K. Krishnamurthy, P. Kalya, University of Missouri-Rolla, Rolla, MO; R. Talwar, Advanced Manufacturing R&D, Boeing – Phantom Works, St. Louis, MO; D. L. Ballard, Air Force Research Laboratory, Wright-Patterson AFB, OH

Summary:

Friction stir welding is receiving increased attention as a joining process to manufacture large aircraft and aerospace structures because of the fine microstructure and related mechanical properties of the joints. To employ this process, it will be necessary to identify and control the process parameters, which is a difficult problem because mathematical models relating the feed rate and speeds to the quality of the weld are not available. This difficulty is exacerbated because these parameters need to be adjusted in real time during welding to accommodate variations such as gaps in butt joints and mismatch in plate thickness, which arise due to, for example, manufacturing tolerance variability.

In this study, a neural network based methodology is considered since the physics of the process is not well understood and a first principles model is extremely difficult to obtain. A neural network is trained to predict the tool force given the traverse rate, tool rotation rate, and plate temperature behind the weld tool. This neural network will then be used to design a neural controller to change the feeds and speeds in real time to maintain a desired tool force even in the presence of defects such as gaps in butt joints.

Preliminary experiments conducted show that the forge load drops in the vicinity of the gap of the butt joint as expected. Further, as the gap does not conduct heat efficiently, the forge load increases as the tool crosses the gap and comes in contact with the plate that is at a lower temperature. A neural network that predicts the forge load during welding of butt joints with gaps will be presented.

Work is in progress to quantitatively show the influence of forge load and plate temperature on the microstructure of the weld, which ultimately determines the quality of the weld. The authors gratefully acknowledge the support of the Air Force Research Laboratory through contract no. FA8650-04-C-704 (Dr. Jaimie S. Tiley, Program Manager).