"Brawn + Brains > 1," Feature Article, Sept. 2002

FEATURE FOCUS: Intelligent Tools
brawn + brains > 1

Having acquired muscle and grace, machine tools begin wising up.

By Paul Sharke, Associate Editor

Unlike the search for intelligent life beyond Earth, the quest to discover machine shop intelligence here at home actually turns up a few examples. In lathes, in threaded holes even—signs of intelligent tools are out there. But what's setting these tools apart from the ones that aren't so smart?

Someplace on the list of intelligence must be shape memory alloys. At one time the domain of rocket science, shape memory alloys are finding their way into more earthbound applications.

This month, for instance, Command Tooling Systems of Ramsey, Minn., plans to unveil its ColdSet tool holder. According to engineering director Bill Keefe, the tool holder relies on the phase transformation of metal to create a clamping force around a bit.

Seeking to reduce errors during high-speed contouring, a researcher at NIST Manufacturing Engineeering Laboratory tests an intelligent controller algorithm using a grid controller.

Alloys of nickel and titanium, named Nitinol after their primary constituents and for the Naval Ordnance Laboratory where they were discovered, can assume two shapes, depending on their crystal structure. Exhibiting martensitic structures at low temperatures and austenitic structures at higher ones, the alloys, in effect, remember two distinct patterns. A popular example is the haphazardly twisted wire that spells out "Nitinol" when it's heated into its austenite phase.

According to Keefe, an external chiller immerses the ColdSet holder in an atmosphere of carbon dioxide cold enough to precipitate a layer of dry ice on its surface. Lowering the tool holder temperature to -100°F takes about a half-minute. As the material cools into its martensitic range, the tool bore expands. Once the holder begins warming, it contracts around the tool.

The ColdSet holder grips tools in the same manner that heat shrink, or thermal lock, systems do. Both set-ups use friction from a force fit to clamp down on the circumference of a tool. But a ColdSet holder moves more as it expands, Keefe said. Where a thermal lock system might provide 0.001 inch of interference, the higher actuation range of the cold system can double that, he explained. The result? A more consistent grip on the tool shank.

The payoff of the cold system comes in holding tools of the smallest diameters—down to 1/8 inch—where the amount of interference the holder develops becomes especially critical.

"The cutting tool shank tolerance has less of an effect on gripping force because the percentage of interference lost is much lower in the shape memory alloy," Keefe said.

Data Grab

Another example of intelligence in the factory is a device that can go forth and capture data about the machining environment. Florence, Ky.-based Balluff Inc. sells what it calls the DataBolt, which does just that. Threaded into an engine block or transmission prior to machining or assembly, the DataBolt records which machines do what. According to product specialist Steve Combs, the DataBolt grew out of the company's development of tooling data carriers for machine tools.

More popular in Europe than in the United States, data carriers lessen the need for manual data entry after tool presetting, Combs said. In a typical application, the presetter measures the height and diameter of the tool once it's been installed in a tool adapter. That information is written to the data carrier memory, which holds up to 2 kilobytes of information. The data carrier follows the tool setup out to the machining center.

Mounted to an air cylinder, a sensor head moves toward a DataBolt. Once it's close enough to read or write inductively, the sensor relays cylinder head machining data to the data carrier.

Once the tool and adapter are parked in a pocket on the machining center, the part program can call up the specifics of that particular tool setup. The machining center then knows the tooling offsets without the machine operator having to key them in manually, Combs explained.

The DataBolt takes this idea a step further. Ordinarily, a complex series of machining or assembly operations might justify creating a record of what holes were drilled or how many seals were installed. But installing a permanent chip in an engine block to do that gets expensive, Combs said, especially if it's an EEPROM chip of the kind Balluff uses.

The DataBolt is threaded into a tapped hole early in the manufacturing cycle. It stays with the part through the machining or assembly steps, recording process details as it goes. At the end of the line, an assembler removes the DataBolt and retrieves its contents. Returned to the head of the line, the bolt repeats a cycle of data acquisition. According to Combs, the EEPROM can write a million times to the same bit. A RAM version can be rewritten to indefinitely, he said.

