Production Workshop Data Monitoring And Anomaly Perception Based On Dynamic Data Flow Analysis

In the era of'Industry 4.0',information systems and physical systems are deeply integrated,the manufacturing industry that needs to be reformed enjoys the convenience brought by data information.In addition,intelligent manufacturing derived from big data and traditional workshops,which has become a research hot spot.Data from the manufacturing plant is collected by sensors,which can indirectly reflect the operating status of the manufacturing plant,such as the process of workpiece processing,machine health,product quality,personnel deployment,order adjustment,etc.In order to tap the hidden information of massive data,artificial intelligence technologies such as neural network and deep learning are combined with workshop production,and the data analysis capability becomes one of the important evaluation indicators of the highly intelligent performance for intelligent plant.Industrial data analysis includes clustering,classification,and regression techniques.With the intelligence of the manufacturing plant being rapidly enhanced,the derived data flow has the characteristics of multi-source,continuity,time-series,novelty,etc.,and the traditional data analysis methods have suffered from low adaptability,poor generalization ability,reduced accuracy,etc.The problem of data stream information that cannot be fully mined is gradually coming into focus.Moreover,due to the complexity and diversity of real scenarios,it is challenging to mine the information of data streams in dynamic environments,especially the study of time series features and concept drift features has become difficult.For the data stream of non-stationary environment,this paper completes the research on monitoring and anomaly sensing of dynamic data flow characteristics,which is mainly studied as follows:(1)For the problem of dynamic data stream monitoring in intelligence plant,the proposed GNG-L model with higher adaptability based on the GNG network in this paper,which is composed of weight adaption,node deletion,and node addition mechanisms.Firstly,the mechanism of weight adaptation is proposed by analyzing the changes of the local characteristics for data streams,which ensures the network topology is adjusted accurately and quickly.Secondly,the adaptive deletion mechanism removes neural nodes that are no longer updated due to the evolution of data streams.Finally,the generation mechanism is trigged when the new feature of data stream evolution needs to be described in new regions of the feature space.(2)Combined with a processing case,the monitoring model ofNO_x and combustion efficiency based on the improved GNG model is built,and then the index screening is completed by using principal component analysis.The strategy of combining monitoring results and index control is proposed to complete the real-time dynamic control of the processing process.(3)Based on the monitoring results of the dynamic data stream feature monitoring model,we complete the data anomaly awareness analysis.Based on the studies in(1)and(2),a hybrid prediction model GS-GMDH is proposed,which can perform timely and accurate feature monitoring and singularity identification for time series time and make prediction analysis.