the cosine distance difference between the smallest and the second smallest entries, combining with the quantity of material key parameter data, and even considering factors such as historical matching frequency. These strategies optimize the matching decision from multiple dimensions, adapt to different scenarios, ensure the selection of the most practical standard entry, and improve the accuracy and reliability of material coding;

[0052] Fifthly, the AI-based automatic BOM material information coding method provided by the present invention discloses the process of format recognition and processing of the bill of materials. The file format is recognized through the file extension and the file header, and the quantization judgment result is made using the format judgment rule library, and the comprehensive score is formed by weighted combination to determine the format, and then the corresponding tool is selected to extract the data, so that the automatic material information coding method can adapt to bills of materials in multiple formats, ensure the accurate extraction of data from bills of materials from different sources, provide effective data support for subsequent standardization and coding work, and ensure the coherence and efficiency of the material management process;

[0053] Sixthly, the AI-based automatic BOM material information coding method provided by the present invention discloses using data mining algorithms to remove noise data. Supported by multi-source data such as material product design drawings and production process documents, by extracting key parameter data, merging to form a comprehensive feature vector, and mining frequent item sets and association rules, the noise in a single bill of materials data can be accurately identified and removed; ensuring the purity of the coded data and laying a solid foundation for the accuracy of subsequent coding.

[0054] Other advantages, objectives and features of the present invention will be partially reflected by the following description, and partially will be understood by those skilled in the art through the research and practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] Figure 1 It is a schematic flowchart of the AI-based automatic BOM material information coding method in a technical solution of the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0056] The present invention will be further described in detail below in conjunction with the accompanying drawings, so that those skilled in the art can implement it with reference to the text of the specification.

[0057] It should be understood that terms such as "having", "including", and "comprising" used herein do not exclude the presence or addition of one or more other elements or combinations thereof.

[0058] As Figure 1 shown, the present invention provides an AI-based automatic coding method for BOM material information, including:

[0059] S1. Unify the data structure of the bill of materials formulated at any stage of the product, and store it as a set of multiple single-item bill of materials data. Any single-item bill of materials data includes at least material name, material group, material specification model, item number, material packaging form, and material usage.

[0060] S2. Use AI to identify the item number and material usage in the bill of materials.

[0061] S3. Screen out each single-item bill of materials data with a unique item number, and extract four data: material name, material specification model, item number, and material packaging form from each single-item bill of materials data. Compare the four data of the single-item bill of materials data with the data in the pre-constructed standard database. If the four data of the single-item bill of materials data exactly match an entry in the standard database, replace the single-item material data with the corresponding entry in the standard database, but retain the item number and the corresponding material usage data in the single-item material data to form single-item material standard data; otherwise, after cleaning the single-item bill of materials data, re-compare it with the data in the standard database. If the four data of the cleaned single-item bill of materials data exactly match an entry in the standard database, replace the single-item material data with the corresponding entry in the standard database, but retain the item number and the corresponding material usage data in the single-item material data to form single-item material standard data.

[0062] S4. Use AI to generate a unique model code based on the material specification model, item number, and material packaging form in the single-item material standard data, and code the material in the form of material group plus unique model code.

[0063] Among them, the method for cleaning the single-item bill of materials data is: correct the spelling mistakes of the material name, material specification model, and material packaging form according to the pre-constructed standard vocabulary list in the field of electronic devices.

[0064] In the above technical solution, a specific implementation of step S1 is as follows: Use a format recognition tool written in Python to analyze the format of the bill of materials developed at any stage of product R & D, production, quality inspection, inventory, etc. By identifying the file extension, such as.xlsx,.csv,.pdf, etc., and the specific identification information in the file header, determine the file format to which the bill of materials belongs. For the identified different file formats, call the corresponding data extraction plug-ins. If it is an Excel file (.xlsx), use the pandas library to read the data, and extract information such as material name, material group, material specification model, item number, material powder packaging form, and material usage according to the pre-set column name mapping relationship; if it is a PDF file, use OCR (Optical Character Recognition) technology, combine regular expressions to match key information, and perform structured processing after extraction. Further, the extracted material information is stored as a collection of multiple single-item bill of materials data according to a unified data structure. Among them, a storage form of each single-item bill of materials data is as follows: Material name: specific name; Material group: affiliated group; Material specification model: detailed specification; Item number: unique number; Material packaging form: packaging type; Material usage: specific quantity.

