Study On Modeling And Optimization Of The Cutting Quality In Plasma Arc Cutting Process

In recent years intelligent and precision cutting technologies are widely used and will be the trend in future cutting industry. Based on this situation, because of the reality problem that it is difficult to predict the shape of cut accurately in practical cutting process, a predicting model grounded on BP (Back Propagation) algorithm and an optimizing method based on SA (Simulated Annealing) algorithm is put forward; the corresponding software is programmed in this paper. Plasma arc cutting process is studied here, which has broad applications and bright future, and the prediction and optimization for the plasma arc cutting parameters in practical applications are achieved.This software takes VC++ as the main development tool, while calling MATLAB engine to realize the three-dimension effect of the data and the platform for human-computer interaction is designed which is simple and convenient, finally the prediction and optimization for cut-quality parameters of plasma arc cutting process are realized. This platform mainly adopts button control-one command, one action, which is uncomplicated and reliable. The software includes four parts, which are network training, forecasting application, three-dimensional display and process optimization respectively, meanwhile offering a range of help files. In all above four parts, according to the actual needs the user can select appropriate data files and also make reasonable adjustment of the network quality parameters such as network structural parameters, training errors and so on. In the software, this part, training process, is the right core of the network, because the learning ability will affect several other functions. After a long term research, this paper proposes several improvements for standard BP algorithm, which are adding momentum term for training, adapting learning rate constantly and modifying activation-function in order to improve the learning speed and learning efficiency. Three-dimensional display mainly call MATLAB engine to form diagram of curve and hook face of data, and it can predict the influential regulation among parameters to realize data-regulation visualization. Optimization process applies SA algorithm widely used to realize two purposes including the best optimization and target value optimization.The experimental results show that the software has strong learning ability and generalization ability, fast convergence and high accuracy. The iterative error and the relative error of training results basically meet the requirements of the system; the errors between the predictive results and the actual results are within the allowable range, while three-dimensional displays accurately reflect the regulation of real experimental data and through a number of iterative search, the SA algorithm acquire the relatively optimum parameters and target parameters, which are not only conform to the reality but also well converge to the optimization target, based on the training data.