Research On On-line Monitoring Technology Of Parts Machining Process Based On Deep Learning

In the process of machining,surface roughness is an important index to measure machining quality,and tool condition is the key factor to affect machining efficiency.On-line monitoring technology can realize real-time monitoring of machining process without stopping the machine or testing at the interval of machining.Therefore,it is of great significance to carry out on-line monitoring technology for surface roughness and tool condition in order to ensure high quality and efficient machining of parts.In this paper,combined with deep learning method,surface roughness prediction and tool condition monitoring technology are studied respectively.Threedimensional acceleration sensor is used to collect dynamic signals in the process of machining,and deep-level features of data are automatically extracted by deep learning method,and surface roughness prediction model driven by data mechanism hybrid is established,The tool condition monitoring model of multi-layer extreme learning machine and the deep transfer learning framework under different machining conditions are established to realize the accurate and efficient monitoring of machining process.The specific work of this paper is as follows:Firstly,the formation mechanism and common monitoring methods of surface roughness and tool wear are introduced.This paper briefly introduces the evaluation standard of surface roughness of parts and related influencing factors,discusses the tool wear forms and common blunt standards,analyzes the causes of tool wear and introduces the common monitoring methods of machining process.Secondly,a prediction model of surface roughness driven by data and mechanism was established.The stacked denoising autoencoder is used as the data-driven prediction method and the unsupervised layer by layer greedy training algorithm is used to extract the features automatically from the collected dynamic signals.The roughness theory model was used as the mechanism driven prediction method,and the surface quality was predicted after combining the theoretical calculated values with the signal characteristics into new vectors.The derivative of the actual output of the model against the theoretical calculated values was added to the network loss function as a constraint.On this basis,the surface roughness classification and regression tasks were completed simultaneously to mine the similar information among the tasks,and then the high precision and strong generalization roughness hybrid driven prediction was realized.Then,the tool condition monitoring model and the deep transfer learning framework are established.The multi-layer extreme learning machine network is used to extract the deep features of the data,improve the calculation speed of the model,and avoid falling into the local minimum in the iteration process.Using Coral loss measured data distribution of the differences between the source domain and target domain and add it to the modeling of loss function.Based on the labeled source domain data and the unlabeled target domain data,a reliable tool condition monitoring model in the target domain was obtained by minimizing the distribution difference between the domains and the classification error of the source domain at the same time,and the end-to-end knowledge transfer was realized.Finally,an online monitoring scheme for the machining process based on vibration signals is designed,a software and hardware system for data acquisition is built and developed,and the proposed deep learning method is tested and verified respectively.The results show that the proposed surface roughness prediction method can effectively improve the convergence rate and generalization ability of the model,and reduce the prediction error of the model.The tool condition monitoring model of multi-layer extreme learning machine can greatly shorten the training time of the model on the basis of ensuring accurate monitoring.After the deep transfer learning,the accuracy of target domain monitoring can be improved as a whole,and the knowledge transfer of tool condition monitoring under different processing conditions can be effectively realized,and the reliability and intelligence of the process monitoring can be improved.