[0065] In the above technical solution, a specific implementation of step S2 is as follows: Use the BERT model as an AI tool and further add a relation reasoning module. In the above technical solution, a specific implementation of step S3 is as follows: Write a Python screening program. This program uses a loop structure to traverse a single bill of materials data set. During the traversal, a hash table is used to record the item numbers that have appeared and their occurrence times. For each bill of materials data, its item number is extracted and the hash table is queried. If the item number does not appear in the hash table, it is added to the hash table and the occurrence time is recorded as 1; if it has appeared, its occurrence time is updated; Single bill of materials data with an item number occurrence time of 1 is screened out. For the screened single bill of materials data, four data items, namely the material name, material specification model, item number, and material packaging form, are extracted. Using an SQL query statement, these four data items are compared with a standard database pre-built in the MySQL database. Each record in the standard database contains complete material information and has been reviewed by industry experts and verified by historical data. If the four data items of the single bill of materials data exactly match an entry in the standard database, the material name, material group, material specification model, item number, material packaging form, and other data of the single bill of materials data are replaced with the corresponding entry in the standard database, while retaining the material usage data in the original single bill of materials data to form single bill of materials standard data. If a complete match cannot be achieved, the cleaning process is entered. According to a pre-built standard vocabulary table in the field of electronic devices, which is stored in JSON format and contains the correct expressions and incorrect spellings of common material names, material specification models, and material packaging forms, the string matching algorithm, such as the KMP (Knuth-Morris-Pratt) algorithm, is used to match the material name, material specification model, and material packaging form one by one. If an incorrect spelling is found, it is corrected according to the correct expression in the vocabulary table; after correction, the SQL query statement is used again to compare with the standard database. If the match is successful, the single bill of materials standard data is formed in the above manner.

[0066] In the above technical solution, a specific implementation of step S4 is as follows: According to material categories such as capacitors, resistors, inductors, integrated circuits, etc., a unique material group coding rule is preset. For example, capacitors are represented by "C", resistors by "R", inductors by "L", and integrated circuits by "IC". An AI is used to generate a unique model coding based on the material specification model, item number, and material packaging form in a single piece of material standard data. The rule is as follows: For the material specification model part: First, condense the key information of the material specification model. Taking capacitors as an example, first extract the key numerical parts of the capacitance value and withstand voltage value, discard the unit information (because the material group already indicates it is a capacitor, and the unit can be defaulted), and perform digital processing. For example, if the capacitance value is 10 μF, take 10, and if the withstand voltage value is 50 V, take 50, and then splice them into 1050. If the material specification model contains multiple parameters, key parameters can be selected according to importance or a fixed order for processing to ensure that this part of the coding reflects the material characteristics to a certain extent and has 4 digits. For the item number part, the item number usually contains letters and numbers. When coding, only take the letter part of the item number and convert the letters to numbers (such as A = 01, B = 02, C = 03,... Z = 26) to ensure that this part of the coding can identify the item number and has 2 digits. For the material packaging form part, a mapping table between the material packaging form and digital coding is established. For example, "surface mount" is coded as "01", "plug-in" is coded as "02", "BGA package" is coded as "03". If there are many types of packaging forms, hierarchical coding or compression coding methods can be used to ensure that the coding length is 2 digits. Finally, the material is coded in the form of material group - unique model coding to ensure the uniqueness and traceability of the coding.

[0067] As another specific embodiment of the above technical solution, as follows:

[0068] During the product R & D stage of an electronic product manufacturing enterprise, the design team formulated detailed bill of materials, which exist in different formats such as Excel spreadsheets, PDF documents, etc. First, step S1 is executed. The enterprise's technical personnel use data processing software to unify the data structure of the bill of materials: By identifying the key information in different format files, information such as material name, material group, material specification model, item number, material packaging form, and material quantity used is extracted and stored as a collection of multiple single-piece bill of materials data.

[0069] Then, step S2 is carried out. An AI tool is used to identify the item number and material quantity used in the bill of materials, fix the material quantity used by the item number, and ensure that the material quantity used is not replaced when replacing the entries in the standard database.

