Research On Intelligent Planning Of Mechanical And Electrical Product Assembly Sequence Based On Knowledge Graph And Its Key Technologies

With the introduction and advancement of "Industry 4.0" and "Made in China 2025",the mining and utilization of data and knowledge have become a meaningful way to upgrade the manufacturing industry.Knowledge engineering-related technologies have once again become a research hotspot in academic and industrial circles.In recent years,knowledge graph technology has developed rapidly.This technology can formally describe objective things and their interrelationships and perform related expressions and related reasoning on data and knowledge in the manufacturing field.The assembly process is a vital link in mechanical manufacturing,and the assembly workload of mechanical products accounts for 20%-70% of the entire product manufacturing workload.Improving the level of intelligence in the assembly process is of great significance.The assembly involves tooling and fixtures,timing,etc.,and the assembly sequence is challenging to generate intelligently.The paper relies on the National Natural Science Foundation and the Guizhou Province Science and Technology Major Special Program Project.It focuses on the fundamental problem of mechanical and electrical product assembly sequence planning.The knowledge graph's assembly sequence reasoning explores the reasoning result evaluation and selection driven by the assembly sequence's rationality and tries to provide theoretical and technical reference for the intelligent generation of the assembly sequence of electromechanical products.Finally,a hollow-cup motor produced by an aerospace company in Guizhou is used as an example to verify the method proposed in this article.The main research work of the thesis is as follows:(1)Aiming at the problem that text-based assembly knowledge is complex to effectively mine and use,and cannot be directly used to guide assembly:(1)A text knowledge extraction method combining pre-trained word vectors,long-term and short-term memory neural networks,and conditional random fields is proposed,ALBERTAtt Bi LSTM-CRF is used for text-based knowledge extraction.The experimental results show that the accuracy,recall,and F1 value of this method is improved by 13.27%,5.33%,and 9.2%,respectively,compared with other methods such as Ro BERTa.(2)As there is no public text data set in the assembly field.This paper constructs an assembly text dataset Assembly Data.In order to overcome the problem of insufficient data caused by the difficulty of data collection and the time-consuming and laborious data labeling of Assembly Data,this paper proposes a combination of active learning The model migration method MTAL is:apply the ALBERT-Att Bi LSTM-CRF knowledge extraction method to the public data set CLUENER2020 to obtain pre-training weights,and then fine-tune the weights obtained on the public data set on Assembly Data to enhance the text data in the assembly field,thereby The problem of fewer data in the assembly text data set is solved,and the validity of the data set is verified.(3)Based on the assembly data set Assembly Data.The MTAL is compared with the ALBERT-Att Bi LSTM-CRF without migration.The experimental results show that the accuracy,recall,and F1 value of the knowledge extraction method proposed in this paper is increased by 3.75%,1.32%,and 2.38%,respectively.(2)For the construction of assembly knowledge graph: The knowledge source of assembly knowledge graph construction includes text data and empirical knowledge.(1)For practical knowledge that it is challenging to use data mining to process and utilize directly.This paper proposes a representation method based on ontology+rules to represent empirical knowledge and then uses the practical knowledge represented by "ontology+rules" as a supplement to text data.(2)The assembly ontology and assembly rules were constructed based on assembly tasks.The correlation analysis between ontology and knowledge graph was carried out.The mapping between resource description framework RDF and graph database Neo4 j was studied RDF data was imported into a graph database to realize Storage between heterogeneous data.Finally,based on knowledge fusion and knowledge storage,a knowledge graph is constructed to lay the foundation for subsequent reasoning.(3)For the completion and reasoning of the knowledge graph:(1)The above knowledge graph lacks part topology information and geometric information,etc.,this paper proposes a knowledge graph complement method MBD-KRL that integrates MBD model information,through the 3DXML in the MBD model Documents and STEP AP242 files for information extraction,to obtain the topological information and geometric information of the parts,complete the link prediction between different entities,improve and complete the knowledge graph,and finally build a relatively complete assembly knowledge graph,and finally in the WN18 and FB15 K data sets The completion method was verified on the previous page.The verification results show that the MBD-KRL completion result is better than the traditional methods such as Trans H,and the Hits@10 index has increased by 5.1% and 6%,respectively.(2)Based on the completed knowledge graph,the generation of assembly sequence is studied.The assembly sequence reasoning based on the knowledge graph is realized,and the inferred assembly relationship's rationality and reliability are verified.(4)Selection and evaluation of assembly sequence: The assembly sequence obtained by the above reasoning is not necessarily unique and cannot be directly used in production.This paper analyzes the association between the assembly sequence and the assembly sequence's rationality and proposes a sequence rationality-driven basis.The optimization method of assembly sequence based on sequential pattern mining.This method defines the assembly sequence unit,merges related sub-assemblies,and normalizes the sequence through assembly semantic similarity,thereby removing sequence redundancy.After comparing with the results of other mainstream sequential pattern mining algorithms such as Clo FAST,the method proposed in this paper is superior to other methods in terms of the accuracy,running time,and memory usage of the mining results.Finally,the frequent assembly sequence pattern with assembly sequence rationality information obtained by sequence pattern mining is calculated with the assembly sequence based on knowledge graph reasoning.The assembly sequence is sorted based on the calculation result.The assembly is assembled according to the sorting product—sequence scheme selection.(5)Developed a prototype system for the intelligent planning of the assembly sequence of electromechanical products,and the system was verified for application in an aerospace company in Guizhou.