This article highlights three case studies illustrating the use of image analysis technology in an industrial setting where notable productivity gains have been realized.

Manufacturing products that adhere to strict industry and corporate quality standards is no easy task. With competition getting more fierce, smarter, and increasingly global - every day, organizations must deal with thinning margins and potentially severe repercussions if inefficiencies and poor productivity are not addressed.

In the quality assurance lab, the situation is no different. Manual laboratory procedures are being slowly replaced with innovative automated solutions that provide consistently accurate, objective and reproducible results. One such technology being implemented into leading labs around the world is image analysis.

A typical PC-based image analysis system consists of a CCD (charged coupled device) camera mounted on a microscope and attached to a frame grabber. With the use of sophisticated image analysis software, laboratory technicians are able to easily manipulate their images into their accurate binary components of interest, and then analyze their binary structures from an analytical, statistical point-of-view.[1]

With the advent of the computer revolution, today's image analysis systems have come a long way since their early beginnings in the 1960s. These systems used integrated light pen technology to digitize and quantify single objects of interest when pointed at a screen. Significant advancements in hardware coupled with new developments in mathematical algorithms applied to image processing have contributed to the growing acceptance of image analysis technology as an effective tool for image quantification.

Fig. 1 Image of screen sieve as captured through image analysis software. Insert: photograph of a sieve for determining particle size.

CALIBRATING ELECTROFORMED SIEVES

At Alcan International Ltd., electroformed sieves are used to measure particle size distribution of alumina powder. As a result of the powder's highly abrasive nature, openings in sieve sheets often become enlarged after numerous uses. To avoid an over-evaluation of particle sizes, sieves are periodically calibrated according to ASTM E161-87 methods. Based on these methods, the Arvida Research & Development Center's (ARDC) metallurgical laboratory (Jonquiere, Quebec) developed a semi-automatic procedure for calibrating sieves using image analysis.

With an automated image analysis system, sieves are placed under an upright Zeiss Axioplan microscope illuminated in transmission mode. The image is captured with a hardware and software solution combining a color, three-chip Sony CCD camera, a Pentium-based PC, and Clemex Vision image analysis software. Fig. 1 illustrates a typical image of a four-inch diameter sieve sheet at a magnification of 35X.

To extract quantitative measures of sieve openings using image analysis software, a sequential list of image processing and measurement instructions, known as a routine, was developed. Integrated within the routine are "waiting steps" which permit the operator, at designated intervals, to manually focus and move the stage from field-to-field. After each step, upon focusing and positioning the stage, the operator clicks on the run icon that instructs the system to calculate user-defined measurements previously established in the routine. Once all calculations for openings in each field have been calculated and compiled by the system, raw data is presented in the form of histograms and in a spreadsheet-like data browser. Reports are generated with data being exported and reused in a Microsoft Word report template.

Since using image analysis technology, the ARDC metallurgical laboratory has experienced significant productivity gains with respect to statistical accuracy and speed of analysis. Conventional methods called for operators to manually inspect sieve sheets with the use of a micrometer. Typically, an operator would move over a random selection of about 20 fields, with a minimum of 5 measurements per field noted. Understanding that this method is highly susceptible to human error and bias resulting from subjective operator interpretations or transcription errors, ASTM E161 method stipulates that several operators should measure a given sample several times in order to obtain more accurate results. According to Jacques Boutin, supervisor of the ARDC metallurgical laboratory, "Image analysis has enabled us to improve the quality of our statistics since we now only have one set of eyes making measurements." He added that the speed of the system has enabled his division to analyze an average sample in 10 minutes rather than the 20 minutes necessary for manual measurements. This does not take into account the fact that each sample would have to be measured by several operators to ensure statistical accuracy.

Fig. 2 Sampling screen from image analysis software.

INCLUSION RATING

At Nucor Cold Finish, Norfolk, NE, a division of Nucor Corp., corporate quality standards must be meticulously met and verified for major customers in the automotive and heavy truck industries. An important portion of the process involves conducting inclusion rating of steels according to internationally recognized industry standard methods.

Prior to using an automated system, inclusion rating was done manually by visually comparing what was observed with a microscope with standard industry charts. A slow, tedious process, this method has since been replaced with an automated, dual screen image analysis system with an inclusion rating software module (Fig. 2).

On a typical day, Cameron Kaufman and the quality assurance team at Nucor would follow several steps when conducting an analysis. Samples would first be sectioned and then subjected to a five-step polishing process. To remove excess dirt, the sample would be cleaned and dried with an alcohol solution. Ready for inspection, one heat (a batch of 6 samples) at a time would be analyzed using the automated system. At a magnification of 100X, the number of heats would be set using the inclusion rating software. By clicking the run button, inclusion ratings would be calculated in less than four minutes per sample (with average sample area of approximately 160mm2).

According to Kaufman, the automated image analysis system for inclusion rating has increased the productivity of Nucor's quality lab by 500% when measuring the number of samples analyzed per month and saved approximately $20,000 per year.

Fig. 3 (a) An electronic image of the original microstructure of HSLA steel. (b) The computer corrected image after filtering in the image analysis software. (c) The results of grain size measurement from the sample.

MEASURING GRAIN SIZE IN HSLA STEELS

As one of Canada's largest integrated steelmakers, Dofasco produces a wide product line including rolled steels in hot and cold rolled; galvanized and GalvalumeT prepainted; tinplate and chromium coated steels in coils, cut lengths and strip; and welded pipe and tubular steels. To evaluate the quality of its steels, Dofasco's R&D Center in Hamilton, Ontario, routinely uses grain size measurement to gage the material's mechanical properties.[2] Manual measurement of grain size can be time consuming, tedious, and subject to operator error. To facilitate the process, Tracy MacPherson, a research specialist, uses a high-resolution image analysis system, equipped with a Nikon Ephiphot microscope, Xillix MicroImager black and white CCD camera, a Pentium-based computer, and Clemex Vision image analysis software.

Measuring grain size in HSLA steel poses a difficult challenge, since grains tend to be very small, and irregular in size, shape and orientation (Fig. 3a). Using image analysis software, an image analysis routine was developed to process the image and extract statistical results according to ASTM E-112 guidelines. At a magnification of 500X, the routine followed several sequential steps; 1) boundaries were sharpened using a sharpen high filter, 2) the grain network was segmented using the grey threshold instruction, 3) a series of dilation and erosion steps were followed to amplify grain boundaries, 4) an object transfer instruction then followed to remove bainite objects with an area smaller than 1 micron square (Fig. 3b), and 5) object measure parameters were implemented according to ASTM E-112 guidelines. Results were then displayed in graphical form for easy analysis (Fig. 3c).

According to MacPherson, the system increased the efficiency of the analysis reducing a grain size measurement from 20 minutes to 5 minutes.

SUMMARY

It is inevitable that industry's continued emphasis on the quality movement will force labs to seriously evaluate their laboratory procedures and find more efficient methods of analyzing materials. New developments in computer technology and imaging software is making image analysis an indispensable ingredient in a microscopist's repertoire of analytical tools. As cost continues to decrease, the popularity of image analysis systems will multiply accordingly. The future may even dictate that every microscope sold will contain some form of image analysis technology. IH

For further information about metallurgical image analysis and equipment, contact the author at Clemex Technologies Inc., Montreal, Quebec, ph: 450.651.6573, or e-mail to ron@clemex.com.