Smarter Still

These two tools could be considered smart in the way that a border collie is more intelligent than the average pound pooch. But no dog ever worked through a problem in tensor calculus. What will it take to make machine tools that really think?

Hans Soons, a program manager at the National Institute of Standards and Technology in Gaithersburg, Md., spelled out his own definition of smart machine tools with five major points. Like a good machinist, smart machine tools know their capabilities and condition, he said. They know, or can figure out, the optimal approach to machining a given part. They can monitor and diagnose themselves. They know the quality of the work they make. And, they learn from their mistakes.

Of course, that's a definition only, as Soons is the first to point out. No such machine tools exist. For Soons and his associates at the Manufacturing Engineering Laboratory, the list represents the wish of people close to the industry who, one day, would like to see autonomous machine tools that produce a first part and every part that follows correctly and without breakdown. Here's his definition, again, in five easy pieces:

A forest of tools on a machining center sits ready for work. Data carriers on each adapter can relay setup information to the machining head as it calls a tool into service.

One: A machine's accuracy degrades over time. Maintenance logs, diagnostic data, and other records of a machine's condition inhabit many storage sites apart from the machine itself. A smart machine would keep its own records, communicate this data through standard protocols, and, from that, improve its operation.

For example, a smart machine furnishing this kind of data to a manufacturing or design engineer could help him decide if it could make a certain part. Better still, the machine could judge this for itself.

Together with representatives from industry, NIST researchers are standardizing machine tool data formats. They are working on ways to predict the tolerance of machined parts from generic data on machine performance.

Two: Today, a machine tool runs off a series of elementary commands that have been generated off-line by a part program. It runs in "dumb" mode, Soons said. The smart machine tool operates instead from a higher-level language.

According to Fred Proctor, who manages a program at NIST on intelligent open-architecture control for manufacturing systems, a machine tool that could work directly from part geometry data would cut out the current step of translating everything into so-called G codes. These codes only specify moves, as in "G1" for moving in a line, or "G2" for moving in a circle. Much high-level information generated in the CAD and CAM phases of development is simply thrown away in this translation to G code, Proctor explained.

The first part to undergo machining by way of STEP-NC. Design and manufacturing details, ordinarily lost in translation, can be passed along to the machine.

Along the trail from art to part, a design engineer generates in CAD the high-level mathematical surfaces that stress analysis and other CAE software then evaluate. From there, the files move to another "domain expert," the manufacturing engineer, who also employs sophisticated knowledge to decide the best approach to making the part. The domain expert determines what tools to use, what speeds are best for removing material, and so on.

In generating G code, all the design intent of the previous intellectual activity diminishes to a series of simple "go here" commands, Proctor said. Yet, the controllers that run today's CNC machines have capacity well beyond that to realize intelligent adaptive control.

"Unbalanced" is what Proctor called this primitive approach to telling an advanced controller how to run a machine. The computer that runs a CNC machine could do a lot, he said, but it has no information to go on.

That's why a draft variation of STEP, the standard for the exchange of product model data, is now undergoing validation. STEP-NC, as the new data input standard is called, will add manufacturing data to the design data. Armed with such a complete set of part data, a CNC machine outfitted with STEP-NC would have the geometry of the workpiece, its allowable tolerance, and materials, as well as any requirements for the tooling and fixturing needed to make it.

'First Part Correct'

From this foundation begins the first course of brick for building a machine controller that can plan its own work. Such a machine could figure out the right sequence of cuts, the correct motion of the axes, the proper feed and speeds, and so forth, so that the very first part it makes of a particular shape comes out exactly correct. Today, the process of making a good part often takes several tries and the input of experts. "First part correct" is a major anticipated benefit of machine tool intelligence.

