Seamless montages comprising many separate im ages allow microstructural analysis at high magnification of a relatively large specimen area.

Fig 1 Zeiss image analysis system

Sandia National Laboratories is using digital image analysis to conduct a new level of quantitative microstructural characterization including quantification of numerous microstructural features over large areas. The Zeiss image analysis system (Fig. 1), part of Sandia's microscopy facility, has the capability to measure a variety of user-defined microstructural features, such as area fraction, particle size and spatial distributions, grain size and orientation of elongated particles. Measurements are made in a semiautomatic mode through the use of macro programs and a computer-controlled translation stage. A routine has been developed to create large montages of 50+ separate images. Individual frames are matched to the nearest pixel to create seamless montages. Three case studies illustrate system capabilities.

Fig 2 A portion of a 3 X 3 montage consisting of nine separate images acquired (using 50X objective lens) at a 0.209 um/pixel resolution Fig 3 Color image of copper-Kovar microstructure

Looking at the 'big picture'

Microscope stage movement and measurement functions are controlled through image-analysis system macro programs. Sandia researchers, in collaboration with researchers at Georgia Institute of Technology (Atlanta), developed a routine that allows semiautomatic acquisition and analysis of montages of several images. Montages of up to 100 separate images are easily generated. For example, Fig. 2 shows a 3 X 3 montage consisting of nine separate images, which were acquired at a resolution of 0.209 um/pixel (50X objective lens) and pasted back together. Matching between individual images is within approximately 10 pixels. The size of the montage is limited only by the size of the computer hard disk on which the images are stored.

Some applications require exact matching between individual montage frames. A separate routine was developed to allow matching of consecutively acquired images to an accuracy of 1 pixel. Operator input is used to match each frame to the previously acquired image to the nearest pixel. The result of this process is a large montage having no discontinuities at the individual frame edges.

The main advantage of a montage is that a relatively large area of the sample (on the order of square millimeters) can be examined at a high magnification. In a sample having a nonuniform microstructure, large areas must be analyzed to eliminate local variations in the size or spatial distribution of the measured particles. This often occurs in samples such as metal matrix composites where the spatial distribution of reinforcements is not uniform. The montage allows large areas to be examined at a sufficiently high resolution to accurately measure the individual particles. This technique also is necessary in samples where a gradient in particle size or density is present-such as in a fractured tensile sample where void density is measured as a function of distance from the fracture surface.

System specifications

The microscope stage can be translated either manually or automatically (via image-analysis software). Color images acquired using a digital color camera have 8 bits of color depth in each of three colors (red, blue and green), providing a total color depth of 24 bits. Color depth refers to the number of brightness levels; 8 bits of color depth corresponds to 256 distinct levels of brightness (i.e., 256 gray levels for a black and white image). Image size is 1312 X 1033 pixels. Image resolution ranges from 2.09 um/pixel for a 5X objective to 0.209 um/pixel for the 50X objective. Powerful macro routines of the image analysis software (Zeiss KS400 version 3.0) can automate most of the measurement and stage control functions. Digital images from other sources (SEM, TEM, scanned images, etc.) are easily imported into the software to allow image processing and analysis.

Image acquisition

Three major image-analysis processes are:

  • Image acquisition and storage
  • Image processing (altering image contrast and brightness, correcting for artifacts such as uneven illumination effects, and thresholding to convert the image into a binary black and white image)
  • Defining and performing the measurements of interest

    The process is illustrated using an acquired color image of a copper-Kovar alloy specimen (Fig. 3). The reddish colored regions are the copper phase, and the yellowish-white regions are the Kovar (nickel-iron alloy) phase. Color balance can be adjusted using software provided with the camera. Colors in the image can be matched to the original colors of the sample if exact color matching is required. However, for most applications, color information is not necessary, and the images are converted to gray level images.

    Fig 4 Processed image in Fig. 3: (a) Converted to gray scale image; (b) Image contrast increased; (c) Corrected for nonuniform illumination effects

    Figure 4a shows a converted gray-scale equivalent of the color image in Fig. 3. Image processing enhances the appearance of the image. Two common image enhancements are optimizing image contrast and brightness and removing effects of uneven illumination from the light source. Contrast should be adjusted to a point where different microstructural features in the image are easily distinguished from one another.

    Figure 4b shows the image from Fig. 3 having enhanced contrast. Enhancing the contrast often exacerbates a dark "halo" near the edges of the image, an effect resulting from nonuniform illumination. If this effect is not removed prior to performing the threshold operation, the lighter phase near the edge of the image may have the same gray level as the darker phase near the center of the image. This makes it difficult to separate the phases using the threshold routine. Illumination artifacts are removed by blurring the image so no details of the enhanced image remain except for the overall gray level. The blurred image represents only the illumination pattern in the image. Subtracting the illumination pattern from the enhanced image produces a uniformly illuminated image (Fig. 4c).

