CN116645129A - Manufacturing resource recommendation method based on knowledge graph - Google Patents

Manufacturing resource recommendation method based on knowledge graph Download PDF

Info

Publication number
CN116645129A
CN116645129A CN202310273431.4A CN202310273431A CN116645129A CN 116645129 A CN116645129 A CN 116645129A CN 202310273431 A CN202310273431 A CN 202310273431A CN 116645129 A CN116645129 A CN 116645129A
Authority
CN
China
Prior art keywords
resource
manufacturing
knowledge graph
processing
demand
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Pending
Application number
CN202310273431.4A
Other languages
Chinese (zh)
Inventor
刘阳
张冠伟
王磊
张奇
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Tianjin University
Original Assignee
Tianjin University
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Tianjin University filed Critical Tianjin University
Priority to CN202310273431.4A priority Critical patent/CN116645129A/en
Publication of CN116645129A publication Critical patent/CN116645129A/en
Pending legal-status Critical Current

Links

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q30/00—Commerce
    • G06Q30/02—Marketing; Price estimation or determination; Fundraising
    • G06Q30/0201—Market modelling; Market analysis; Collecting market data
    • G06Q30/0202—Market predictions or forecasting for commercial activities
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/30—Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
    • G06F16/36—Creation of semantic tools, e.g. ontology or thesauri
    • G06F16/367—Ontology
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F40/00—Handling natural language data
    • G06F40/20—Natural language analysis
    • G06F40/279—Recognition of textual entities
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G06N3/084—Backpropagation, e.g. using gradient descent
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N5/00—Computing arrangements using knowledge-based models
    • G06N5/02—Knowledge representation; Symbolic representation
    • G06N5/022—Knowledge engineering; Knowledge acquisition
    • G06N5/025—Extracting rules from data
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q30/00—Commerce
    • G06Q30/02—Marketing; Price estimation or determination; Fundraising
    • G06Q30/0201—Market modelling; Market analysis; Collecting market data
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
    • G06Q50/04—Manufacturing
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • Y02P—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN THE PRODUCTION OR PROCESSING OF GOODS
    • Y02P90/00—Enabling technologies with a potential contribution to greenhouse gas [GHG] emissions mitigation
    • Y02P90/30—Computing systems specially adapted for manufacturing

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Business, Economics & Management (AREA)
  • Physics & Mathematics (AREA)
  • Strategic Management (AREA)
  • General Physics & Mathematics (AREA)
  • Finance (AREA)
  • Development Economics (AREA)
  • Accounting & Taxation (AREA)
  • Data Mining & Analysis (AREA)
  • General Engineering & Computer Science (AREA)
  • Computational Linguistics (AREA)
  • Entrepreneurship & Innovation (AREA)
  • Artificial Intelligence (AREA)
  • Economics (AREA)
  • Marketing (AREA)
  • General Business, Economics & Management (AREA)
  • Health & Medical Sciences (AREA)
  • General Health & Medical Sciences (AREA)
  • Software Systems (AREA)
  • Computing Systems (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Evolutionary Computation (AREA)
  • Mathematical Physics (AREA)
  • Game Theory and Decision Science (AREA)
  • Human Resources & Organizations (AREA)
  • Databases & Information Systems (AREA)
  • Audiology, Speech & Language Pathology (AREA)
  • Animal Behavior & Ethology (AREA)
  • Tourism & Hospitality (AREA)
  • Primary Health Care (AREA)
  • Manufacturing & Machinery (AREA)
  • Biomedical Technology (AREA)
  • Biophysics (AREA)
  • Molecular Biology (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

The invention relates to a manufacturing resource recommendation method based on a knowledge graph, which comprises the following steps: establishing a supply and demand information model in the manufacturing field; constructing a manufacturing field ontology model, showing the concept and the relation of the demand information and the manufacturing resources, and taking the concept and the relation as a mode layer of a manufacturing field knowledge graph; carrying out knowledge extraction; realizing visual expression by using a knowledge graph; the knowledge graph embedding is realized, and in the training process, an error triplet, namely a negative sample, is obtained by randomly replacing a head entity; continuously optimizing the loss function by a random gradient descent method to obtain a qualified embedded vector; after training the vector representation of the entity by using a TransE model, measuring the coincidence degree of each resource by using the calculation of vector values, and calculating the similarity between vectors by using a cosine similarity calculation method; and on the basis of feature matching, increasing resource Qos service matching.