Three: an intelligent machine would monitor, diagnose, and optimize itself. That would mean detecting thermal growth and compensating for the machining errors it produces. Or, detecting chatter and adjusting the speed and feed rate to eliminate it. Or, gauging tool wear and calling up a tool change.

Because just-in-time manufacturing builds no padding into production runs, monitoring machine health is another critical goal. Condition-based maintenance will figure prominently into the works of future machine tools. To assure that all that monitoring takes place, manufacturers will come to rely on smart sensors.

According to Kang Lee, the group leader of sensor development at the Manufacturing Engineering Lab, the need for sensors capable of self-identification, digital communication, high-speed networking, and distributed software control prodded both NIST and industry to develop IEEE standard 1451. The standard covers transducer-to-microprocessor interfacing, plug-and-play transducers capable of self-identification, microprocessor-to-network interfacing, and transducer application portability. As a result of the standard, transducer makers have just one interface to maintain.

STEP-NC defines a data input standard for machining centers. An extension of STEP, the standard for CAD systems, STEP-NC overlays manufacturing information on top of design data.

Work began only months ago on the latest version of the standard, which examines wireless sensors, Lee said. Wireless sensing will probably be limited to condition monitoring in the near term, he said. Real-time control without physical connections would require more work and remain an undertaking for the future, he expected. Critical control could ill afford the noise and operating vagaries that plague cell phone and other wireless users today.

Yet, wireless sensors could eliminate costly cable runs that sometimes exceed the price of the sensors themselves. Pulling new cables through existing plants can be a monumental task. Wireless sensors would fit many applications where installation is now too costly.

One problem for sensor users is maintaining relevant calibration data as the sensors remain installed for 15 years or more and begin approaching obsolescence. The standard calls for an electronic tag, known as TEDS, for "transducer electronic data sheet," as a way past this problem. The tag remains an element of the sensor memory. Replacement sensors having this feature could simply be plugged into the cable and would be able to identify and update calibration data automatically.

Assembly operations like this tube line also benefit from machine intelligence. Data carriers on each hanger can gather process information for archiving.

The "holy grail" of condition-based maintenance, the instrumented spindle that knows just how much life it has left before needing service, sits a ways off, Soons said. Detecting the onset of failure may be a more realistic near-term goal, he added.

Four: Smart machine tools would be able to acquire data during, and after, machining, then use the information in estimating the quality of a finished part. Ideally, on-machine inspection would look at complex parts in addition to the simple shapes that are checked in this way today. The challenge lies in unhitching the accuracy of a finished part from errors of the machine tool.

Five: Learning may be the truest sign of intelligence. NIST researchers are looking at ways to use part inspection data to update a machine's internal error model. Another example of learning: a machine tool that knows where to find information on cutting an exotic material for the first time.

Call to the Workshop

Some elements of intelligent machine tools are close at hand. STEP-NC, for instance, is reaching commercial availability through the efforts of Troy, N.Y.-based Step Tools Inc. Smart sensors are well on their way to commercialization.
Some aspects of intelligent machines, like detecting and adjusting the offsets in part location and tool dimensions, are almost routine. Yet, other aspects of smart machine tools are a long way off. Some areas await basic research.

Making the goal of intelligent tools no less challenging is the difficulty of getting machine tool makers, sensor manufacturers, and other suppliers to the machinists' trade to cooperate and share ideas in a normally competitive atmosphere. That might be the NIST Manufacturing Engineering Lab's biggest job.

Along those lines, NIST, the National Science Foundation, and the Integrated Manufacturing Technology Initiative will host a workshop on smart machine tools in December. They will invite a representative group of industry, academic, and government practitioners to assess needs and opportunities for increasing the intelligence of machine tools.

The purpose? To build a collaborative route over which manufacturing engineers, designers, and machine tools will share intelligence. Smart machine tools will soon incorporate the knowledge of domain experts into their processes. Designers and manufacturers, in turn, will gain a tool-level understanding of how the machining environment and machine condition affect those processes.