'Expert' computers 'see' color the way you do.

Artificial intelligence adds speed, automation, and ease of use of color formulation and QC tasks. It also adjusts the computer's sense of color to your eye.

Getting the color of your plastic product right the first time is often your first and most crucial test of quality, whether you are a color-concentrate manufacturer, resin producer, compounder, molder, or extruder. Certainly, color control is paramount to your customer's overall satisfaction. We all know how difficult it has become to ensure adherence to color specifications across multiple approval points and within an increasingly complex cycle of tighter manufacturing tolerances, global competition, and just-in-time scheduling.

Increasingly sophisticated and/or flexible instrumentation - including more affordable benchtop and handheld spectrophotometers and colorimeters - have addressed some of these problems by providing a consistent means of measuring and communicating color as it moves through the production cycle.

But it is the latest technological advances in software - some only a year on the market - that are making the most impact. Today's artificial-intelligence software can address a deeper level of color control than ever before. It can help manage recycle inventory, reshade off-color batches, and reduce concentrate and pigment costs. On the production line, this technology allows operators to measure batches against standards, adjust for film thicknesses or substrates, and easily set color-acceptance tolerances according to the ways in which we really see color.

COLOR BY THE NUMBERS

Since the first color-formulation algorithm was reported more than 50 years ago, a number of mathematical theories have been used to create color-matching and control functions that produce repeatable standards. The bases of today's color science include the classical equation of Kubelka and Munk as well as later theories that sought to perfect color matching through models based on tristimulus (three coordinates) or spectral (wavelength) data. Common to all is that they define a three-dimensional "color space" according to linear mathematical equations. The result is a somewhat arbitrary view of color that fits neatly into an equation but does not always track with how the human eye actually perceives color.

Computerized color-formulation systems based on these equations are subject to the same mathematically imposed limitations. That is, they give us means to define and communicate relative color specifications, but these methods do not mirror the human visual system. For example, software that utilizes colorimetric formulations treats green and red color space as equal, when in fact most humans can detect far more nuances in green than in red, so green actually occupies more "space" in our visual system.

In the plastics industry, there are many other factors that come into play when matching samples visually. Is the sample fiat or does it have curves? Is it glossy or matte, smooth or textured? Under what conditions has the sample been selected and viewed to match its color? Such variables affect our perception of color and therefore influence its measurement and the ease with which we can achieve visual agreement with that measurement.

Different gloss levels between batch and standard can easily translate into costly rejects, particularly in dark colors. So can the opacity of the colorant and the opacity and inherent color of the matrix in which it is incorporated. If you have ever been given a fabric or paint sample and asked to reproduce the color in a high-gloss plastic chip, you are well aware of the painstaking trial-and-error testing that often ensues in trying to get the match against all of those variables.

Consider also how differing light sources affect color perception (a phenomenon known as metamerism), and the potential for confusion increases. A sample will look quite different under daylight than it will under fluorescent lighting of the plant floor. In fact, the discrepancy between tristimulus color space and how we view it has grown as the environments in which we measure and attempt to achieve color control have broadened. Computer technology has allowed color formulation to migrate from the lab and out into many areas of operation, from the inspection bench to the production floor.

'INTELLIGENT' SOFTWARE

In the production laboratory, master shaders can correct for many of those real-world variables based on their vast experience. When they receive a new target color to match, they compare that target to a personal file of previous matches. based on their visual assessment of the color difference, they select a known recipe that is the closest match to the new target. They then adjust the recipe until they obtain an acceptable match. It is this deeper level of color control - adjusting a theoretical match based on practical experience - that many computerized color-matching systems nowadays can simulate.

Here are some of the specific advantages you can expect from a color-matching system that has true "expert technology" - i.e., the ability to take theoretical predictions and adjust them based on the experience of the system's knowledge base:

* Precise color matching - employing advanced mathematics to predict colorant behavior, including automatic compensation for differences in gloss between two samples.

* Automatic search-and-correct functions - using information stored in your color library (i.e., proven colorant combinations) as the starting point for the new formula.

* Risk-free formula modifications - allowing users to try new combinations or formulas by modifying individual components on screen without risking production or product trials.

* Transparent and translucent matching - reducing colorant costs by automatically accounting for the interaction between pigment opacity and plastic opacity in order to minimize the amount of pigment used to get just the desired overall product opacity. Some systems can also match both opaque and transparent targets measured by either reflectance or transmission.

* Material management and workoff - automatically storing recycle and regrind as formulas and colorants and the ability to characterize regrinds by a single measure. Ideally, an expert system should also provide a method of recycle inventory tracking with automatic updates. That ensures that new formulas using recycle optimize the use of materials.

In the area of quality control, virtually nothing elicits more controversy than how to measure, quantify, and communicate color differences objectively - a consequence of all the real-world variables previously mentioned. In the past, the best solutions were based on complex mathematical distortions of color space in the area near to the standard. These distortions required manual adjustment of several empirical parameters.

Now, through the latest in artificial intelligence, you can develop pass/fail tolerances from visual assessments of your historical batch data. Some of the most advanced color quality-control software uses non-symmetrical ellipsoids to define the "acceptability area" of your products in color space. This more closely approximates the non-spherical color space of human vision - the way the human eye sees color - rather than the arbitrary, mathematically defined color space of pure theory. Software based on real-world models of color space can make pass/fail judgements founded on those made by your customers or your own staff.

Such systems further accommodate differences in color perception through an easily implemented, operator-specific procedure. Each sample is visually examined by a human observer, who makes a pass or fail judgement. The data are then run through a "training" program that determines the acceptability tolerance that will include all the "pass" samples and exclude all the "fail" samples. The result is automatically stored as a defined "tolerance area" for that standard.

Such a system can also allow for visual data conflicts - i.e., disagreements between different human observers about whether a sample is a "pass" or a "fail." The software outlines a zone of uncertainty so that future samples that fall into this area can be flagged for visual inspection. After a visual decision has been made, the data set should be updated automatically to reflect the new information. Intelligent software continuously "learns" the bias of your own visual observers to anticipate their pass/fail judgments

COMPATIBILITY & EASE OF USE

These advantages of "expert-based" software aren't much good if the design is awkward or difficult to use. Even the most complex technology should seamlessly integrate with your normal color-management routine so that the system continuously improves first-shot matching without any special effort on the users' part. The entire system should ideally allow the user to change screens and report data in formats tailored to fit individual needs.

Look for an expert software system that supplies you with tools such as user-defined job identifications for quick and painless job tracking. Also, you will be able to make the right color decision faster if the system delivers on one screen all the information you need to evaluate all formulas and corrections, rather than forcing you to scan several screens of information or sets of analyses.

Your expert system should not only be compatible with a range of other instruments and systems, but should also provide a wide variety of output capabilities to make optimal use of data. Make sure each formula can be examined or exported in a variety of ways, from showing just the single "best" match to multi-formula displays with cost and color-difference data to expanded formulation displays that include all fillers, resins, and other non-pigment ingredients. The system should also export formulas in a human-readable format such as ASCII. This ensures access to the data by other departments in your organization or your customer's third-party programs. Software that uses industry standards such as Microsoft Windows will also communicate more readily with inventory or other computer applications in your organization.

Danny C. Rich, Ph.D., is manager of advanced colorimetry and metrology at Datacolor International, Lawrenceville, N.J., a supplier of color-measurement hardware and software.