| In the overall promotion of Industry 4.0,5G network,industrial Internet and machine learning technology play a vital role.Utilize the advantages of 5G’s large broadband and low latency to ensure high-speed transmission of massive data,connect industrial equipment to the cloud,and apply it to scenarios such as remote equipment control,unmanned intelligent inspection,machine vision quality inspection,etc.,to realize intelligent manufacturing and automatic control And optimize scheduling,thereby improving production efficiency and product quality,reducing costs and risks.Based on the massive data collected by the industrial Internet,the establishment of accurate and reliable artificial intelligence models driven by data plays an important role in the production and decision-making of enterprises.In this paper,aiming at the load balancing problem of the power system caused by disorderly charging behavior and the problem of insufficient detection accuracy of semiconductor manufacturing defects,a machine learning algorithm is studied to achieve peak-shaving and valley-filling of power grid load and improve the detection accuracy of printed circuit board defects.The research content mainly includes the following aspects.(1)Aiming at the difficulty of defect identification and detection in semiconductor manufacturing,a variety of data enhancement methods are used to classify and identify based on deep residual convolutional neural networks;in view of the difficulty of obtaining a large number of defect samples in actual production scenarios,combined with optimized unsupervised anomaly detection Algorithm GANomaly,using only normal samples for training,has an average detection accuracy of 87% in the data set in this paper.This method provides an idea of unsupervised anomaly detection for wafer map recognition.(2)Aiming at the problem of insufficient accuracy and low efficiency of printed circuit board defect detection,based on the YOLOv5 s algorithm,through a series of improvement measures such as transfer learning,improved anchor frame size clustering algorithm,integration of Vision Transformer structure,and addition of attention mechanism,in this paper The average detection accuracy rate in the data set can reach99%,and the detection speed can reach 100 FPS,which can adapt to the scene of noncontact intelligent detection of defects in industrial production.(3)Aiming at the load impact problem brought by the disorderly charging behavior of electric vehicles to the power system,the scheme of using singular spectrum analysis to predict long-term short-term memory network and gated recurrent unit is analyzed.Feasibility in the power load forecasting task,on this basis,based on the non-dominated sorting genetic algorithm to solve the time-of-use electricity price,intelligently guide users to charge in an orderly manner,and compare the initial disordered charging behavior,and realize the power grid Load shaving peaks and filling valleys ensures the safe and stable operation of the power system.To sum up,in the field of power system and integrated circuit manufacturing,this paper establishes a machine learning model by using the massive data collected by the industrial Internet,and optimizes the load of the power grid through a multi-objective optimization algorithm,so that the load impact problem of disordered charging behavior can be obtained.The balance guarantees the stable operation of the power system;the use of YOLOv5 s algorithm and data enhancement algorithm improves the detection accuracy and detection speed,improves production efficiency and product quality,and provides strong support for enterprises to reduce costs and improve quality.The research results of this paper provide new ideas and methods for the development and intelligent upgrading of the industrial Internet. |
PDF Full Text Request