Description

Manufacturing resource recommendation method based on knowledge graph
Technical Field
The invention belongs to the technical field of digital processing in manufacturing industry, and particularly relates to a manufacturing resource recommendation method based on a knowledge graph.
Background
Under the background of continuous development of economic globalization and intelligent manufacturing technology, the scale of manufacturing industry is continuously enlarged, but manufacturing modes and technology platforms are behind, the positions of manufacturing resources of a plurality of enterprises are dispersed, the resources are not closely related, an efficient and synergistic product manufacturing method is difficult to form, and unreasonable resource use and excellent resource waste are caused during the product manufacturing processing. Along with the proposal of cloud manufacturing, manufacturing enterprises start to use technical services such as cloud computing, big data and the like and express various manufacturing resources, and help users to acquire resource services with higher quality, but at present, the development of cloud manufacturing is not perfect, a technical system is not complete enough, and the problems of messy information mode, incomplete resource summarization, unreasonable resource recommendation mode and the like still exist, which can cause the manufacturing enterprises to solve problems in a slower speed and a low resource recommendation quality when coping with the product manufacturing and processing demands of users.
The patent aims at the problems, designs a manufacturing resource recommending method based on a knowledge graph, expresses the manufacturing and processing requirement information and the manufacturing resource information of a product in the form of the knowledge graph, and achieves the aim of recommending the manufacturing resources required by the product processing through the manufacturing resource recommending method based on the knowledge graph. The patent aims at aiming at the product manufacturing processing, exploring a construction method of a knowledge graph in the manufacturing field, recommending manufacturing resources based on the constructed knowledge graph, and improving the service quality and the resource utilization rate of enterprises.
Disclosure of Invention
The invention aims to provide a manufacturing resource recommending method based on a knowledge graph, which is used for extracting enterprise manufacturing resource information and product processing requirement information, representing the supply and demand information in the manufacturing field in a knowledge graph mode, recommending manufacturing resources by utilizing a knowledge graph embedding method and service attribute information of the resources, and completing service response of the product manufacturing processing requirement with high quality. The technical proposal is as follows:
a manufacturing resource recommendation method based on a knowledge graph comprises the following steps:
first, establishing a supply and demand information model in the manufacturing field
Based on part characteristics, product manufacturing requirements are decomposed into different requirement units, and a product requirement information model is established:
Ru＝{ID,RTy,RN,RFu,RTq,RC}
ru represents a demand unit; ID represents a demand number; RTy represents a demand type, which includes site resources, equipment resources, and human resources; the RN represents a demand name, wherein the demand name comprises machining sites and assembly sites, equipment resource demands, the demand name comprises turning equipment and milling equipment, and the demand name comprises machining workers and assembly workers; RTq represents a time quota, also known as a machining man-hour quota; RC represents a cost limit; RFu represents a part demand feature unit;
constructing a corresponding manufacturing resource information model according to the establishment specification of the demand information model:
MR＝{MRID,MRAt,MRF,MRAb,MRQos}
wherein MR stands for manufacturing resource, comprising five different attribute units: MRID is a manufacturing resource identifier, MRAt represents a basic attribute of a resource, MRF represents a functional attribute of the resource, MRAb represents a capability attribute of the resource, and MRQos represents a quality of service attribute of the resource;
secondly, constructing a manufacturing field ontology model, showing the concept and the relation of the demand information and the manufacturing resources, and taking the concept and the relation as a model layer of a manufacturing field knowledge graph;
thirdly, knowledge extraction is carried out: performing data annotation by using a BIO annotation method, and completing an entity recognition task by using a BiLSTM-CRF model, wherein the input of the BiLSTM is a trained word vector and reflects each word in a sentence, and after model training, the output of the BiLSTM is the score of a word corresponding to each category;
fourthly, realizing visual expression by using a knowledge graph: storing and expressing the obtained data by adopting a Neo4j graph database, storing the obtained entity and relationship data into a structured CSV file, writing Neo4j import nodes and relationship programs, importing the data, and realizing the generation of a knowledge graph in the manufacturing field;
fifthly, knowledge graph embedding is achieved: embedding a knowledge graph by adopting a TransE model, mapping the entity and the relation to the same space, and training the existing triplet T= (h, r, T) in the knowledge graph in the manufacturing field to ensure that the vector of the entity and the relation meets h+r-t=0;
in the training process, obtaining an error triplet, namely a negative sample, by randomly replacing a head entity; continuously optimizing the loss function by a random gradient descent method to obtain a qualified embedded vector;
step six, training the vector representation of the entity by using a TransE model, measuring the coincidence degree of each resource by using the calculation of vector values, and calculating the similarity between vectors by using a cosine similarity calculation method;
the knowledge graph stores product manufacturing and processing demand information and manufacturing resource information, a demand unit in a product demand information model and a resource unit in a manufacturing resource information model are represented by nodes and relations, vectorization and similarity query of different nodes are utilized, manufacturing resource characteristics are matched according to manufacturing and processing characteristics in the demand unit, and therefore resource units matched with the demand unit are obtained;
