Application of the Neural Network to CNC Controller Parameters Optimization

• Main Authors: Li-Hsien Peng, 彭立賢
• Other Authors: Jenq-Shyong Chen
• Format: Others
• Language: zh-TW
• Published: 2019
• Online Access: http://ndltd.ncl.edu.tw/cgi-bin/gs32/gsweb.cgi/login?o=dnclcdr&s=id=%22107NCHU5311062%22.&searchmode=basic

碩士 === 國立中興大學 === 機械工程學系所 === 107 === CNC machine tools play an important role in the mechanical industry. When the CNC machine tool is used for machining, there are three kinds of the processing indexes such as speed, accuracy and surface quality. Due to the Industry 4.0. the products are gradually oriented towards the trend of low volumes and high product variety. After the machine tool has been shipped from the factory, it will be set a set of standard controller parameter. But this set of controller parameter can’t conform all kinds of processing requirement. Therefore, it is extremely important to adjust the controller parameters for different requirements of workpiece quality. This study conducts the time-consuming full factor experiment from controller software, and uses this data to train the pre-trained model. Through the concept of transfer learning, the pre-trained model’s parameters are transferred to the model which was trained by the machine tool to conducts the Taguchi orthogonal table experiment. Finally, the genetic algorithm is used to find the best combination of parameters for different processing requirements. Use the position loop feedback signal from machine to verify the optimization. The optimized result of the genetic algorithm is compared with the original parameters and the best parameters obtained from the Taguchi experiment. The best parameters are obtained from the Taguchi experiment. The speed index of the speed priority criterion is optimized by 95%. The corner’s accuracy index of the precision priority criterion is optimized by 63.81%. The corner’s surface quality index of the surface quality priority criterion is optimized by 87.02%. The best parameters are obtained through genetic algorithm optimization. The speed index of the speed priority criterion is optimized by 95%. The corner’s accuracy index of the precision priority criterion is optimized by 55.18%. The corner’s surface quality index of the surface quality priority criterion is optimized by 83.93%. Another part of research is to develop a rapid measurement system for machine tools. This measurement system can dynamically measure tool center point of the machine tool. Repeated experiments are carried out for different processing paths and feed rates. The experimental results show that accuracy of the measurement system is about 0.04 mm. Three processing indexes which are optimized by intelligent method will be verified by this measurement system. It is more similar to the actual machining condition by measuring the path of tool point center. A highly accurate measurement system with intelligent parameter optimization will greatly increase the efficiency of the machine''s tuning.