[0070] Next, perform step S3. The technician uses a screening program to screen out each single Bill of Materials (BOM) data with unique part numbers. Taking capacitor materials as an example, when the screening program reads a BOM data of a capacitor with a part number of "R005", the program compares it with the recorded part numbers and finds that this part number does not appear repeatedly and belongs to a unique part number. Then, four data, namely the material name, material specification model, part number, and material packaging form, in this single BOM data are extracted and compared with a pre-constructed standard database. During the comparison process, it is found that the material name of this capacitor, "electrolytic capacitor", is different from the expression "electrolytic capacitor" in the standard database and cannot be fully matched. Therefore, according to the standard vocabulary list pre-constructed in the field of electronic devices, the material name is corrected, and "electrolytic capacitor" is modified to "electrolytic capacitor". After comparing with the standard database again, the corresponding entry is successfully matched, and this single material data is replaced with the corresponding entry in the standard database, while retaining the part number and the original material usage data.

[0071] After completing the above steps, enter step S4. The technician uses a coding algorithm to code the materials in the single material standard data. Taking this capacitor as an example, according to the coding rules, for the material specification model part, the capacitance value of 10 μF is taken as 10, and the withstand voltage value of 50 V is taken as 50, and then the two are concatenated into 1050. For the part number part, the letter part is converted into a numerical value of 18. For the material packaging form part, it is converted into a number of 04. The material group is "C", and a "-" is used to connect the material group and the unique model code. Then the code is C-10501804.

[0072] In the above technical solution, the present invention has at least the following technical effects:

[0073] 1. By unifying the data structure of the Bill of Materials in step S1, the originally chaotic material information becomes orderly and standardized, which is conducive to data sharing and collaboration among various departments within the enterprise, reducing communication costs and errors caused by inconsistent data formats. In step S2, the part number is bound to the material usage to ensure the accuracy of the material usage in the BOM. In step S3, it is compared with the standard database and cleaned according to the standard vocabulary list to further ensure the accuracy and standardization of the material information, improving the quality of the material information and providing a reliable data basis for subsequent production, procurement and other links;

[0074] 2. The screening and comparison mechanism in step S3 can quickly and accurately identify the BOM data that needs to be standardized, and correct and replace it in a timely manner, which greatly shortens the time for sorting and standardizing the material information, improves the efficiency of material management, and greatly speeds up the progress of new product R & D and production;

[0075] 3. In step S4, encoding is performed according to the various characteristics of the materials, making the encoding closely associated with the key information of the materials. This not only ensures the accuracy of the encoding but also enables the detailed information of the materials to be quickly traced through the encoding during subsequent production, inventory management, etc. Once a quality problem or a change in production requirements occurs, the relevant materials can be quickly located, and corresponding measures can be taken, improving the enterprise's control ability over materials and the flexibility in dealing with problems.

[0076] In another embodiment of the present invention, for the single material list data that still cannot be fully matched with the data in the standard database after cleaning in step S3, the following further processing is performed:

[0077] A1. Using the pre-constructed knowledge graph in the field of electronic components, find the key parameter data of the material from this single material list data. The key parameter data of the material includes component type, package form, and performance parameters.

[0078] A2. Unify the expression forms of the same type of key parameters, and through a vector conversion model, convert the key parameter data of the material into low-dimensional digital vectors.

[0079] A3. After assigning weights to different key parameter data, combine all the key parameter vector data by weighting to generate a comprehensive vector of this single material list data.

[0080] A4. Through the cosine distance algorithm, calculate the cosine distance between the comprehensive vector of this single material list data and the comprehensive vectors of each entry in the standard database. If the obtained cosine distance exceeds the set matching threshold, then replace this single material data with the corresponding entry in the standard database, but retain the material usage data in the single material data to form single material standard data.

[0081] In the above technical solution, the present invention further optimizes the processing method for the single material list data that still cannot be fully matched with the data in the standard database after cleaning. A specific implementation manner is as follows:

[0082] Set a single material list data (taking a capacitor as an example) that still cannot be fully matched with the standard database after cleaning. This single material list data is as follows:

[0083] Material name: Capacitor; Material group: Electronic components; Material specification model: 10μF 50V; Designation: R005; Material package form: SMD; Material usage: 5.

[0084] Using the pre-constructed knowledge graph in the field of electronic components (such as the Corning knowledge graph), find the key parameter data of this capacitor. These key parameter data include but are not limited to:

[0085] Component type: Extracted from the material name, which is a capacitor; Package form: SMD; Performance parameters: Extracted from the material specification model, which is 10uF 50V, representing the capacitance value of 10uF and the rated voltage of 50V respectively.