step seven, on the basis of feature matching, resource Qos service matching is added, and a resource candidate set with best service quality is selected from manufacturing resources with basic features meeting the similarity requirement, so that the final candidate resource has good performance on functional attributes and service quality;
the resource Qos service is defined as four indicators of process age PT, cost level CL, process quality PQ and satisfaction S,
the resource processing aging expression is as follows:
wherein T is a Average time for completing its processing task for the selected resource, T ave The average time for completing the processing task for all the same type of processing resources;
the cost level is expressed as:
wherein C is a Cost of completing its processing task for the selected resource, C ave The average cost of completing the processing task for all similar processing resources in the enterprise;
the processing quality is expressed as follows:
wherein Q is a For the number of qualified tasks in all processing tasks of the selected processing resource, Q e To complete the total number of processing tasks;
satisfaction is expressed as:
wherein S is i Scoring the user satisfaction degree obtained by each product manufacturing and processing in the past;
and calculating weights of four indexes according to a coefficient of variation method, calculating Qos scoring scores of manufacturing resources by using the four indexes, sorting the manufacturing resources according to scoring conditions, setting up resource selection scores, and selecting one or more groups of manufacturing resources with highest ranking as a result of a manufacturing resource recommendation method based on a knowledge graph.
Further, in a first step, the first step,
basic attributes of manufacturing resources include the following three types of information:
MRAt＝{MRn,MRl,MRw}
wherein MRn represents a resource name, a static attribute of the resource, MRl represents a resource position, dynamic information and MRw represents the working condition of the resource;
the functional attributes of the manufacturing resource include the following three types of information
MRF＝{MRt,MRe,MRp}
Wherein MRt represents a machining feature type, MRe represents a machining process type, and MRp represents an applicable machined part;
the capability attributes of the manufacturing resource include the following three types of information
MRAb＝{MRz,MRm,MRpow}
Wherein MRz represents the maximum processing size, MRm represents the processable material, MRpow represents the processing power;
the quality of service attribute of the manufacturing resource includes the following three types of information
MRQos＝{MRtim,MRc,MRa}
Wherein MRtim represents processing time, MRc represents processing cost, and MRa represents processing precision.
According to the invention, the manufacturing field supply and demand information model construction module can lead the description method of the product manufacturing processing demand information and the manufacturing resource information, the manufacturing field ontology modeling module and the knowledge extraction module can fill the knowledge graph from the concept and the data, the knowledge graph visual expression module and the knowledge graph embedding module can visually express the entity and the relation and form the embedded expression, the feature matching module based on entity similarity calculation and the manufacturing resource recommendation generation module based on resource service can realize the recommendation of the optimal manufacturing resource from the basic feature and the resource service quality.
The existing manufacturing resource matching model often cannot fully integrate the internal manufacturing resources of enterprises, meanwhile, the description mode of the resource information is disordered, the difficulty of manufacturing resource matching is greatly increased, accurate manufacturing resource recommendation is difficult to provide for users, meanwhile, the manufacturing resource recommendation index is single, and the actual experience of the users on the manufacturing resource service is considered. Therefore, compared with the prior art, the method and the system can fully mine all data information in enterprises, integrate the data information in a standardized way, and realize manufacturing resource recommendation with higher quality according to manufacturing requirements and service quality.
Drawings
FIG. 1 is a system structure diagram of a manufacturing resource recommendation method based on a knowledge graph
FIG. 2 is a schematic diagram of a supply and demand information model in the manufacturing field
FIG. 3BiLSTM-CRF model basic framework schematic
FIG. 4 is a schematic diagram of a knowledge graph of the manufacturing process requirements of a product
FIG. 5 is a schematic view of a knowledge graph of a manufacturing resource
FIG. 6 schematic view of a TransE model
FIG. 7 is a schematic diagram of a product manufacturing process requirement similar unit
Detailed Description
The following describes specific embodiments of the present invention. Fig. 1 is a system configuration diagram of a manufacturing resource recommendation method based on a knowledge graph of the present invention.
As shown in fig. 1, the manufacturing resource recommendation method based on a knowledge graph (hereinafter referred to as recommendation method) of the present invention mainly includes a manufacturing domain supply and demand information model building module, a manufacturing domain ontology modeling module, a knowledge extraction module, a knowledge graph visual expression module, a knowledge graph embedding module, a feature matching module based on entity similarity calculation, and a manufacturing resource recommendation generation module based on resource service.
The recommendation system is mainly oriented to the product manufacturing process, and mainly aims to solve the problems that when manufacturing resources are selected for the product manufacturing process at present, all manufacturing resources of enterprises are difficult to fully utilize due to scattered manufacturing resources, and the quality of manufacturing resource recommendation is low due to a single manufacturing resource recommendation mode. According to the method, a formalized and unified demand information model and a manufacturing resource model are constructed, an ontology model is constructed on the basis of the demand information model and the manufacturing resource model to serve as a mode layer of a manufacturing domain knowledge graph, resource information is extracted by a knowledge extraction technology to serve as a data layer of the domain knowledge graph, the construction of the manufacturing domain knowledge graph is completed, and the visual expression of the knowledge graph is realized by Neo4 j.
