Research On Key Semantic Inference Methods For Industrial Control Protocols Based On Multimodal Industrial Data

The Industrial Internet connects equipment,production lines,plants,suppliers,products and users together in an open and interconnected industrial network platform,enabling automated and intelligent production methods and promoting the manufacturing industry in the direction of high quality.However,at present,different devices connected to the Industrial Internet usually adopt private industrial control protocols,which can often only communicate between devices of the same manufacturer,and cannot communicate between devices of different manufacturers,which greatly hinders the pace of China’s intelligent manufacturing.The premise of interconnection and interoperability of production factors in the Industrial Internet is the interoperability and translation of industrial control protocols,and the problem of resolving unknown industrial control protocols must be solved.The industry generally uses protocol reverse analysis techniques to obtain protocol specifications such as protocol format and semantics,of which the most critical issue is how to parse the field semantics of IPC protocols.To this end,this thesis studies the semantic identification of industrial control protocols and proposes a semantic inference method for key fields of industrial control protocols based on industrial field multimodal data.The main research content is as follows:(1)Study the semantic channel mining method for multimodal industrial data.According to the characteristics of industrial field environment and multimodal industrial data,we design a method to collect industrial control image modal information and log modal information,and then propose a semantic channel mining method for image modal data based on image differential association and a semantic channel mining method for log modal data based on natural language processing algorithm.Finally,this paper designs a large number of experiments to verify the effectiveness of the proposed semantic channel mining method based on the collected multimodal data.The experimental results show that the accuracy rate of semantic channels for image modality recognition reaches 94%,and the accuracy rate of semantic channels for log modality recognition reaches 96%.(2)Research on key semantic inference algorithms for industrial control protocols.By mining frequent events and extracting candidate message regions,a protocol semantic analysis algorithm based on sequence alignment is proposed to infer the spatio-temporal correlation between semantic channels and protocol message stanzas,find out the message fragments in messages that are positively correlated with the change pattern of semantic channels,and obtain the protocol message locations corresponding to key semantics.The experimental results show that this method achieves an accuracy of 85% for semantic channel detection and 73% for field detection,while the recognition rate and coverage rate of message formats both reach 100%.Compared with other semantic inference algorithms,this method shows significant superiority in semantic channel detection,message segment identification performance and message format reconstruction performance.