Research On Prediction Of Precision Milling Quality Of High-frequency Component Based On Neural Network

With the popularization of 5G technology,a golden period of leapfrog development for wireless communication is coming.The antenna is a key component in wireless communication technology and its electromagnetic performance is affected by the manufacturing accuracy.Therefore,the use of intelligent manufacturing technologies to predict the processing quality of antenna prototypes and the realization of the intelligent production of antennas,are of great significance to the development in the field of communication.Faced with the above problems,this thesis focuses on the prediction of the manufacturing accuracy of high-frequency-components in the precision milling process.The main contents of this thesis are as follows:First,based on the structural characteristics,material characteristics and processing characteristics of the high-frequency-components,the influencing factors of the processing accuracy of the high-frequency-components were analyzed,and the milling and data acquisition plans were determined;according to the structural characteristics of highfrequency-components,they were categorized into three basic structures,namely square holes,slits and round holes.Then,experimental samples were designed and the experiments were completed.Second,the quality data and processing data of the high-frequency-components obtained from the experiment were analyzed and processed;for the quality data,it was concluded that the absolute size and the dimensional accuracy were mutually independent in this experiment;for the monitoring data,the cutting force was segmented according to the processing time of each basic feature,and then based on traditional feature extraction methods,the time domain features and wavelet packet features were extracted from the cutting force data.After that,39 groups of features were selected based on the corrleation analysis and,together with cutting parameter data,form a sample dataset for prediction models.Then,the dimensional accuracy of high-frequency-components was studied and predicted by BP Neural Networks and Deep Learning Neural Networks,respectively.The F1-Score of the three processing features under the BP model reached 70% on the validation set,indicating that the model achieved satisfactroy accuracy for all three basic features;based on the GAF algorithm,the time-domain signal was converted into two-dimensional image data,which isused as input to the MLP-CNN.The results showed that the prediction accuracy of the model was significantly improved.The sizes of the square holes,round holes and slit were predicted with 88.27%,84.44%,76.08% accuracy on the validation set,respectively.Finally,a real-time analysis system for the manufacturing of high-frequencycomponents was developed and built using C#.The system had three functions: data acquisition based on NI DAQmx,quality prediction based on Python,and a process parameter history database based on MYSQL.It could be used to predict the manufacturing accuracy of high-frequency components in practice.