After the knowledge graph of the manufacturing field is built, the knowledge graph embedding technology is utilized, the embedded representation of the entity in the knowledge graph of the manufacturing field is learned by using the TransE model, the resource unit matched with the demand unit is obtained through the calculation of the entity similarity, and finally, the manufacturing resource with the highest service quality is selected through the calculation of the resource service Qos.
The invention forms a standardized information expression method through the construction of the supply and demand model in the uniform manufacturing field, so that the information of the product manufacturing and processing requirements and the information of the manufacturing resources in enterprises can be integrated with high efficiency, and the management and the calling of the resources are convenient; the knowledge graph of the manufacturing field can be constructed to visually express the knowledge of the manufacturing field, and the entity and the relation can be expressed more directly, quickly and accurately by embedding the knowledge graph; in addition, compared with the general recommendation method, the manufacturing resource recommendation method integrating the feature and the resource service dual information is more comprehensive, and meets the requirements of users on the service quality of manufacturing resources in the intelligent manufacturing environment.
The structure and operation of each module will be described in turn.
1. Manufacturing field supply and demand information model construction module
The supply and demand information model in the manufacturing field comprises a demand information model and a manufacturing resource information model, and in order to facilitate the realization of a recommendation method, two models with uniform construction forms are required to be constructed, wherein manufacturing and processing information contained in the models corresponds to each other. Fig. 2 is a schematic diagram of a supply and demand information model in the manufacturing field.
Based on part characteristics, product manufacturing requirements are decomposed into different requirement units, and a product requirement information model is established:
Ru＝{ID,RTy,RN,RFu,RTq,RC}
ru represents a demand unit, ID represents a demand number, RTy represents a demand type, such as a site resource, an equipment resource, a human resource, etc., RN represents a demand name, such as a machining site, an assembly site, etc. included in the site resource, a turning equipment, a milling equipment, etc. included in the equipment resource, or a machining worker, an assembly worker, etc. included in the human resource, RTq represents a time allowance, also called a machining man-hour allowance, such as a time of turning an outer circle, an inner hole, RC represents a cost allowance, RFu represents a part demand feature unit, and for different resources, element forms thereof are also different, taking an equipment resource demand as an example, which can be expressed as:
RFu＝{RFc,RFm,RFma,RFs,RFa}
wherein RFc represents machining characteristics such as excircle, hole and shaft, RFm represents machining modes such as turning and milling, RFma represents part material characteristics such as 45 steel and 40cr, rfs represents part size characteristics such as diameter of hole machining or length and width of plane machining, and RFa represents required machining precision characteristics such as IT8 and ra6.3.
Constructing a corresponding manufacturing resource information model according to the establishment specification of the demand information model:
MR＝{MRID,MRAt,MRF,MRAb,MRQos}
wherein MR represents a manufacturing resource and comprises five different attribute units, MRID is a manufacturing resource identifier, MRAt represents a basic attribute of the resource, MRF represents a functional attribute of the resource, MRAb represents a capability attribute of the resource, and MRQos represents a quality of service attribute of the resource.
The basic attribute mainly comprises the following three types of information:
MRAt＝{MRn,MRl,MRw}
wherein MRn represents a resource name, a static attribute of the resource, MRl represents a resource position, dynamic information, and MRw represents a working condition of the resource.
The functional attributes mainly comprise the following three types of information
MRF＝{MRt,MRe,MRp}
Wherein MRt represents a machining feature type, such as an outer circle type machining, MRe represents a machining process type, such as a turning machining, and MRp represents a suitable machined part.
The capability attribute mainly comprises the following three types of information
MRAb＝{MRz,MRm,MRpow}
Wherein MRz represents the maximum processing dimension, MRm represents the processable material, and MRpow represents the processing power.
The service quality attribute mainly comprises the following three types of information
MRQos＝{MRtim,MRc,MRa}
Wherein MRtim represents processing time, MRc represents processing cost, and MRa represents processing precision.
After the construction of the supply and demand information model in the manufacturing field is completed, the product manufacturing and processing requirements and the manufacturing resource information can be represented by the model.
2. Manufacturing field ontology modeling module
The ontology is a knowledge representation method, which guides the construction of the ontology model in the manufacturing field according to a demand information model and a manufacturing resource information model constructed by a supply and demand information model construction module in the manufacturing field, adopts an ontology construction seven-step method to represent the concept and the relation of demand information and manufacturing resources and is used as a model layer of a knowledge graph in the manufacturing field. The construction scope is to finish all manufacturing resources of the machined parts and enterprises, and the requirement information of the past products and the reserved manufacturing resource information of the enterprises in the long-term manufacturing and machining task process are considered. And (3) continuously subdividing the manufacturing domain knowledge into two concepts of manufacturing resource knowledge and product manufacturing processing knowledge according to the related requirements of the product manufacturing processing by adopting a top-down hierarchical structure method, and continuously subdividing the two concepts downwards according to a manufacturing domain supply and demand information model. The manufacturing resource knowledge is continuously divided into human resources, equipment resources, technical resources, site resources and external resources, different resources are divided into different equipment, technologies and the like, and the product manufacturing and processing knowledge is divided into different demand units. Four concepts of hierarchical relationship and interrelationships are established, namely part-of (relationship between local and whole), kind-of (relationship between parent class and subclass), instance-of (relationship between class and instance), attribute-of (attribute of class, including object attribute and data attribute).
