Design And Implementation Of Switch Installation Error Correction System Based On Deep Neural Networks

With the development of industry and the advancement of science and technology,more and more new technologies are gradually being applied to industrial production.This has greatly reduced the unnecessary burden on workers,making the work of workers more and more simple and error-free.It also greatly improves the production efficiency.Among these new technologies,deep neural networks have been in full swing in recent years and have been widely used in various fields.Of course,the application in image recognition is indispensable.Image recognition technology based on deep neural networks is now more mature,and the error rate of recognition is now much lower than human eye recognition.So at present,some industrial inspections are gradually transformed from human visual inspection to computer vision inspection.Nowadays,in the workshop production line of an automobile factory,it is necessary to manually install a car switch button panel and manually check whether the installation is correct.Because workers need to be familiar with the production plan and switch button information prior to installation,this increases the difficulty of the work.In addition to repeating the same operation for a long time,the workers are very prone to fatigue,and it is inevitable that there will be mis-installation and missing-loading.If the wrong panel is installed on the vehicle and you want to change it later,the cost is enormous.So in order to solve this problem,this paper designs and implements of switch installation error correction system based on deep neural networks.The system mainly implements two functions: The first part is to realize the semi-automated guide function,so that the workers can correctly install the switch without understanding the production plan;the second part is to test the completed panel to ensure that the switch button is installed correctly.For the first part of the system's functions,this paper uses programmable logic controllers(PLCs)to coordinate industrial control units,sensors,and other components to work together.It can guide the worker to take the corresponding key when installing the switch,without having to understand the production plan can be installed correctly.The second part of the function of the system is the focus of this paper.First,we segmented and sampled all the switch button images,and made more training samples.Then we designed the corresponding neural network structure and trained the best model.Finally,we applied the model to switch detection.In order to complete the functional requirements of the second part,this paper has conducted in-depth research on the deep neural networks and conducted detailed studies on several models that are currently relatively hot,such as: deep belief network,recurrent neural network and convolutional neural network.Familiar with the principles and training methods of these three neural networks,this article selected the above three networks as the network structure of the project to train 3000 switching samples on the network,and then selected the best network model for this project.This paper finally selects the convolutional neural network model as the network model of the project.Its recognition accuracy is close to 98%,which is much higher than the accuracy of human eye recognition.This method fully meets the production requirements.After the completion of the system,the production test of this system has resulted in satisfactory results.The system runs smoothly,stably and has strong robustness.It fully meets the production requirements of the depot.This system has been officially applied in workshop production,and it runs smoothly and stably in the workshop.