[0086] To ensure that the key parameter data from different sources or in different formats can be uniformly processed subsequently, it is necessary to standardize the expression forms of these relationship parameters. For example: for the component type capacitor, it is uniformly expressed as capacitor (which can be refined into more specific types according to actual needs, such as electrolytic capacitor, ceramic capacitor. For simplicity, it is assumed that no further refinement is made here); for the package form SMD, it is uniformly expressed as SMD; for the performance parameters 10μF 50V, it is uniformly expressed as 10μF50V. The key parameter data of the material after the unified expression form is converted into a low-dimensional digital vector through a vector conversion model (for example: Word2Vec model). For example: capacitor is converted into the vector [1,0,0]; SMD is converted into the vector [0,1,0]; 10μF 50V is converted into the vector [0,0,1]. Further, weights are assigned to different key parameter data. For example: the weight of the component type is 0.4; the weight of the package form is 0.3; the weight of the performance parameters is 0.3. Then the comprehensive vector of this single material list data = 0.4×[1,0,0] + 0.3×[0,1,0] + 0.3×[0,0,1] = [0.4,0.3,0.3]. Assume that the comprehensive vector after conversion of entry 1 in the standard database is [0.5,0.3,0.2], and the comprehensive vector after conversion of entry 2 is [0.3,0.4,0.3].

[0087] Further use the cosine distance formula (for example, use the cosine_similarity function in the scipy library) to calculate the cosine distance between this single material list data and entry 1 and entry 2. Assume that the cosine distance between this single material list data and entry 1 is 0.8, and the cosine distance between this single material list data and entry 2 is 0.6. Set the matching threshold to 0.7, then replace this single material data with entry 1 but retain the material usage data in the single material data to form the single material standard data.

[0088] The above technical solution uses a pre-constructed knowledge graph in the field of electronic components to find the key parameter data of materials from a single bill of materials data, covering multiple dimensions such as component type, package form, and performance parameters, overcoming the limitations of matching only based on material name, specification model, part number, and package form. This method can more deeply understand the essential characteristics of materials and avoid matching failures caused by inconsistent surface information. For example, for a capacitor, not only its simple name and specification are considered, but also specific types of capacitors (such as electrolytic capacitors, ceramic capacitors, etc.) and more detailed performance parameters (such as capacitance value, withstand voltage value, error range, temperature characteristics, etc.) are extracted, making the description of materials more complete and improving the matching accuracy with entries in the standard database. The key parameter data of materials is converted into low-dimensional digital vectors and weighted combined, and then the similarity with each entry in the standard database is calculated through the cosine distance algorithm, realizing the quantitative evaluation of the similarity between materials. Compared with traditional text matching methods, this calculation method based on vectors and cosine distance can more accurately measure the similarity between materials. Especially when facing materials with similar but not exactly the same descriptions, it can more reasonably find the closest matching item. For example, for the subtle differences in the descriptions of the same type of capacitor by different manufacturers (such as "10uF±5% 50V" and "10uF 50V"), through vector conversion and cosine distance calculation, their similarity can be better judged, rather than simply considering them as different materials.

[0089] In another embodiment of the present invention, if the number of entries in A4 whose obtained cosine distance exceeds the matching threshold is greater than 1, then:

[0090] Sort the multiple entries whose cosine distance exceeds the matching threshold in ascending order of cosine distance, and select the entry with the smallest cosine distance as the corresponding entry in the standard database. Sorting the multiple entries whose cosine distance exceeds the matching threshold in ascending order of cosine distance and selecting the entry with the smallest cosine distance as the corresponding entry in the standard database, this strategy ensures that when multiple similar entries exist, the most similar one is preferentially selected, reducing incorrect matches caused by inaccurate partial information and ensuring that the single-piece material standard data finally formed is as close to the actual situation as possible

[0091] In another embodiment of the present invention, if the number of entries in A4 whose obtained cosine distance exceeds the matching threshold is greater than 1, then:

[0092] Sort the multiple entries with cosine distances exceeding the matching threshold in ascending order of cosine distance, select the two entries with the smallest and the second smallest cosine distances, further calculate the difference between the cosine distances corresponding to the two entries. If the difference is not greater than 0.05, select the entry with the smallest cosine distance as the corresponding entry in the standard database; otherwise, further count the number of material key parameter data of the two entries, and select the entry with the larger number of material key parameter data as the corresponding entry in the standard database. Select the entry with the larger number of material key parameter data as the corresponding entry in the standard database. This strategy comprehensively considers the similarity between entries and the richness of key parameters, avoids misselection due to minor cosine distance differences, and at the same time takes into account the integrity of key parameter information, improving the rationality of selecting the most matching entry.

