Prediction of heat-affected zone using Grey theory

摘要

The amount of deformation and the residual stress, which influence welded plates, are related directly to three aspects of the welding process: the method of heat input; the welding speed and the geometrical conditions; limitations of the weldment. The temperature behavior in the heat-affected zone usually determines the strength of the welded plate as a whole. It is important to carry out research into the temperature behavior in this zone, not only because the temperature in this area influences the material composition (the continuous cooling transformation (CCT) diagram) but also because it can be used to supply a feedback signal to the welding system for monitoring and control of the welding process. This study uses Grey theory to predict the temperature message of the heat-affected zone. The first stage of the project is to use differential equations to construct a two-dimensional temperature model of the dynamic conditions during the welding process, and then to model the temperature distribution of the dynamic welding temperature field in order to obtain the range of the heat-affected zone. In addition, the modeled message of the heat-affected zone around the welded bead, and temperature data collected at a dynamic state measurement point, which moves at the same uniform speed as the heat source, are used as inputs to a Grey theory predictor. Using a G(1, 1) modeling method and a G(1, 1) unified-dimensional new message method the temperature behavior in the heat-affected zone is then predicted. The number of temperature data points collected and input to the Grey predictor models varies between 5 and 15. The weld heat input source is rated at 2800-5400 W. The welding speed is 0.6-1.2 cm/s. From the results, it is learnt that the higher the number of temperature data points input to the predictor models, the more precise is the prediction obtained. Additionally, it is found that the G(1, 1) unified-dimensional new message method is more accurate than the G(1, 1) method of prediction, and that the average error is approximately 0.2-0.3 percent.