After concepts, relations and attributes of the ontology are defined, a model of the ontology in the manufacturing field is built by adopting a Prot g ontology modeling method.
3. Knowledge extraction module
Knowledge graph is essentially a network knowledge representation based on graph model, and the core is a triplet of "entity-relationship-entity". And extracting entities and relations in the data by adopting a knowledge extraction method to serve as a data layer of the knowledge graph. The knowledge in the manufacturing field has a lot of non-standard semi-structured or unstructured data which cannot be directly called, and the knowledge extraction function is to extract key information from the non-standard data for storage. Knowledge extraction mainly comprises entity extraction, relation extraction and attribute extraction.
Preprocessing data before knowledge extraction is performed, marking the text, and performing data marking by adopting a BIO marking method, wherein B, I, O is divided into a beginning character, an intermediate character or a tail character which represent an entity and other contents which do not belong to the entity. For example, when the product is manufactured and processed, a certain requirement is that an inner turning base is arranged on a vertical lathe turntable, a small end faces downwards, a large end faces upwards, and the inner turning base is calibrated and fixed, so that the large end excircle runout is ensured to be within 0.05, the inner profile of the semi-finish turning is divided into at least 5 layers, and the rotating speed is not lower than 80r/min, wherein the circular runout and the rotating speed are entities needing to be extracted.
The entity extraction is the extraction process of named entities in the text, and common entity extraction methods include a rule and dictionary-based method, a machine learning method and a deep learning method, and in order to ensure the effect of entity extraction, a BiLSTM-CRF model is adopted to complete the entity extraction task.
FIG. 3 is a schematic diagram of a basic framework of BiLSTM-CRF model, wherein BiLSTM is a two-way long-short term memory network, which adds a reverse calculation process relative to the long-short term memory network LSTM, and compared with the case that LSTM can only transmit information from front to back, biLSTM can fully utilize the context information of an input sequence by simultaneously accessing the input sequence to a forward LSTM and a backward LSTM and commonly accessing the two LSTM layers to an output layer. The CRF conditional random field is a decision model that functions to analyze the output and find the relationship in it for predicting the sequence. In the BiLSTM-CRF model, the input is a word vector and the output is a tag sequence. Specifically, the input of BiLSTM is a trained word vector, which reflects each word in a sentence, and after model training, the output of BiLSTM is the score of a word corresponding to each category.
And (3) recognition effect analysis: the invention uses the manufacturing resources in an enterprise and the manufacturing processing of actual products to verify the invention content, integrates the internal data and text resources to obtain 10353 pieces of data, carries out pretreatment according to the BIO labeling method, and carries out pretreatment on the data set according to 3:1:1 is divided into a training set, a verification set and a test set. In order to evaluate whether the training effect reaches the expected value, model evaluation indexes are set to be the accuracy, the recall rate and the F1 score, the model has reached a stable state after 30 training rounds, the accuracy is 96.12%, the recall rate is 95.64%, and the F1 score is 95.51%, so that the entity identification requirement is met.
By adopting the relation extraction method based on the ontology, the ontology is the expression of the concept, and the entity is the instantiation of the ontology, so that the relation between the concept and the concept in the ontology can be inherited to the entity, a large number of entities can be obtained after the entity extraction is completed, meanwhile, the entity can be labeled, the entities are paired according to the label, and the triples such as an entity A lathe and an entity B JIMK460 can be obtained, and the triples (A, part-of, B) can be obtained.
4. Knowledge graph visual expression module
Aiming at the problems of inconvenient searching, unobvious data structure and difficult mining of data relationship in the past data expression, a Neo4j graph database is adopted to store and express the data obtained by the three modules.
Neo4j is a graph database with very wide application, and by adopting the Cypher language, not only can the visualization of the knowledge graph be realized, but also the tasks of inquiry, node matching and the like can be completed. And storing the obtained entity and relation data into a structured CSV file, writing Neo4j import nodes and relation programs as shown in table 1, and automatically importing the data to realize the generation of the knowledge graph in the manufacturing field.
Table 1Neo4j importation node and relationship program
Fig. 4 is a schematic diagram of a knowledge graph of a manufacturing requirement of a certain product according to the present invention, and fig. 5 is a schematic diagram of a part of a knowledge graph of a manufacturing resource according to the present invention.
5. Knowledge graph embedding module
The knowledge graph consists of entities and relations, and in order to realize the structural representation of the product manufacturing and processing requirement information and the manufacturing resource information in the knowledge graph, the edges of the knowledge graph are expressed in the form of fact triples:
T＝(h,r,t)
wherein h represents a head entity, t represents a tail entity, and r represents a relationship of the head entity and the tail entity.
Knowledge graph embedding can represent entities and relationships in the knowledge graph as low-dimensional vectors, and is a method for entity alignment. Knowledge graph embedding provides a scoring function to represent the rationality of the triples, and then the score of the triples is continuously improved through model learning and training of the embedded representation of entities and relations. And realizing knowledge graph embedding by adopting a TransE model, and mapping the entity and the relation to the same space.
