Research On Production State Detection Algorithm Based On Deep Learning

With the rapid development of intelligent manufacturing and Internet of Things technology,the demand for production status detection system in the production process of industrial products is increasing day by day,and efficient and reliable detection algorithm is essential for the detection system.Due to the wide use of deep learning,the production status detection system using this technology has been applied to industrial production at present,which can improve the production efficiency in the industrial assembly line and achieve the purpose of convenient management.How to design and implement a production state detection system with good detection effect,strong stability and saving human resources has become a research hotspot in this field.Currently,the production status detection system still has the following deficiencies:(1).The current system does not fully utilize the collected relevant data,and the image data and the sensor numerical data are not used uniformly,which leads to incomplete training of the production state detection algorithm;(2).The loss function in the production state detection algorithm has a low matching degree with the optimization target,and does not have specificity,and does not subdivide the tasks under different scenes;(3).The whole system lacks a remote visual and manageable platform.In industrial production,acquisition equipment,control equipment and other related equipment are numerous,and the management is relatively complex.Moreover,a convenient operation interface platform is also needed for the replacement of models.Aiming at the above problems,this paper designs and implements a production status detection algorithm system based on deep learning,which mainly includes the following contents:(1).This paper studies the method based on machine learning,realized the image data and the feasibility of the sensor is a unified analysis of numerical data,which,in view of the image data using convolution neural network to extract the features of numerical data for sensor adopts features combination and crossover method for processing,then the two data characteristics for Mosaic unified use,eventually improve the utilization rate of the data.(2).In this paper,the degree of match algorithm,the network Loss function,in view of the low matching degree this shortage,this paper proposes a new Loss function Loss Loss function(Gap),Gap Loss instead of cross entropy Loss function is adopted to improve the model training,the experimental results show that compared with the cross entropy Loss function,this method can accelerate the convergence rate of the model,the production status is the classification of the detection system model is more accurate and more effective.(3).Finally,the production status detection system is implemented in this paper.The platform combines edge side and cloud,integrates real-time production status detection,data management and data analysis into one,which ensures the stability and reliability of the system while convenient operation,and also verifies the feasibility of the optimization algorithm.