self-learning control strategy with application to milling system.
by：QY Precision 2019-11-03
1. Introduction to Computer Numerical Control operations (CNC) The use of large parameters such as cutting depth and feed rate will significantly increase productivity. However, with the increase of cutting depth and feed, the increase of cutting force, the thermal expansion of tool tip, tool and workpiece deflection, the vibration of the machine will increase, thus reducing the accuracy of the workpiece. Therefore, although the cutting depth is different, keeping the cutting force at the cutting edge at the appropriate value is a way to ensure that the size error allows. Due to the nonlinear and time constraints of the system, the processing process is difficult to control. The parameters are different due to the change of cutting depth. Therefore, the use of adaptive control systems to increase productivity is achieved by automatically controlling the feed rate to maintain a constant spindle load ( Liu & Wang, 1999). Due to problems such as instability and large transient overshooting when machining conditions deviate from controller design (Balic, 2000) The application of modern adaptive control algorithms has been proposed. Model Reference Adaptive Control (MRAC)scheme(Tomizuka et al. , 1983), self- Optimization control strategy ( Huang Lin, 2002) And other adaptive control methods ( Zuperl & Cus, 2003) Experimental tests have been conducted. [ Figure 1 slightly] The whole process needs to be modeled to design an adaptive controller. Due to highly nonlinear and time-dependent, it is difficult to model the milling process dynamics Different cutting properties. Therefore, a milling process control method based on adaptive learning is proposed in this paper. Adaptive learning control system adaptive learning control program on knowledgelinelearning. It consists of forward neural network and fuzzy feedback mechanism. The input of the controller is the error of the cutting force and the change of the cutting force error. Neural network prediction inverse- Using the dynamic model of the controlled process and the fuzzy feedback mechanism, the connection weight of the neural network is modified adaptive. Through these two elements, the inverse dynamics model of the controlled object can be modified adaptive according to the change of cutting conditions, thus automatically obtaining the adjustable feed speed of the constant milling force. In order to verify the effectiveness of the adaptive learning control system, experimental cutting test was carried out. 2. Architecture diagram of adaptive learning control 1 shows the structure diagram of the adaptive learning control system for milling to realize automatic opening Line adjustment of constant milling force feeding speed [F. sub. ref]. [ Figure 2: During milling, the milling force increases when the cutting depth increases. In order to avoid tool breakage, the control system will reduce the feed speed immediately. When the cutting depth is reduced by hours, the system produces a larger feed rate to maintain a higher cutting efficiency. The developed system controls the peak milling force F during the tooth cycle. The measured milling force F passes through the thread extension line (TDL) The output vector contains a filter that measures the value of the milling force delay. Then, enter the delay value of the milling force to multiple Neural network in front of layer. There are 4 neurons in the input layer of the neural network, 6 neurons in the inhibition layer, and 1 neuron in the output layer. Set the learning rate and theme parameters to 0. 01 and 0. 5 respectively. The limiter binds the command signal to avoid any damage caused by excessive speed. Found through extensive testing and simulation, using 3- 4-layer forward neural network6-1 type (Figure 2). In this study, many different neural networks have been tested and simulated ( Perception, Hebbian, bp Network). In the current work, two supervised neural networks for modeling are compared. The first is the back propagation neural network (BP) The hidden layer has a sigmoid transfer function and the output layer has a linear transfer function; The second is the radial basis Network (RBN) Has a Gaussian activation function. Fuzzy feedback mechanism (Figure 3) It consists of a blur, a knowledge base, a fuzzy reasoning engine and a fuzzy solver. For the fuzzy feedback mechanism, the input scale factor (0. 516/0. 0331)andoutput (5)were chosen. [ Figure 3 slightly]3. Experimental test and discussion in the experiment, a ball Vertical Milling (R218-16B20-030) Two cutting edges are installed on the Heller bea01 CNC machining center with the Fagor CNC controller. Cutting blade r2 18-16 03 M-Mwith 12[degrees] The front corner was selected. The cutting conditions are: milling width [R. sub. D] = 16mm, milling depth [A. sub. D] = 4mm and cutting speed left, L& Wang, C. (1999). Intelligent Adaptive Control of milling process. International Journal of Computer Integrated Manufacturing, Volume 112, 453-Tomizuka, M. 460; Oh, J. H. & Dornfeld, D. A. (1983). Model Reference Adaptive control for milling process. Seminar Notes on manufacturing processes and robotics systems, New York, 55-63 Zuperl, U. & Cus, F. (2003). Optimization of cutting conditions using neural networks. Robot. comput. integr. manuf. ,Vol. 19, 189-