Research Of Self-adaptive Machining System For CNC Milling Based On The Neural Network

We often come up against the condition of the chipping allowance changing fast in NC machining, especially in machining complex curved surface. In this condition lower feed rate is adopted, that leads to lower machining efficiency. Constant cutting load self-adaptive control system can solve this problem well.The self-adaptive machining system for CNC milling was implemented with the following steps. First the mapping model between the cutting parameters and characteristic currents about cutting load was set up by using Neural Network. Then the characteristic current was monitored and compared with the current about nominal cutting load. According to the compared result, the feed rate was adjusted to keep the cutting load invariable. As a result, the machining efficiency is improved greatly and machine tool is well protected. Moreover, products of high quality are also guaranteed.The main tasks we have finished in this thesis are showed as follow.Firstly, inputs of Neural Network were researched. The milling is a process which is very complex, high nonlinear and close coupling. In this thesis we obtained the inputs of Neural Network based on analyzing the factors which can influence the cutting load.Secondly, the characteristic current which is the output of Neural Network was researched. To aim at the difficulties in the method of measure cutting load directly, we used the indirect method based on measuring on-line spindle currents. In this thesis, by analyzing the milling process and theory on motors, the corresponding relationship between the spindle motor currents and cutting load has been confirmed. Then the characteristic current matched along with cutting load was extracted availably by processing the spindle currents. And the characteristic current was used as the output of Neural Network. The time lag of the characteristic current was also solved in this thesis.Finally, we designed the framework of self-adapting control system, include both hardware and software. The functional modules were also developed. In order to set up the mapping model between the cutting parameters and characteristic currents about cutting load, two different Neural Networks were established in the thesis. And the performance of the two Neural Networks was testified by using the experimental data. The machining experiment showed the system works well.