Fig. 6 is a schematic diagram of a transitional model, where the relation r is regarded as a translation operation from the head entity h to the tail entity t, and the training goal of the model is to train the existing triples in the knowledge graph in the manufacturing domain to make the vectors of the entities and the relations satisfy h+r-t=0, that is, the closer the new vector summed by the head entity and the relation vector in the triples is to the tail entity vector, the higher the correctness of the triples is.
For the correct triplet, the lower its potential energy, the closer the two entities are represented, while for the wrong triplet the higher its potential energy is expected to be the better. And selecting as much entity data as possible from the knowledge graph, extracting triples in the entity data, and constructing a relation index for the entity at the same time so as to construct a training set required by the experiment. Based on open source machine learning library TensorFlow, firstly preprocessing data, storing triples by txt files, setting parameters of a TransE model, setting an input training set to be the determined triples, entity sets and relation sets, setting an edge margin super-parameter gamma to be 1, setting the dimension k of an embedded vector to be 100, setting the learning rate to be 0.001, and setting the iteration times to be 100.
Because the triples existing in the knowledge graph are all correct, for model training, a replacement method is used for obtaining the wrong triples, in the training process, head and tail entities are randomly replaced, the wrong triples, namely negative samples, are obtained, and then a loss function is continuously optimized through a random gradient descent method, so that qualified embedded vectors are obtained.
6. Feature matching module based on entity similarity calculation
After training the vector representation of the entity by using the TransE model, measuring the coincidence degree of each resource by using the calculation of vector values, calculating the similarity between vectors by using a cosine similarity calculation method, wherein the calculation formula is as follows
Wherein a is i ∈A＝[a 1 ,a 2 ,...,a n ]，b i ∈B＝[b 1 ,b 2 ,...,b n ]N is a vector dimension value.
The knowledge graph stores product manufacturing and processing demand information and manufacturing resource information, a demand unit in a product demand information model and a resource unit in a manufacturing resource information model are represented by nodes and relations, vectorization and similarity query of different nodes are utilized, manufacturing resource characteristics are matched according to manufacturing and processing characteristics in the demand unit, and therefore the resource unit matched with the demand unit is obtained.
Fig. 7 is a schematic diagram of similar units of product manufacturing requirements, including a local knowledge graph of a certain requirement unit in a certain excircle product manufacturing requirement knowledge graph in an enterprise and a local knowledge graph of a certain 2 resource units in a manufacturing resource knowledge graph. The method comprises the steps of taking a required resource unit as a target feature through knowledge graph embedding, obtaining a 100-dimensional vector through entity training, searching for a similar feature vector through cosine similarity, obtaining a plurality of similar manufacturing resource units after preliminary screening of a candidate set in a manufacturing resource knowledge graph, and sequencing the units from high to low according to the similarity. Table 2 is the ranking information of the partially similar units.
Table 2 ranking information of partially similar units
7. Manufacturing resource recommendation generation module based on resource service
In general, manufacturing resource recommendation only meets the adaptability of features, and only considers whether the processing attribute of the resource meets the manufacturing requirement, but in the intelligent manufacturing environment, service evaluation of the manufacturing resource by a user is more and more important, and the service quality of the manufacturing resource is also becoming an important investigation factor.
The invention adds a resource Qos service matching based on the feature matching, and aims to select the resource with the best service quality from manufacturing resources with basic features meeting the requirement of similarity, so that the final candidate resource has good performance on the function attribute and the service quality.
Qos service is defined as four indexes of processing time period PT, cost level CL, processing quality PQ and satisfaction S, and resource processing time period is expressed as
Wherein T is a Average time for completing its processing task for the selected resource, T ave And the average time for completing the processing task for all the same type of processing resources.
Cost level is expressed as
Wherein C is a Cost of completing its processing task for the selected resource, C ave Is an enterpriseAnd the average cost of finishing the processing task by all the similar processing resources.
The processing quality is expressed as
Wherein Q is a For the number of qualified tasks in all processing tasks of the selected processing resource, Q e To complete the total number of processing tasks.
Satisfaction is expressed as
Wherein S is i The user satisfaction degree obtained by each product manufacturing process in the past is scored.
The Qos parameter of each resource can be used for constructing an evaluation matrix Q, wherein the Q comprises 4 evaluation indexes of n manufacturing resources.
The values of the processing quality PQ and the satisfaction S are in [0,1], and can directly participate in calculation, and the resource processing aging PT and the cost level index CL need to be normalized, and the calculation method is as follows
Wherein PT 'and CL' are indexes of processing timeliness and cost level of the normalized resources, P (i) is all the resources with processing time records in the historical database, and Q (i) is all the resources with cost records.
And determining the weight coefficient of each index by adopting a variation coefficient method:
wherein V is i As a coefficient of variation, sigma i Is the standard deviation of the two-dimensional image,is the average number of each index.
The weight omega can be obtained according to the variation coefficient i Is that
By the method, weight calculation is carried out on candidate manufacturing resources obtained through feature matching, qos scoring scores are obtained through weighting summation of four indexes, the manufacturing resources are ranked according to scoring conditions, and table 3 serves matching ranking information for the manufacturing resources.
Table 3 manufacturing resource service match ordering information
Setting up a resource selection score according to the service ranking of the manufacturing resources, and selecting one or more groups of the manufacturing resources with the highest ranking as the result of a manufacturing resource recommendation method based on the knowledge graph.