[0093] In another embodiment of the present invention, step A4 further includes the following steps:

[0094] B1. Establish a historical matching database and calculate the frequency of each entry in the standard database being selected during the historical matching process;

[0095] B2. When the number of entries with cosine distances exceeding the matching threshold is greater than 1, select the entry with the highest frequency of being selected during the historical matching process as the corresponding entry in the standard database. The above technical solution utilizes the experience of historical data. For some common material matching scenarios, the appropriate entry can be more accurately selected according to past matching experience, especially applicable to materials with fuzzy descriptions or frequently appearing in different projects, improving the reliability and stability of the matching.

[0096] In another embodiment of the present invention, step S1 further includes format recognition of the bill of materials formulated at any stage of the product and selection of the corresponding processing tool to preliminarily detect the data in the bill of materials, and return the bill of materials lacking any data such as material name, material group, material specification model, item number, material packaging form, and material usage to the bill of materials provider to supplement the missing information.

[0097] In the above technical solution, by identifying the format of the bill of materials (BOM) developed at any stage of the product and using the corresponding processing tool to preliminarily detect the data in the BOM, the integrity of the BOM can be systematically checked. For a file containing multiple single BOM data entries, this technical solution can ensure that each data entry contains key information, such as material name, material group, material specification model, item number, material packaging form, and material quantity. Selecting the corresponding processing tool to parse the BOM can convert BOMs from different sources and in different formats into a unified data structure. This helps eliminate data processing difficulties caused by format differences, enabling subsequent operations (such as screening, matching, coding, etc.) to be based on a consistent data format, thereby improving the coherence and standardization of the entire material information processing process.

[0098] In another embodiment of the present invention, in the step S1, the steps for unifying the data structure of the BOM developed at any stage of the product are as follows:

[0099] Identify the format of the BOM developed at any stage of the product, including:

[0100] C1. Identify file format A through the file extension and identify file format B through the file header;

[0101] C2. Use multiple rules in the file format judgment rule library to respectively judge file format A and file format B, quantify the judgment results of various rules into scores, assign weights to each score, and form the comprehensive scores of file format A and file format B according to the weighted combination of scores and weights. Select the file format with the larger comprehensive score as the file format of the BOM;

[0102] According to the identified file format, select the corresponding processing tool to extract various data in the BOM and generate single BOM data with a unified data structure.

[0103] In the above technical solution, the present invention further optimizes the method for unifying the data structure of the BOM. Among them, when identifying the format of the BOM, a specific implementation method is as follows:

[0104] Assume that the extension of the read BOM file is ".csv" (file format A), and the file header of the file contains special identification information and is identified as file format B.

[0105] The file format judgment rule library contains multiple rules. Among them, the rules for file format A (.csv) are as follows:

[0106] Check whether the file is separated by commas. The score for this rule is 30 points and the weight is 0.3;

[0107] Check whether the file contains a header. The score for this rule is 40 points, and the weight is 0.5;

[0108] Check whether the file has specific suffix information. The score for this rule is 20 points, and the weight is 0.2.

[0109] The rules for file format B are as follows:

[0110] Check whether the file header contains the "ProductBOM" identifier. The score for this rule is 50 points, and the weight is 0.6;

[0111] Check whether the date format in the file header conforms to "YYYY-MM-DD". The score for this rule is 30 points, and the weight is 0.4.

[0112] For file format A (.csv), upon inspection, the file is comma-separated, scoring 30 points; it contains a header, scoring 40 points; but the suffix information does not meet the requirements, scoring 0 points. The comprehensive score = 0.3×30 + 0.5×40 + 0.2×0 = 29 points. For file format B, upon inspection, the file header contains the "ProductBOM" identifier, scoring 50 points; the date format meets the requirements, scoring 30 points. The comprehensive score = 0.6×50 + 0.4×30 = 42 points.

[0113] In summary, since the comprehensive score of file format B is greater than that of file format A, we select file format B as the file format for this bill of materials.

[0114] The above technical solution uses the file extension to identify file format A, uses the file header to identify file format B, and judges these two formats respectively according to multiple rules in the file format judgment rule library. Quantify the judgment results of various rules into scores, and then form a comprehensive score through weighted combination of scores and weights, so as to accurately judge the true format of the bill of materials. This multi-dimensional format judgment method avoids the one-sidedness and inaccuracy of judging file formats relying solely on a single factor (such as file extension). Identifying and detecting the format at the initial stage of data processing can discover and solve potential format problems at the source, and prevent errors caused by format problems in subsequent data processing steps (such as screening, cleaning, encoding, etc.).