    Fig 5 Image of Fig. 4c that has been converted to a binary black and white image by means of thresholding. The white area accounts for 68.5% of the total area. Fig 6 Typical LENS microstructure showing distinct scallops in the layers containing melt-pool traces, which are into and out of the plane of the paper

    Area fraction of the second phase is determined by thresholding the image; i.e., converting the image to a binary black and white image where the phase to be measured is white (gray level = 256) and the rest of the image is black (gray level = 1). The binary image is shown in Fig. 5, where the phase to be measured (Kovar) is white. In this example, image analysis indicates the white area occupies 68.5% of the image area.

    Thresholding probably is the most important step in the image-analysis routine. Most gray level images do not have a sharp boundary at the interface between two different phases. Instead, there is a boundary region over a distance of several pixels where the gray level gradually transitions from one gray level to the other. Small changes in the threshold parameters alter the size of the regions of interest by changing the specific location in the boundary region where the particle is defined in the binary image.

    The relatively large variation in measured area fractions with small changes in threshold values is associated with the geometry of the particles in the image. When large numbers of small particles are present, the number of pixels at the edge of the particles represent a large fraction of the total number of pixels associated with the particles. Increasing the radius of the particles by one pixel uniformly around their edge adds a large number of pixels to the measured area, which can have a significant effect on the measured area fraction.

    When making multiple measurements of a single sample, it is important to have consistent focus levels, image contrast, and threshold levels in each frame. Otherwise, each measurement (or set of measurements) will yield different values. Also, each operator might use slightly different threshold conditions. Measurements should not be considered an absolute measure of area fractions or other microstructural features. Samples measured by different operators can yield slightly different absolute values, but the trends between multiple samples should be consistent for different operators. In most applications, measurements should have a relative accuracy of between 5 and 10%. The accuracy among different operators can be improved by performing several test measurements to determine specific threshold levels to be used by all operators.

    Case studies

    Image analysis was used in several studies including examination of the size variation of relatively large microstructural features (analogous to grain size determinations), analysis of strengthening particles in an aluminum alloy and relating microvoid features to their geometric location within a test sample. These case studies illustrate some of the measurements that can be made using the image analysis system.

    Fig 7 Outlines of LENS scallops used to determine size variation

    Variability of LENS melt features

    Laser engineered net shaping (LENST) is a solid free-form fabrication process in which a solid metallic part is built using a laser to melt powder particles. A CAD file governs the movement of the laser such that a complex part is created by successive deposition of consecutive melt paths and layers. A typical LENS microstructure is shown in Fig. 6. The part is built having alternating layers oriented 90? to one another. Distinct scallops are evident in the layers containing melt pool traces, which are into and out of the plane of the paper.

    Variations of laser power and other processing parameters during deposition can result in nonuniform melt pools, leading to voids and inhomogeneities in the microstructure, which can degrade mechanical properties. Research is focusing on developing a feedback control loop to monitor the melt pool for size and temperature. This information is used to control the power source to maintain uniform deposition conditions. Bulk LENS samples fabricated with and without the feedback control loop are characterized using image analysis.

    The sizes of approximately 200 individual scallops are analyzed in each sample, and a 4 X 5 frame montage is acquired from a controlled and an uncontrolled sample using a 5X objective lens. This corresponds to an area on the actual samples of roughly 1 cm2. The scale of the montages does not warrant pixel to pixel matching of successive frames. Thus, the frames match with an accuracy of 10 to 20 pixels. Because the contrast of the scallops is not sufficient enough to allow thresholding, a transparency is placed over the montages instead, and the scallops are manually traced. The transparencies are scanned and imported into the image analysis software to measure scallop characteristics. Figure 7 shows the scanned representation of the sample used to make measurements. Measured features for each scallop include area, maximum horizontal dimension (width), maximum vertical dimension (height) and the center of gravity. Size and location of features are easily collected and related to one another.

    Fig 8 Sample image from montage of 6061-T6 aluminum alloy. Darker particles are Mg2Si and lighter particles are AlFeSi (aligned in the rolling direction). The different gray levels of the particles allow each population of particles to be analyzed independently.

    Damage evolution determination in Al alloys

    Strengthening precipitates in 6061-T6 and 5086-O aluminum alloys cause different mechanical responses to applied loads-most notably in the strain hardening characteristics. The physical process of damage nucleation varies with material, microstructure and loading conditions. Analysis of void nucleation further explains the effects of microstructural features and mechanical behavior on damage evolution. Also, measurements provide the information necessary to develop nucleation models for these materials.

    Interrupted tensile, compression and torsion tests were conducted to strain levels of 25%, 50% and 75% of the tensile failure strain in both alloys. Measurements of damage (in the form of debonded and cracked particles) are made as a function of loading condition and strain level. After testing, specimens are sectioned and polished on a plane perpendicular to the fracture surface. Polishing artifacts including Mg2Si particle pull out (appearing as a void) are ruled out by viewing at higher magnification.