Claims (2)

1. A manufacturing resource recommendation method based on a knowledge graph comprises the following steps:
first, establishing a supply and demand information model in the manufacturing field
Based on part characteristics, product manufacturing requirements are decomposed into different requirement units, and a product requirement information model is established:
Ru＝{ID,RTy,RN,RFu,RTq,RC}
ru represents a demand unit; ID represents a demand number; RTy represents a demand type, which includes site resources, equipment resources, and human resources; the RN represents a demand name, wherein the demand name comprises machining sites and assembly sites, equipment resource demands, the demand name comprises turning equipment and milling equipment, and the demand name comprises machining workers and assembly workers; RTq represents a time quota, also known as a machining man-hour quota; RC represents a cost limit; RFu represents a part demand feature unit;
constructing a corresponding manufacturing resource information model according to the establishment specification of the demand information model:
MR＝{MRID,MRAt,MRF,MRAb,MRQos}
wherein MR stands for manufacturing resource, comprising five different attribute units: MRID is a manufacturing resource identifier, MRAt represents a basic attribute of a resource, MRF represents a functional attribute of the resource, MRAb represents a capability attribute of the resource, and MRQos represents a quality of service attribute of the resource;
secondly, constructing a manufacturing field ontology model, showing the concept and the relation of the demand information and the manufacturing resources, and taking the concept and the relation as a model layer of a manufacturing field knowledge graph;
thirdly, knowledge extraction is carried out: performing data annotation by using a BIO annotation method, and completing an entity recognition task by using a BiLSTM-CRF model, wherein the input of the BiLSTM is a trained word vector and reflects each word in a sentence, and after model training, the output of the BiLSTM is the score of a word corresponding to each category;
fourthly, realizing visual expression by using a knowledge graph: storing and expressing the obtained data by adopting a Neo4j graph database, storing the obtained entity and relationship data into a structured CSV file, writing Neo4j import nodes and relationship programs, importing the data, and realizing the generation of a knowledge graph in the manufacturing field;
fifthly, knowledge graph embedding is achieved: the method comprises the steps that a TransE model is adopted to realize knowledge graph embedding, entities and relations are mapped to the same space, a relation r is regarded as translation operation from a head entity h to a tail entity T through training of a triplet T= (h, r, T) existing in the knowledge graph in the manufacturing field, and the training target is that vectors of the entities and the relations meet h+r-t=0 through training of the triplet existing in the knowledge graph in the manufacturing field, namely, the closer new vectors and tail entity vectors are after summation of the head entity and the relation vectors in the triplet, the higher the correctness of the triplet is;
in the training process, obtaining an error triplet, namely a negative sample, by randomly replacing a head entity; continuously optimizing the loss function by a random gradient descent method to obtain a qualified embedded vector;
step six, training the vector representation of the entity by using a TransE model, measuring the coincidence degree of each resource by using the calculation of vector values, and calculating the similarity between vectors by using a cosine similarity calculation method;
the knowledge graph stores product manufacturing and processing demand information and manufacturing resource information, a demand unit in a product demand information model and a resource unit in a manufacturing resource information model are represented by nodes and relations, vectorization and similarity query of different nodes are utilized, manufacturing resource characteristics are matched according to manufacturing and processing characteristics in the demand unit, and therefore resource units matched with the demand unit are obtained;
step seven, on the basis of feature matching, resource Qos service matching is added, and a resource candidate set with best service quality is selected from manufacturing resources with basic features meeting the similarity requirement, so that the final candidate resource has good performance on functional attributes and service quality;
the resource Qos service is defined as four indicators of process age PT, cost level CL, process quality PQ and satisfaction S,
the resource processing aging expression is as follows:
wherein T is a Average time for completing its processing task for the selected resource, T ave The average time for completing the processing task for all the same type of processing resources;
the cost level is expressed as:
wherein C is a Cost of completing its processing task for the selected resource, C ave Adding for all the same kind in enterpriseAverage cost of payroll resources to complete the processing task;
the processing quality is expressed as follows:
wherein Q is a For the number of qualified tasks in all processing tasks of the selected processing resource, Q e To complete the total number of processing tasks;
satisfaction is expressed as:
wherein S is i Scoring the user satisfaction degree obtained by each product manufacturing and processing in the past;
and calculating weights of four indexes according to a coefficient of variation method, calculating Qos scoring scores of manufacturing resources by using the four indexes, sorting the manufacturing resources according to scoring conditions, setting up resource selection scores, and selecting one or more groups of manufacturing resources with highest ranking as a result of a manufacturing resource recommendation method based on a knowledge graph.
2. The method for recommending manufacturing resources based on a knowledge graph according to claim 1, wherein, in the first step,
basic attributes of manufacturing resources include the following three types of information:
MRAt＝{MRn,MRl,MRw}
wherein MRn represents a resource name, a static attribute of the resource, MRl represents a resource position, dynamic information and MRw represents the working condition of the resource;
the functional attributes of the manufacturing resource include the following three types of information
MRF＝{MRt,MRe,MRp}
Wherein MRt represents a machining feature type, MRe represents a machining process type, and MRp represents an applicable machined part;
the capability attributes of the manufacturing resource include the following three types of information
MRAb＝{MRz,MRm,MRpow}
Wherein MRz represents the maximum processing size, MRm represents the processable material, MRpow represents the processing power;
the quality of service attribute of the manufacturing resource includes the following three types of information
MRQos＝{MRtim,MRc,MRa}
Wherein MRtim represents processing time, MRc represents processing cost, and MRa represents processing precision.
CN202310273431.4A 2023-03-20 2023-03-20 Manufacturing resource recommendation method based on knowledge graph Pending CN116645129A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202310273431.4A CN116645129A (en) 2023-03-20 2023-03-20 Manufacturing resource recommendation method based on knowledge graph