[0115] In another embodiment of the present invention, the unified data structure in step S1 includes the following fields:

[0116] The material name field is used to clearly identify the specific name of the material, and its data type is string;

[0117] The material group field is used to group materials according to classification criteria such as function, source or production stage, and its data type is string;

[0118] The material specification and model field is used to record various technical parameters and model specifications of the material, and its data type is string;

[0119] The item number field is used to identify the specific location number of the material in the product structure, and its data type is string;

[0120] The material packaging form field is used to record the packaging type of the material, and its data type is string; and

[0121] The material usage information field is used to record the quantity of the material, and its data type is integer.

[0122] In the above technical solution, by clearly defining a unified data structure, including the material name field, material group field, material specification and model field, item number field, material packaging form field, and material usage information field, and specifying the data type for each field, the information in the bill of materials is clearly classified and presented. Specifying the data type for each field ensures data consistency and standardization, avoids various errors caused by inconsistent data types, and improves the data quality and reliability of the entire material information system.

[0123] In another embodiment of the present invention, the method for screening the item numbers in the bill of materials with uniqueness in step S3 includes: reading a piece of bill of materials data, comparing its item number with the recorded item numbers, and updating the frequency information. If the frequency information of this item number is greater than 1, it is determined as a non-unique item number, accurately screening unique item numbers, reducing the risk of item number confusion, and helping to improve the efficiency and pertinence of material information screening. Ensuring the uniqueness of the item numbers in the bill of materials data participating in subsequent processing can improve the quality of the entire material information processing. Because subsequent operations (such as comparing a single piece of bill of materials data with a standard database or performing coding) are based on accurate item numbers, it can reduce incorrect matches or inaccurate coding results caused by item number problems.

[0124] In another embodiment of the present invention, noise data needs to be removed before cleaning a single piece of bill of materials data. The method for removing noise data includes:

[0125] D1. Taking a single piece of bill of materials data as the core data source, determining its material name information, and extracting the key parameter data of this material;

[0126] D2. Then using the product design drawings of this material, production process documents, and material descriptions provided by suppliers as data sources, and after unified processing for the data format differences of different data sources, extracting the key parameter data of the material from one or more sources;

[0127] D3. Combine the data extracted in steps D1 and D2 to synthesize a comprehensive feature vector to represent the material;

[0128] D4. Use data mining algorithms to mine the frequent item sets and association rules among different features in the feature vector;

[0129] D5. When reading a single material list data to be denoised, first determine its material name information, match the single material list data with the same material name in the single material list data, then extract the frequent item sets and association rules among the corresponding different features, and perform denoising operations on this single material list data to be denoised according to these frequent item sets and association rules.

[0130] The above solution further optimizes the method for removing noise from single material list data. A specific implementation is as follows:

[0131] The data of a single capacitor material list is as follows:

[0132] Material name: Capacitor; Material group: Electronic components; Material specification model: 10μF 50V; Location number: R005; Material packaging form: SMD; Material usage: 5.

[0133] Take the above single material list data as the core data source, determine its material name information as "Capacitor", and extract the key parameter data of the material. The key parameter data includes capacitance value: 10μF; Rated voltage 50V; Tolerance ±10%.

[0134] The other data sources of this material are the product design drawings of the material, production process documents, and material descriptions provided by the supplier. Among them, the key parameter information extracted from the product design drawings (CAD files corresponding to CAD file parsing tools) is length: 5mm; Width: 3mm; Pin pitch: 2mm. The key parameter information extracted from the production process documents (PDF files corresponding to PDF parsing tools) is operating temperature range: -20°C to 85°C. The key parameter information extracted from the material description provided by the supplier (text files corresponding to text parsing tools) is lifespan: 1000 hours; Maximum ripple current: 1A.

[0135] Integrate the above-extracted information into a comprehensive feature vector, in the form of [capacitor, 10 μF, 50 V, ±10%, SMD, R005, 5, 5 mm, 3 mm, 2 mm, -20°C to 85°C, 1000 hours, 1 A]. Use data mining algorithms (such as the Apriori algorithm) to mine the frequent item sets and association rules among different features in the feature vector. First, use the create_transaction_list function to convert the above comprehensive feature vector into a transaction list, and then use pandas to convert the transaction list into a data frame for subsequent use of the Apriori algorithm. Set the apriori function, set the minimum support to 0.1, and use the column names as item set elements to find the frequent item sets.

