on-scan-it—mining useful data in the finishing shop

1

Click here to load reader

Upload: anselm

Post on 30-Dec-2016

222 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: On-scan-it—Mining useful data in the finishing shop

- Mining Useful Dat

Reviewed by Anselm Kuhn Metal Finishing Information Services Ltd., Stevenage, Hertfordshire, England, United Kingdom; E-mail, [email protected]

o behind the scenes in Wall Street where the researchers hang out, and you’ll find the term “Data Mining” is very big indeed. These financial

whiz-kids learned some time ago that there is a mass of data that can be extracted, deconstructed, and recon- structed t buried in Company Reports, Press Releases, and magazine articles. Most importantly, this mass of information can be combined and analyzed to yield infor- mation that those who originally provided it, never thought was accessible.

Wall Street doesn’t have a monopoly on numerical data. Past issues of Metal Finishing are stuffed with valuable quantitative information. Sometimes it’s in tabular form. But mostly, that data is presented as graphs. Graphs are the greatest way of visualizing data, allowing us to spot trends at a single glance. But what if we want to access the original data from which they were plotted? Suppose we want to compare the plating rate of Joe Doe’s electroless nickel with that from some other supplier? The only way to do this is to grab the original data from the various sources, and replot them onto a single graph. Or maybe Joe Doe (a simple-minded sort of guy) plotted several graphs, showing deposition rates as function of temperature, pH, agitation, and nickel ion concentration? We’re smarter and recognize this can be expressed as a multi-variate regression term which we might feed into an SPC program.

Of course it’s always been possible to grab a square and a ruler, drop perpendiculars, measure their length and thus estimate the original data points. Now there is a computer program that does it all for you. UN-SCAN- IT takes a scanned image of a graph and in simple fash- ion, extracts the original data from which it was plotted. The result is a set of x,y data points. Then what ? The simplest thing is to use this approach to marry data from a range of sources. To facilitate this, the program allows you to export this data to another graph-plotting package where we can either plot different lines for Joe Doe’s data and other people’s, or simply plot all the points to which we can fit a single line.

But the software will do much more than that. “Rogue points” can be removed and new data added. One can measure areas under selected portions of a graph, one can transform the data, simplify it, produce a first or sec- ond derivative, smooth it, and fit straight lines or poly- nomials to it. The package incorporates its own graph- plotting routine where the user can inspect and operate on the data.

The software will also do other, real crazy things. Take a digital photo of, say, a gearwheel or any essentially 2D object. UN-SCAN-IT will digitize it (no scanner needed) and then calculate the surface area. Which, if we’re set- ting total plating currents for a given current density on a complex shaped job, could save an hour or so (and give the right answer!). Miscalculate the surface area of an awkwardly-shaped component, and the job costing will be wrong. UN-SCAN-IT will help out there too.

How else might a finisher find this useful? Most obvi- ously, as noted, is to bring together data from various sources into a single graph, to compare data from differ- ent sources. A special unscanning mode allows us to ana- lyze cyclic voltammograms, such as those from Cyclic Stripping Voltammetry (CSV). Analysis of spectra and chromatograms (Ion Chromatograms or HPLC) is more accurate than just measuring peak height. Using the derivative function, we can take those graphs that plot concentration vs. time (for example electrochemical recovery of metals from solution) and recast that data to show the actual rate at which the metals are being removed. The same for plots of coating thickness vs. time. This is the information we really need - yet how rarely it is shown. The same is true with many waste- water treatment processes. We need to know when we’ve reached the point of diminishing return. The first deriv- ative provides the answer. For those of us who want to know “what’s under the hood,” mining the data from graphs, seeing whether it fits first-order or second order kinetics, modelling empirical fitting equations - stuff we need to make predictions, estimate production rates - this program allows us to do all of that. In one mode, the software can handle scatter plots, which makes one wonder how it would cope with a photo of a randomly rusted salt spray panel, or a pitted Hull Cell panel - both of these are analogs of scatter plots.

This user-friendly package sells in the $350 price range and comes nicely-presented on CD-ROM with a real manual in a ring-binder. Installation was trouble- free, the software is thoroughly professional, and cus- tomer service was helpful. More complete information is available on the Silk Scientific website. For anyone trad- ing in numbers, anyone into analyzing and comparing quantitative results and data, this is a very handy tool that actually delivers on its promise. The days of “rule- of-thumb” have passed, and today’s finishers deal in real numbers, This little package could earn its keep in the finishing shop.

February 2004 29