| The mutual reference and integration of the new generation of artificial intelligence technology and advanced manufacturing technology is the development trend of intelligent manufacturing.As the basic resource of the manufacturing industry,CNC machine tools have an important impact on the implementation of intelligent manufacturing due to their high speed,high precision,and intelligence.Through the analysis and modeling of the CNC machine tool feed system,people can have a deeper understanding of the internal characteristics of the feed system,and can better realize the control and compensation.However,due to the complexity of the internal structure of the actual system and the external operating environment,the accuracy of the model based on assumptions and simplified mathematical methods is difficult to meet the increasingly high contour error compensation application requirements.Therefore,this article is based on the running process data of the machine tool feed system and applies a new generation of artificial intelligence technology to conduct in-depth research on the accurate modeling of the feed system and the method of contour error compensation.While improving the machining accuracy of the machine tool,it explores a new generation of manual labor.The fusion method of intelligent technology and mathematical method and its application technology in the field of intelligent manufacturing.Studying and cognizing the dynamic response characteristics of the feed system is the basis for establishing its accurate model and realizing error compensation.In this paper,by studying the actual response law of single-axis and multi-axis synthesis of the feed system,and the simulation response law of its multi-domain physical model,the actual response of the feed system to the command is divided into repeatable mathematical components and repeatable non-mathematical components.Furthermore,the judgment index and calculation formula of the model prediction accuracy and whether the model can be used for compensation are put forward,which lays the foundation for the research of accurate modeling of the feed system and the method of contour error compensation.Aiming at the problem that the mathematical model based on hypothesis and simplification is only an approximation of the real system with a certain deviation and the parameter identification is complicated,this paper adopts the adaptive transformation of the classic long short-term memory network(LSTM)and proposes a nonlinear autoregressive long short-term memory network(NAR-LSTM)establishes a data-driven model of the feed system.To improve the prediction effect of highly non-linear motion states such as start-up and reversal,through automatic classification of motion states,samples are extracted and sub-models of different motion states are constructed.The prediction error of the built model at different feed speeds of the test curve is 2.2 ～ 6.3 μm,reaching the micron-level prediction accuracy,but the prediction accuracy of the model at high speed is still insufficient.Aiming at the problem that a single mathematical method or data-driven method has limited ability to describe the input-output mapping relationship of the feed system,this paper proposes a mathematical equation based on residual learning and a NAR-LSTM fusion modeling method.This method divides the feed system model into two components: the basic model based on mathematical equations and the residual model based on NAR-LSTM network.The basic model is used to describe the repeatable mathematical component in the actual response of the feed system,and the residual model is used to simulate the repeatable nonmathematical component.The fusion model reached a prediction accuracy of 2.4 ～ 4.5 μm on the test curve,which further improved the prediction accuracy and compensability of the model.Aiming at the problem that the iterative calculation of compensation value based on the offline model is inefficient and difficult to meet the problem of online real-time calculation,this paper proposes the reinforcement learning calculation of the compensation value of the contour error and the dual-code joint control compensation method.In order to quickly calculate the compensation value,an improved time series deep Q-network(TS-DQN)method is proposed,which simplifies the iterative calculation of the compensation value in each control cycle to the compensation parameter identification and the compensation value calculation;in order to send the compensation value to the CNC The system assists the processing in real time,and designs a contour error compensation method based on dual-code joint control.The compensation value is compiled into the i-code file,and the G-code and i-code are synchronously analyzed and controlled during processing.Experimental results show that this method can significantly improve the contour accuracy of parts.To verify the effectiveness of the method proposed in this paper,the integration and experimental verification of the feed system modeling and machining contour error compensation algorithm in the numerical control system are completed.In the actual machining of the round table and the convex hemisphere on the vertical machining center,the contour error was reduced from 17.3 μm to 7.6 μm,and from 16.8 μm to 7.1 μm;the cam,intake cam and exhaust cam lift were machined on the cam grinder.The errors were reduced from 30.7 μm to15.2 μm,and from 20.2 μm to 13.5 μm,respectively,which verified the effectiveness and versatility of the feed system modeling and contour error compensation methods proposed in this paper.The research work in this paper provides a certain theoretical basis and technical guarantee for the modeling of machine tool feed system and contour error compensation research,and will play a positive role in promoting the popularization and practical application of contour error compensation. |
PDF Full Text Request