Research And Application Of Predictive Visual Function Block Diagram Running Resources

With the rapid development of intelligent manufacturing,the role of machine vision in the manufacturing process is becoming more and more important.At present,the mainstream PLC development platform has also begun to increase the machine vision function.Machine vision is packaged as an algorithm module on the PLC development platform.Developers can perform secondary development on the basis of the algorithm module,which has the advantages of short development cycle and easy maintenance.Based on the vision platform,predicting the running resources of the machine vision algorithm module can optimize industrial production scheduling,such as predicting the execution time of machine vision can help PLC to set the task execution cycle,predicting the minimum variable type group of machine vision can reduce its running space.This paper is dedicated to the research and application of framework for predicting visual FBD running resources on the Visual Function Block Diagram(FBD)platform.The research contents are as follows:(1)A framework for predicting visual FBD running resources through related features is proposed.The framework uses program instrumentation to obtain runtime features of visual FBD instead of directly using input parameters as features.Because there is image input for visual FBD,the framework uses convolutional neural networks to extract the content features of the image,then screen the features to further improve feature engineering,and finally build a model to predict the running resources of visual FBD.(2)According to the above framework,a solution for optimizing the variable type of visual FBD was constructed.Based on the phenomenon that the variable type prediction is too small,which will lead to the collapse of visual FBD,this paper solves this problem by defining the loss function.Aiming at the difficulty of obtaining labels,a special automatic label acquisition program was developed.(3)According to the framework,a program for predicting the execution time of visual FBD is constructed.Since the time required for prediction must be smaller than the execution time of visual FBD,this paper filters the features according to the time consumption of instrumented features,and then executes instrumented program to obtain the execution time of visual FBD.Finally,according to the execution time of visual FBD,three sets of schemes for extracting image content features were designed.(4)This paper designs and deploys a module to optimize the variable type of the visual FBD and a module to predict the execution time of the visual FBD on the visual FBD platform.The feasibility of the two modules is confirmed by examples.In the experimental part,this paper uses the pictures in the PASCAL VOC data set as the picture set and Canny edge detection as the research object for the experiment.In the experiment of optimizing the visual FBD variable type,the instrumented feature combined with the image content feature was used as the input,and the CNN after adjusting the loss function was used as the model.The accuracy of the model reached 87.37%,and the illegal variable type accounted for only 1.28%.In the experiment of predicting the execution time of visual FBD,using the instrumented feature combined with the image content feature as input and XGBoost as the model,the prediction result shows that the value of MAPE is only 2.09%.The two experiments have fully reached the expectations and can meet the actual application.