[0136] When reading another single bill of materials data to be denoised as follows:

[0137] Material name: capacitor; material group: electronic components; material specification model: 10 μF 50 V; item number: R005; material packaging form: SMD; material quantity: 5.

[0138] According to the frequent item sets and association rules mined previously, we can find that information such as tolerance, size, operating temperature range, lifespan, and maximum ripple current is usually associated with "capacitor". Therefore, if this data to be denoised lacks this information, we can supplement or modify it according to the frequent item sets and association rules to make it more complete and accurate, and the result is:

[0139] Material name: capacitor; material group: electronic components; material specification model: 10 μF 50 V; item number: R005; material packaging form: SMD; material quantity: 5; length: 5 mm (supplemented); width: 3 mm (supplemented); pin pitch: 2 mm (supplemented); operating temperature range: -20°C to 85°C (supplemented); lifespan: 1000 hours (supplemented); maximum ripple current: 1 A (supplemented).

[0140] The above technical solution realizes the multi-dimensional collection of material information by taking the single bill of materials data as the core data source and combining multiple data sources such as product design drawings, production process documents, and material descriptions provided by suppliers, and extracting the key parameter data of materials; merging the information from different data sources and synthesizing a comprehensive feature vector helps to uniformly represent the scattered material information, condensing the information of multiple dimensions into one vector, and providing a comprehensive and structured data form for subsequent data processing. By mining frequent item sets and association rules, targeted denoising operations can be performed to improve the consistency and reliability of quantity processing. After denoising, matching errors or coding errors caused by noise data can be avoided, and the reliability and success rate of subsequent processing steps are improved. For example, for the bills of materials of "capacitors" provided by multiple different suppliers, the same denoising logic can be used, so that the finally processed data has the same data structure and information integrity, avoiding data inconsistency problems caused by different information expressions of different suppliers.

[0141] The equipment quantities and processing scales described here are used to simplify the description of the present invention. Applications, modifications, and variations of the AI-based BOM material information automatic coding method of the present invention will be apparent to those skilled in the art.

[0142] Although the embodiments of the present invention have been disclosed as above, they are not limited to the applications listed in the specification and embodiments. It can be fully applied to various fields suitable for the present invention. For those familiar with the field, additional modifications can be easily made. Therefore, without departing from the general concept defined by the claims and the equivalent scope, the present invention is not limited to the specific details and the illustrated and described examples here.

Claims

1. An AI-based automatic coding method for BOM material information, characterized in that Including: S1. Unify the data structure of the bill of materials (BOM) formulated at any stage of the product, and store it as a set of multiple single BOM data. Any single BOM data includes at least material name, material group, material specification model, item number, material packaging form, and material usage. S2. Use AI to identify the item number and material usage in the BOM. S3. Screen out the single BOM data with unique item number data, and extract four data: material name, material specification model, item number, and material packaging form from each single BOM data. Compare the four data of the single BOM data with the data in the pre-constructed standard database. If the four data of the single BOM data exactly match an entry in the standard database, replace the single BOM data with the corresponding entry in the standard database but retain the item number and the corresponding material usage data in the single BOM data to form single BOM standard data; otherwise, after cleaning the single BOM data, compare it with the data in the standard database again. If the four data of the cleaned single BOM data exactly match an entry in the standard database, replace the single BOM data with the corresponding entry in the standard database but retain the item number and the corresponding material usage data in the single BOM data. For the single BOM data that still cannot exactly match the data in the standard database after cleaning, further process as follows: A1. Use the pre-constructed knowledge graph in the field of electronic components to find the key material parameter data from the single BOM data. The key material parameter data includes component type, packaging form, and performance parameters. A2. Unify the expression forms of the same type of key parameters, and convert the key material parameter data into low-dimensional digital vectors through a vector conversion model. A3. After assigning weights to different key parameter data, weighted combine all the key parameter vector data to generate a comprehensive vector of the single BOM data. A4. Calculate the cosine distance between the comprehensive vector of the single BOM data and the comprehensive vectors of each entry in the standard database through the cosine distance algorithm. If the obtained cosine distance exceeds the set matching threshold, replace the single BOM data with the corresponding entry in the standard database but retain the material usage data in the single BOM data to form single BOM standard data. S4. Use AI to generate a unique model code according to the material specification model, item number, and material packaging form in the single BOM standard data, and code the material in the form of material group plus unique model code. Among them, the method for cleaning the single BOM data is: correct the spelling mistakes of the material name, material specification model, and material packaging form according to the pre-constructed standard vocabulary in the field of electronic devices.