    Two types of reinforcing particles present in 6061-T6 alloy are Mg2Si and AlFeSi. The particles are easily distinguishable from one another due to their different morphologies and gray levels (Fig. 8). Measurements of interest include the number density of each particle type, the number density of cracked particles, the number density of debonded particles, the average size of virgin particles and the average size of damaged particles. The measurements are compared with baseline data from untested samples.

    To generate adequate statistical data, 25 frames were acquired and pasted together to an accuracy of within I one pixel. Original frames were acquired using a 50X objective lens to ensure adequate particle resolution. An 1.3 mm X 1.0 mm area was measured. The contrast of each frame was adjusted, and uneven illumination effects were corrected on each gray-level image prior to pasting into the montage. The montage was sufficiently homogeneous allowing the use of one threshold setting to delineate particles (13 - 108 for Mg2Si and 119 - 152 for AlFeSi) over the entire area.

    Specimen measurements are shown in Table 1. Average values are reported, but values for each particle are stored in a data file. The aspect ratio of each particle can be calculated by measuring maximum and minimum dimensions of the particles. This example demonstrates that a large number of particles can be analyzed to obtain a good statistical distribution.

    Void damage evolution in 304L stainless steel

    The fracture mechanism in AISI Type 304L stainless steel under many mechanical and environmental conditions involves the nucleation, growth, and coalescence of microvoids. Modeling microvoid fracture requires inputs derived from experimental characterization of the damage process. Microvoid damage in 304L (low-sulfur) stainless steel is quantified as a function of position in the test specimen, applied strain/ hydrostatic stress, temperature and hydrogen content.

    Different mechanical and environmental effects are examined by altering testing variables and the condition of notched tensile specimens. Hydrostatic tensile stress in the notched region is varied by altering the notch radius. The dependence of void damage on applied strain is assessed by interrupting the experiments at different applied axial displacements. The effect of hydrogen content is examined by charging the notched tensile specimens in hydrogen gas until a uniform concentration of hydrogen is established. The effect of temperature is examined by heating the specimens during tensile testing.

    Microvoid damage is produced in a controlled manner using cylindrical tensile specimens in which circumferential notches having a circular profile are machined. Microvoid damage evolves in the notched region due to the high triaxial tensile stress that develops as a result of geometrical constraint. Tensile tests are interrupted at prescribed fractions of the fracture displacement (_Lfract.). Notched regions are sectioned and polished parallel to the tensile axis to reveal the microvoid damage. Measurements of interest include area fraction, absolute area, number density, microvoid size distribution and intervoid distances.

    Measurements are presented for sharp notched specimens (notch ligament radius/ notch profile radius = 3.2) tested at 800C (1470F) without exposure to hydrogen. Microvoid area fraction is measured as a function of position across the notch ligament diameter. Five fields of view are acquired across the minimum diameter of the notch ligament using a 20X objective lens. Each frame covers an area approximately 550 um X 700 um.

    The color image acquired for each field of view is converted to a gray-scale image. Contrast enhancements are not applied to the image because the gray level distribution of the pixels is essentially binary (i.e., black voids and white matrix). Non-uniform illumination effects are corrected.

    Conversion of the gray level image to a binary black and white image required careful analysis. The automatic threshold routine selected an upper-bound threshold value of about 180, which is considerably below the peak on the histogram that corresponded to the matrix phase. Increasing the upper-bound threshold parameter results in an increase in the number of smaller voids in the binary image. A threshold level of about 220 captures the majority of the smaller voids, but also increases the measured area of many larger voids. There is a trade off between including smaller voids and increasing the area of larger voids. To make the measurements consistent, a single upper threshold setting of 220 is used for all specimens. Many fields of view near the edges of the samples and in those specimens having smaller strains consist mainly of smaller voids, which would not be captured using a lower threshold setting.

    Fig 9 Plot showing microvoid area as a function of location for three different strain levels. Also shown are photomicrographs illustrating the damage present at the highest and lowest strain levels.

    Figure 9 shows a representative sample of measurements of microvoid area as a function of distance across the minimum notch ligament diameter for three different strain levels. Two gray-level micrographs also are included, corresponding to the mid-diameter fields of view at the lowest and highest strain levels. The results are consistent with theoretical expectations-microvoid damage increases as a function of increasing strain and is highest near the center of the specimen where the highest hydrostatic tensile stress is located.

    This article is derived from Sandia Report SAND2000-8237. For more information: contact Scott Vaupen, Sandia Business Development Support, Sandia National Laboratories, P.O. Box 969 MS9951, Livermore, CA 94551; tel: 925-294-2322; fax: 925-294-3020; e-mail: sbvaupe@sandia.gov