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202310273431.4A CN116645129A (en) 2023-03-20 2023-03-20 Manufacturing resource recommendation method based on knowledge graph

Publications (1)

Publication Number Publication Date
CN116645129A true CN116645129A (en) 2023-08-25

Family

ID=87614164

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202310273431.4A Pending CN116645129A (en) 2023-03-20 2023-03-20 Manufacturing resource recommendation method based on knowledge graph

Country Status (1)

Country Link
CN (1) CN116645129A (en)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116993237A (en) * 2023-09-21 2023-11-03 北京上奇数字科技有限公司 Enterprise recommendation method and system based on cosine similarity algorithm
CN117150138A (en) * 2023-09-12 2023-12-01 广东省华南技术转移中心有限公司 Scientific and technological resource organization method and system based on high-dimensional space mapping

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117150138A (en) * 2023-09-12 2023-12-01 广东省华南技术转移中心有限公司 Scientific and technological resource organization method and system based on high-dimensional space mapping
CN116993237A (en) * 2023-09-21 2023-11-03 北京上奇数字科技有限公司 Enterprise recommendation method and system based on cosine similarity algorithm

Similar Documents

Publication Publication Date Title
CN106651424B (en) Power user portrait establishing and analyzing method based on big data technology
CN111612549B (en) Construction method of platform operation service system
Mena‐Chalco et al. Brazilian bibliometric coauthorship networks
CN116645129A (en) Manufacturing resource recommendation method based on knowledge graph
CN112463980A (en) Intelligent plan recommendation method based on knowledge graph
CN110826320A (en) Sensitive data discovery method and system based on text recognition
CN107016068A (en) Knowledge mapping construction method and device
Jusoh et al. Application of MCDM approach to evaluate the critical success factors of total quality management in the hospitality industry
CN114860916A (en) Knowledge retrieval method and device
CN101000624A (en) Method, system and device for implementing data mining model conversion and application
US20240127143A1 (en) Method, device and storage medium for information processing based on data interaction
CN113626571A (en) Answer sentence generating method and device, computer equipment and storage medium
CN113313470A (en) Employment type evaluation method and system based on big data
CN116501979A (en) Information recommendation method, information recommendation device, computer equipment and computer readable storage medium
CN112162972A (en) Human resource bidirectional recommendation system based on data mining and privacy protection technology
CN115982429A (en) Knowledge management method and system based on flow control
Papić et al. Multi-criteria decision-making in the tourism domain: The past, present and future of the research field
CN110727801A (en) Ontology-based fuzzy evaluation search engine advertisement optimization method
CN114493172B (en) Emergency capacity allocation plan deduction method and system
He et al. [Retracted] Design and Application of College Student Management System Based on Big Data Technology
CN113222471A (en) Asset wind control method and device based on new media data
Jia [Retracted] Exploratory Research on the Practice of College English Classroom Teaching Based on Internet and Artificial Intelligence
CN117556118B (en) Visual recommendation system and method based on scientific research big data prediction
KR20040023376A (en) Real-time quality measurement method of bibliographic database
Müller et al. Forecasting the Impact of Artificial Intelligence on the Labor Market Through Natural Language Processing of Job Posting

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination