CHARACTERIZING THE MICROSTRUCTURE OF A GTA WELD IN-PROCESS USING HIGH-SPEED, HIGH-MAGNIFICATION, DIGITAL IMAGING

A.C. Hall*, G. A. Knorovsky, C. V. Robino", J. Brooks*,

D. O. Maccallum, M. Reece", G. Poulter"

ABSTRACT

A high quality zoom lens and a high-speed CCD camera have been used to image gas tungsten arc (GTA) welds in stainless steel. Both the trailing (solidifying) edge and the leading (melting) edge of the weld pool have been observed. A number of solidification phenomena have been captured including: dendrite growth, melting, and weld ripple formation. Significant information about the evolution and structure of the solid-liquid interface can be extracted from these videos using computerized image analysis techniques. This information can be directly related to the microstructure of the finished weld. Video clips will be presented, techniques for extracting microstructural information from those clips will be discussed, and the extracted information will be related to the microstructure of the finished weld.

KEYWORDS

WELD VISUALIZATION, MACHINE VISION, WELD RIPPLE, SOLIDIFICATION, MELTING

INTRODUCTION

Over the last decade, significant advances have been made in digital photography and related computer technology. Combination of these technologies has given rise to a field known as machine vision. It is now relatively straightforward to capture and process images using a computer. Robust software exists that allows the user to extract quantitative measurements from digital images. This technology has found widespread application in the world's manufacturing industries [1-3]. Parts are routinely inspected using machine vision technology to see that they have been assembled properly or manufactured to desired tolerances. Surprisingly, this technology has not been widely adopted in the scientific community. It affords the researcher with a powerful new tool set that allows large amounts of information about dynamic physical processes to be accessed and extracted in an automated fashion. Solidification of metals is a field that is ripe for the exploitation of such technology.

Conventional techniques for inspecting the solid-liquid interface in a metal typically involve rapidly quenching a sample and then polishing, etching, and examining the quenched sample [46]. Experimental techniques based on such schemes are very useful but have a significant

Sandia National Laboratories; Albuquerque, NM *Sandia National Laboratories; Livermore, CA

limitation. They only provide the researcher with a "snapshot" of the solid-liquid interface. No direct information about the motion or the evolution of the solid-liquid interface is available from a quenching experiment. The only experimental schemes that do provide information about the motion and the evolution of the solid-liquid interface are based on visualization of transparent materials (like succinonitrile) that solidify in a manner analogous to metals [7, 8]. These experiments have provided a wealth of information about dendrite growth and the evolution of the solid-liquid interface. However, these experiments have a significant limitation. Transparent materials often do not exhibit the complex phenomena that occur during the solidification of multi-component engineering alloys. Solidification mode transformations, which are commonly observed in stainless steels [9], are an example of complex phenomena that are not exhibited by transparent materials. Complex phenomena like this demand new experimental techniques that can provide information about the dynamic behavior of solid-liquid interfaces in engineering materials.

In this paper, we describe a technique for using machine vision technology to characterize the behavior of the solid-liquid interface in a gas tungsten arc (GTA) weld. Dendrite growth, melting, and weld ripple formation have all been observed at high spatial and temporal resolution. A large amount of quantitative data can be extracted from these experiments in a relatively straightforward manner. An example showing how data is extracted and related to the microstructure of the finished weld is discussed in detail.

EXPERIMENTAL PROCEDURE

The solid-liquid interface in a GTA weld was filmed using a Kodak EktaPro camera equipped with a Navitar® Zoom 6000 lens. The Kodak EktaPro camera is capable of capturing digital images at up to 12,000 frames/second. In practice, we have found that filming at 2000 frames / second provides temporal resolution that is more than adequate for a GTA weld. The Navitar® Zoom 6000 lens is designed specifically for machine vision. It is a parafocal zoom lens system capable of imaging objects within fields of view ranging from 146.6mm to 0.03mm on each side.

The camera and welder are fixed with respect to each other and the sample is translated below them. See Figure 2. This arrangement keeps the solid-liquid interface stationary with respect to the camera.

Figure 2: Schematic showing the arrangement of the camera, sample, and welding torch

The entire apparatus (camera, motion control stage, and welding torch) is contained inside a glove box so that an inert atmosphere can be maintained around the weld. This prevents oxide from forming on the surface of the weld pool and obscuring the phenomena of interest. Welds were made at a variety of amperage, arc length, and speed combinations. Solidification and melting phenomena were filmed by pointing the camera at the trailing or leading edge of the weld pool.

RESULTS AND DISCUSSION

Figure 3 is a still image taken from a video of a 150 amp GTA weld in 304 stainless steel traveling at 4.2 mm/sec. Dendrites can be seen at the solid-liquid interface.

Figure 4 is a still image showing melting in Inconel 718. Liquation of the interdendritic laves and/or niobium carbide eutectic-like constituents can be seen at the solid-liquid interface. Figure 5 is a still image showing weld ripples in 304 stainless steel. Multiple weld ripples can be seen behind the solid-liquid interface.

Figure 4: Melting at the solid-liquid interface in Inconel 718, liquation can be seen at the interface

Data Extraction

A large amount of information about the solid-liquid interface and its motion with time is present in these video recordings. Quantitative information can be extracted from these video clips in a relatively straightforward manner using computerized image analysis schemes. We have accomplished this using National InstrumentsTM Lab View and IMAQ Vision software.

Our image analysis scheme proceeds as follows. First, a digital movie is recorded as a series of image files. Each movie typically contains between one and two thousand image files each of

which are 500KB in size. Image analysis software is used to open each image file in sequence and measure a feature of interest (a dendrite arm spacing, the location of an interface, etc...). All measurements are made in a pixel coordinate system that uses the top left corner of the image as an origin. After each measurement is made, it is associated with the frame number of the image and written to an output file. Frame number and pixel coordinates must then be converted into meaningful units like time and position. Frame number is easily translated into time by multiplication with the framing rate. A dimensional measurement like primary arm spacing requires a simple unit conversion to an appropriate length measurement. Filming an accurate scale provides this, as well as a velocity calibration. A measurement like interface velocity requires a frame of reference correction in addition to a unit conversion. A frame of reference correction is necessary because all measurements are made in the camera frame of reference. If information like interface velocity is to be translated into the sample frame of reference the stage velocity must be measured accurately and accounted for.

Example: Weld Ripple

In the video that Figure 5 was extracted from the motion of the solid-liquid interface that is associated with the formation of a weld ripple was seen. That motion was quantified by measuring the position of the solid-liquid interface in each frame of the video. The interface position is determined from a marked intensity change in the digital image. Figure 6 is a graph of this information. The upper curve shows interface position in pixels coordinates with respect to time.

Figure 6: Data from machine vision analysis of a rippling weld.

The raw data associated interface position with frame number. Converting frame number to time was simple. This video was filmed at 1000 frames / second with an exposure time of 1/1000