2. The AI-based automatic coding method for BOM material information according to claim 1, wherein If the number of entries with the cosine distance exceeding the matching threshold in A4 is greater than 1: Sort the multiple entries with the cosine distance exceeding the matching threshold from small to large according to the cosine distance, and select the entry with the smallest cosine distance as the corresponding entry in the standard database.

3. The AI-based automatic coding method for BOM material information according to claim 1, characterized in that If the number of entries in A4 whose cosine distance exceeds the matching threshold is greater than 1, then: Sort the multiple entries whose cosine distance exceeds the matching threshold in ascending order of cosine distance, select the two entries with the smallest and the second smallest cosine distances, further calculate the difference between the cosine distances corresponding to the two entries. If the difference is not greater than 0.05, select the entry with the smallest cosine distance as the corresponding entry in the standard database; otherwise, further count the number of material key parameter data of the two entries, and select the entry with the larger number of material key parameter data as the corresponding entry in the standard database.

4. The AI-based automatic coding method for BOM material information according to claim 1, wherein A4 also includes the following steps: B1. Establish a historical matching database and calculate the frequency of each entry in the standard database being selected during the historical matching process; B2. When the number of entries whose cosine distance exceeds the matching threshold is greater than 1, select the entry with the highest frequency of being selected during the historical matching process as the corresponding entry in the standard database.

5. The AI-based automatic coding method for BOM material information according to claim 1, wherein Step S1 also includes format recognition of the bill of materials developed for any stage of the product and selecting the corresponding processing tool to preliminarily detect the data in the bill of materials, and returning the bill of materials lacking any data such as material name, material group, material specification model, item number, material packaging form, and material usage to the bill of materials provider to supplement the missing information.

6. The AI-based automatic coding method for BOM material information according to claim 5, characterized in that, In the step S1, unifying the data structure of the bill of materials developed for any stage of the product includes the following steps: Perform format recognition on the bill of materials developed for any stage of the product, including: C1. Identify file format A through the file extension and identify file format B through the file header; C2. Use various rules in the file format judgment rule library to respectively judge file format A and file format B, quantify the judgment results of various rules into scores, assign weights to each score, and form the comprehensive score of file format A and the comprehensive score of file format B according to the weighted combination of scores and weights. Select the file format with the larger comprehensive score as the file format of the bill of materials; According to the identified file format, select the corresponding processing tool to extract various data in the bill of materials and generate a single bill of materials data with a unified data structure.

7. The AI-based automatic coding method for BOM material information according to claim 6, characterized in that The unified data structure in step S1 includes at least the following fields: Material name field, used to clearly identify the specific name of the material, and its data type is string; Material group field, used to group materials according to classification criteria such as function, source, or production stage, and its data type is string; Material specification model field, used to record the technical parameters and model specifications of the material, and its data type is string; Item number field, used to identify the specific position number of the material in the product structure, and its data type is string; Material packaging form field, used to record the packaging type of the material, and its data type is string; and Material usage information field, used to record the quantity of the material, and its data type is integer.

8. The AI-based automatic coding method for BOM material information according to claim 1, characterized in that, The method for screening the item number data in the bill of materials in step S3 to have uniqueness includes: reading a bill of materials data, comparing its item number with the recorded item numbers, and updating the frequency information. If the frequency information of the item number is greater than 1, it is determined as a non-unique item number.

9. The AI-based automatic coding method for BOM material information according to claim 1, wherein Before cleaning a single bill of materials data, noise data needs to be removed. The method for removing noise data includes: D1. Taking the single bill of materials data as the core data source, determining its material name information, and extracting the key parameter data of the material; D2. Then taking the product design drawings of the material, production process documents, and material descriptions provided by the supplier as data sources. After performing unified processing for the data format differences of different data sources, extracting the key parameter data of the material from one or more sources; D3. Merging the data extracted in steps D1 and D2 to synthesize a comprehensive feature vector to represent the material; D4. Using a data mining algorithm to mine the frequent item sets and association rules between different features in the feature vector; D5. When reading a single bill of materials data to be denoised, first determine its material name information, match the single bill of materials data with the same material name in the single bill of materials data, then extract the frequent item sets and association rules between the corresponding different features, and perform denoising operations on the single bill of materials data to be denoised according to these frequent item sets and association rules.

Images