second. Dividing the frame number by 1000 converted the X-axis to time in seconds. Converting the Y-axis was also straightforward. At this magnification 360 pixels is equivalent to 2 mm, so dividing the Y-axis coordinates by 180 gave the interface position in mm. This position was measured relative to the camera. To measure interface velocity in the frame of reference of the sample we had to account for the stage velocity. This was done by numerically differentiating the position versus time data and adding a vector representing the stage velocity to the differentiated curve. The addition of this vector placed the data in the frame of reference of the sample. The converted data is also shown in Figure 6. The interface behavior associated with the formation of a weld ripple can now be clearly seen. When a ripple is formed the solidliquid interface appears to go thorough a cycle of advancing and remelting. Evidence of this change in interface velocity can be seen in the microstructure of the finished weld. Figure 7 is a micrograph showing the transverse cross section of a rippled weld in 304 stainless steel. Light colored bands in the shape of the solid-liquid interface can be seen in the image. Similar features can be seen in longitudinal cross sections of rippled welds. These features are associated with the ripple formation seen in the video. As shown in Figure 7, the ripple bands extend all the way to the bottom of the weld pool. Figure 7: Cross section of a GTA weld showing evidence of rippling in the microstructure. Like the metallographic cross section shown in Figure 7, the video clips mentioned above are two-dimensional representations of three-dimensional phenomena. Microstructural evidence of cyclic interface motion deep within the weld pool suggests that data gathered at the surface of the pool can be correlated with phenomena that occur deeper in the pool. SUMMARY AND CONCLUSIONS The example given above illustrates how machine vision technology can be used to measure the solid-liquid interface velocity associated with weld ripple formation. Computerized image analysis techniques were used to process a large numbers of images. This allowed solid-liquid interface velocity to be measured with high temporal resolution. A cyclic solid-liquid interface velocity was observed and was associated with microstructural features present throughout the depth of the weld. This suggested that data collected from this type of experiment can provide information about phenomena deep within a weld pool. By applying the same technique to the other videos mentioned in this paper, primary and secondary dendrite arm spacing, and local melting velocity can also be measured as a function of time and welding conditions. Clearly machine vision has tremendous potential as a data collection technique for scientific experiments. This potential is not limited to solidification science; in fact machine vision could be used to study almost any dynamic phenomena. In the case of solidification science it has allowed us to view and quantitatively measure the dynamic behavior of the solid-liquid interface in an engineering material under actual welding conditions. ACKNOWLDEGMENT Sandia is a multiprogram laboratory operated by Sandia Corporation, a Lockheed Martin Company, for the United States Department of Energy under Contract DE-AC04-94AL85000 1. 2. 3. 4. 5. 6. 7. 8. 9. REFERENCES Zuech, N., When you're hot, you're hot: Machine vision and the semiconductor roller Babb, M., Machine Vision Moves into The Mainstream of Manufacturing. Control Hogarth, S., Machines with vision. Manufacturing Engineering, 1999. 122: p. 102-104. DuPont, J.N., et al., Solidification of Nb-bearing superalloys: Part II. Pseudoternary Huang, X.B., et al., Effect of solidification rate on the microstructure of a Ni-base Glicksman, M.E., et al., Dendritic Growth of Succinonitrile in Terrestrial and Elmer, J.W., S.M. Allen, and T.W. Eagar, Microstructural Development during |