The Research Of Artificial Neural Network Modulating The Laser Welding Craft Of Ti₃Al

Laser welding process includes a series of complicated mutual influenced physical and metallurgical processes. In laser welding, laser ?plasma ?accumulated effects in wall mutual effects would make the search of laser welding processes very complicated undoubtedly. Thus it has important meanings that we researched physical or metallurgical phenomenon and affected mechanism in laser welding processes,thus modulate the shape of welding seams and design craft parameters of laser welding in reducing deformation of welding seams,lessenning flaws of laser welding and rising the quality of welding structure. In case the laser welding phenomenon is realized by computer simulation, we could easily determine the welding outcome and rather better welding parameter through computer systerms. Based on the research of craft experiments, The paper utilized shine upon ability on nonlinear relations and extraordinary ability on dealing with no precise or no reliable messages ,changed the past passive position on modulating the shape of welding seams which depends on experiences and intuitions transformed abstracted display of the shape of welding seams into digital presentations of the shape of welding seams,so the reliability and accuracy has been risen greatly. Simultaneously,based on the base of modulating the shape of welding seams accurately ,we exchanged the imported volume and exported volume ,built pattern ,simulated craft parameters of laser welding on computer platform to get the ideal shape of welding seams.The paper uses studid technological means,carrid on example calculation which aimed at laser welding problem of Ti - Al intermetallic chemical compound, and the calculated result accords with the practiced results.Laser welding process is very difficult to establish exact mathematical model by appling analysis method .And craft parameters of laser welding can be determined through a lot of exprements ,which not only cost much ,but also makes welding work load very large.When modulation is under way,firstly,BP neural network is set up by adopting means of building pattern frontally,which was realized by the MATLAB software , then it should be trained by training data in experiments.Afterwards,the BP neural network should be tested and verified ,then the shape of welding seams should be simulated by utilizing BP neural network which has been tested qualified.Thentake this as a foundation, we exchanges the imported volume and exported volume,designs BP neural network which should be trained and tested qualified,then modulates craft parameters of laser welding to get rather ideal the shape of welding seams.In order to improve modulated precision, the paper conducts suitable and careful choose in training data, which makes the quality of training data increase; according to the training and exported result, the paper revises the structure of BP neural network, conducts correction with the nodes of hidden layer, and at the same time, the paper could prevent excessive train of BP neural network; the paper uses improved simulated annealing artificial neural network which replaces BP neural network to modulate nonlinear relations between the shape of welding seams and craft parameters of laser welding. The practice verifies that these means raised modulated precision, improved forecasted effects in varying degrees.Through research and practical verify, the paper studies the relations between the shape of welding seams and craft parameters of laser welding, establishes technological feasible means of simulating the shape of welding seams and designing craft parameters of laser welding to obtain the rather ideal shape of welding seams. The paper provides the rather reliable foundation and technology safeguard on enhancing welding craft level, ensuring the quality of welding structure, realizing the real time supervise and control of welding quality.