Study On Artificial Intelligent Control Method For Consuming Electrode Gas Shielded Arc Welding Inverter

The arc-welding inverter power supply, using the power electronic instruments as the switch components, introduces the closed-loop feedback control system to meet the welding technological requirements. Because of its well performance, fast dynamic response and having facility in realizing real-time control, it has been recognized as the most advanced welding supply. And it will be more and more widely used. The actual welding process is very complicated, so the traditional control method is unmanageable or bad to control the result. This thesis focuses on the application of intelligent technology in arc-welding supply.The arc-welding supply is a typical control system synthesizing the linear link and nonlinear link. Taking no account of the changing load, the small-signal linear model is established. During the short transition of CO2 welding process, the two states, load and short, will be changing continually. So, to the arc-welding inverter supply, it's working in the large signal perturbation state. A large-signal nonlinear model is created to simulate the system dynamic response. The simulation results, voltage, current and current change rate , satisfy the arc welding technological requirement.According to the nonlinear model established previously, this thesis uses neural network to identify the system. To setup the network, train the network and simulation by use of the NN toolbox of Matlab 6.x. During the self learning process, the adaptive learning rate and momentum gene are introduced to accelerate the rate of convergence and advance the identify accuracy.The technology of the Expert System(ES) and NN combination is one of the most important research aspects among artificial intelligent technology. The difference and complementarity between the ES and NN are analyzed in the thesis. According to their complementarity, the knowledge acquisition and representation utilizing the NN is introduced into the ES. Then the arc-welding parameter selecting ES based on NN is established. The working process of this system is represented in this thesis as well.The NN learning block, knowledge base, inference block and the database system are designed in the thesis. The NN learning block, using BP network, is determined by comparing the simulation results. Normalizing the training data and then training the NN, the knowledge about network configuration and the weights is stored into the knowledge base. Thereby the knowledge is finished. The forward inference rule is adopted in thisthesis to deduced making use of the knowledge in the knowledge base. The database is set up using SQL Server. Under the IDE of Visual C ++ 6.0, the established database file is programmed and visited using MFC through ODBC interface. The established system can deduced the satisfied welding current and voltage with the given base metal type, thickness and the welding wire diameter.