Research On The Intelligent Electrical Upsetting Process And Prediction Of Main Parameters Using BP Neural Networks

The electrical upsetting process has a lower cost of equipment, higher precision of the forging billet and better qualities.The process parameters of electric upsetting include:velocities of upsetting cylinder and anvil cylinder,heating current, preheating length and time,etc. Because different materials have different mechanical, electrical and thermal characteristics, and there are effects between different parameters, the determination of those parameters becomes a comples problem. Besides those there are many assistant parameters. Good products could be manufactured only on the condition that those parameters could be matched each other. If some of them are easy to be unstable, and mismatch appears among some related parameters, which will lead the inevitable quality defects and a reduction of rate of finished product during electrical upsetting procedure.Based on the summarization of the current status and the developing trend of valve producing technique, the principle, characteristics and existing problems of electrical upsetting is discussed in this paper,and the changing disciplinarian and characters of all parameters of the technique are also analyzed after a detailed consult of all national and international related technical documents.In this paper, depending on the "Spherically upsetting method" of electrical upsetting of intelligent, the theoretical calculation method of heatting current and pressure in Spherically upsetting is utilized, calculating each process of parameters such as heatting current and pressure and making paragraphs settings for it based on the upsetting displacement. Practical result shows the new designing process is reasonable, with high reliability,which not only improves the quality of upsetting production but also improves the efficiency of production and reduces the rejected rate of upsetting production.Recently most research on electrical upsetting process focus on the qualitative analysis. Albeit these methods adopt numerical simulation and academic deducing, these research is difficult to use in the real production, because there are many factors influence the process, and the models of these methods are set up from hypothesis, which is undependable and unbelievable. Artificial Neural Networks has the merit of appropriately describing the objects, which have black box and none-linear attitude. Therefore, the procedure of certifying electric upsetting parameters is seem as the black box; the factors that influence the parameters are used as inputs; and the parameters are outputs. This model is set up,Studying samples were modeled and optimized with Levenberg—Marguardt algorithm of BP neutral network models, and enough suitable data samples are used to train ANN. The prediction results of the neutral network were compared with the determined results indicated that BP network not only does this method describe the procedure of process, but also can preview reasonable control parameters.