| 《The Car Industrial Industry Policy》promulgated by State Council in March, 1994, definitely requests car industry to become pillar industry of national civil economy in 2010, and arouses related industry to quickly develop. During the period of the automotive design and manufacture, it is the main factor of auto product development speed and quality that the design and manufacture level of the auto body die especially auto panel die. Whether the drawbead's design is right or not decides the success of stamping. But, because of practical problem's complexity, drawbead structure design and it's position depend on manufacture experiences and experiments, actually still in"trial and try", which reduced efficiency of production and using effect, and increased production costs. In recent 20 years, due to the fast development of computer technology, CAD/CAE/CAM has been widely used in the designing and making of auto body panels in order to effectively increase the competitive ability with high quality, in a short time and at a low cost. However, till today, it is still very difficult to carry out the 3D simulation of drawbead behavior because of the small size and the complex shape of the drawbead. In this case, a very fine small mesh is required on the drawbead in order to account the contact between the sheet and the drawbead. The compute time is lengthened and more memory is required, and it is also disadvantage to modify die's geometry. Therefore, a better equivalent method is urgently needed in order to describe drawbead's character, among them, the equivalent drawbead is a mature method.Artificial Neural Network which is short for ANN, borrowing ideas from structure and character of human's brain, is a large-scale parallel distributed information processing and nonlinear system which is composed by large numbers of processing units. It simulates human's brain in information processing, memory and searches function. There are some remarkable characteristics in neural network such as: nonlinearly mapping ability; don't need accurate mathematics model; be good at learning useful knowledge from the input and output data; easily realize parallel calculation etc.. BP neural network is a current network model which is the most mature and widely be used in practical engineering. In this paper author makes full use of its function approximation, nonlinear mapping, robust and fault-tolerant ability to calculate drawbead force and anti-calculate drawbead's geometrical parameters.In this paper, the author do some researches that calculating drawbead force and anti-calculating drawbead geometrical parameters, and inlaying arithmetic into KMAS's pretreatment which is a stamping CAE analysis software. The main contents of this paper as follow: Chapter 1 introduces the importance of calculating drawbead force and anti-calculating drawbead geometrical parameters in stamping, general research situation of calculating drawbead force and artificial neural network, points out the main contents of this thesis. Chapter 2 represents the mechanics of drawbead force, starting with the mechanics of incurving and recurving when panel passes drawbead, discusses drawbead force's influence by drawbead friction restraining force, resisting force by strain hardening, drawbead's pattern, drawbead's geometrical parameters, material characteristics, lubricate and deformation velocity etc., in this way, makes certain BP neural network's input and output. Then chapter 3 introduces ANN's basic knowledge, work principle, network model's determination and algorithm category in detail, thereby, lays the foundation for the following application. Chapter 4 gets network's training sample and testing sample by PAM-STAMP's drawbead module, dwells on the design of network's structure, the selection of transmission function, initialization of weight value, modification method of weight value, training method and its parameters, and sets up a BP neural network model aiming at this problem by using Microsoft Visual C++, realizing calculating drawbead force by BP neural network, also tests the neural network which finished its training. Chapter 5 realizes anti-calculating drawbead geometrical parameters by using some principles which are represented in chapter 4, also tests the neural network which finished its training. Chapter 6 gives a sum-up and expectation, from the result of testing we can conclude that calculating drawbead force and anti-calculating drawbead geometrical parameters by using BP neural network are feasible and effective. Some insufficiencies also pointed out: BP neural network's convergence velocity is too slow, sometimes it is hard to bear; the best unit quantities of hidden layer are hard to determine; while joining new sample, the network has to retrain.For this problem, BP neural network can deal with complex nonlinear relationship between input and output, basing on the data obtained, predict result exactly after training. Actually, if increasing more samples, at the same time, the arithmetic of calculating drawbead force and anti-calculating drawbead geometrical parameters is more improved, we can firmly believe that this method will develop greater application in more fields.
|
PDF Full Text Request