Study On Controlling Cutting Quality Of WEDM

Wired Electrical Discharge Machining (WEDM) is a non-contact cutting process whichhas the advantages of high precision, no cutting force, high flexibility etc. It is widely used tocut difficult-to-cut materials and also in mould industry. The cutting model of WEDM is oneof the topics on the study for wired electrical discharge machining technology. Therefore ithas a great significance on achieving the high performance, high precision, high automationmachining with WEDM.WEDM is a complex cutting process with multiple parameters and uncertainty.Therefore it is hard to establish accurate mathematical models to reflect relationship betweenthe cutting parameters and the technological targets. Because artificial neural network (ANN)has very strong self-learning, adaptive and nonlinear mapping capability, so it is a kind ofeffective nonlinear modeling method. A cutting quality prediction model was proposed basedon artificial neural network theory and a large number of experimental data.According to the specialty and control demand of WEDM and the problem exsisted inHigh-speed Wired Electrical Discharge Machining (WEDM-HS), a Medium-speed WiredElectrical Discharge Machining (WEDM-MS) system in which the IPC and MPC2810areused for the control core based on Windows XP was studied and developed using modulestructures. It can realize communication between IPC and microcontroller. A friendlyhuman-computer interaction interface was developed using VB6.0. The CNC system havegoing back when shortage of circuit. Auto cutting and multi-cutting function can also berealized automatically by setting proper parameters. The system uses laser scale to from theclosed loop control. The cutting quality and precision can be greatly improved.Cutting Cr12using WEDM was experimentally studied by applying the orthogonalexperimental method. The mapping relationship among pulse width, pulse interval, peakcurrent, workpiece thickness, voltage, wire speed as well as technological targets (cuttingspeed and surface roughness) were established through a lot of testing, analysis andprocessing. And25groups of representative test data were obtained.Based on the experiment data and artificial neural network theory, a BP neural networkcutting quality prediction model were trained and validated in the MATLAB Neural NetworkToolbox. And verifications indicate that the relative error between the predictive cutting speedand actual one is from5.1%to9.8%, the relative error between the predictive surfaceroughness and actual one is from4.5%to6.8%. The trained BP model has reasonableprediction accuracy and generalization ability. It reflects nonlinear mapping relationship between the cutting parameters and the technological targets. This model has certainapplication value for the cutting parameters selected reasonably during the cutting process.