Building 3D Machining Knowledge Into CAM

A long time in coming, "generative" or "knowledge-based" machining is finally becoming a matter of practicality for more shops. If those terms conjure a fuzzy image, we are speaking here of NC programming systems that allow users to capture their preferred machining process strategies in software and then automatically execute those strategies when presented with an appropriately structured set of CAD data.

That knowledge-based machining took so long to arrive has had much to do with the complexities of dealing with the underlying software that made it possible to begin with. Engineers well versed in the arcane languages of expert systems technology were necessary to capture a shop's methods in elaborate sets of rules. There was so much work to be done on the front end just to capture enough knowledge to be of use that only companies with both formidable resources and resolve came to deploy this most promising technology in a meaningful way.

Newer CAM technology tends to take a more pragmatic, if less purist, approach to the automation of NC process planning and tool path generation. The better systems today pack quite a bit of machining "smarts" in at the "factory," but also provide the opportunity for the user to interact with the software in a conventional, interactive way. CAM users can still make every little decision if they want when preparing a program, but it makes sense to go with the default settings whenever possible because they are getting better all the time. For the more adventurous users, some systems provide the opportunity to more fully automate and customize CAM functionality through the use of expert system-like shells, macro languages or other means.

But that raises an old question about knowledge-based machining technology: In which applications can shops justify investing time into automating the part programming process, and in which should they leave well enough alone?

The answer depends on the shop, of course, but it may not at all be what you'd expect. Take Hunter Innovations, for example. This ten-person company in Sacramento, California, specializes in custom product development and manufacturing. In terms of machining, they do the really hard, low volume work that nobody else wants to do, which includes the machining of custom titanium orthopedic implants. Small company. Geometrically complex workpieces. Quantity-of-one manufacturing. It's hardly where you'd expect to find much automation of any kind. Yet automate they have. As far as Hunter's system is concerned, all parts within a family are virtually the same. They input data from a CT scan and, shortly thereafter, output a series of programs that account for the entire process for machining hip replacement implants.

And what once was complex is now quite simple. The programming is done in half the time. Equally important, the proven process generates a good part the first time, every time, for a customer that can't afford mistakes. Here's how the shrewd deployment of knowledge-based machining technology makes it possible.

Making The Difficult Routine

Hunter Innovations is not your typical parts supplier. Founded just 18 months ago by Doug Powell, who is the technical leader, and Joe Hansen, who directs commercial affairs, the original mission was to concentrate on product development for a select group of customers. And while the company has met with considerable success in this regard, they have really excelled in process development for manufacturing families of complex parts. "We're an idea group," says Mr. Hansen, "that can start with a concept and work it all the way through the CNCs."

One of Hunter's key assets is its youth. At 29 years of age, Mr. Powell is the senior member of the entire staff. While that may sound like inexperience to some, he points out that everyone in the company has grown up with computers, and so they collectively are extremely comfortable with the computer-aided design and manufacturing tools that have become central to the discipline of rapid product and process development. Moreover, no one on the team is heavily rooted in conventional wisdom as to how things are "supposed to be done," which has helped the company take a fresh approach to business and technical problem solving.

Hunter's early work was focused primarily on developing valve prototypes for the waterworks industry, where Mr. Powell had worked before the startup of the company. Then came an opportunity to work with Hayes Medical (Sacramento), a designer of custom orthopedic implants used for joint replacement. Hayes needed help in the machining of the one-of-a-kind hip implants--help that was beyond the capabilities of their other suppliers.

Custom implants are a relatively small niche in the giant orthopedics industry. The majority of hip, knee and shoulder replacements are accomplished with standard implants. Custom implant components become necessary when the patient's anatomy is deformed or otherwise unusual enough that a standard size product won't do. In these cases the bone structure must be "measured" through the use of CAT or MRI scans. Then an implant must be designed to fit the specific requirements, and then machined to specification.

Hayes, itself an innovative producer, had already developed software that automates much of the process of building the custom implant geometry. Hunter would take it from there, building a system spanning from their CAD/CAM department to the shop floor.

The Process

The implant design process begins in the nuclear medicine department of a hospital where a CT (computerized tomography) scan is conducted of the joint to be replaced. The scan provides a series of cross sections that together create a three dimensional "picture" of the entire bone structure. The 3D image is actually a digital file comprised of thousands of X,Y and Z coordinate points--not unlike what would be created by digitizing a physical model--but also with a density value attached to each point.

Hayes Medical created their own "AIDA" knowledge-based software that is used to capture and analyze this scan data. The specific workings of the software are a closely guarded secret, but the upshot of it is that a new file is created that defines the specific geometry of the implant component to be made for the patient. The new file is essentially a series of cross sections, each layer of which is referred to as a "spline," and this is the base data that Hunter works with in their own system.

Hunter has built their knowledge base within an Esprit CAD/CAM system from DP Technology (Camarillo, California). Esprit is capable of conventional, interactive 3D design and NC programming functions, but also provides a range of tools by which users can capture recurring machining processes so that they do not have to be regenerated time after time. These automation capabilities can be used rather simply--for example, to string together a hole-drilling and tapping sequence. Hunter, however, has taken them to the extreme.

Unlike a conventional NC programming process that is driven by a linear sequence of decisions applied to specifically defined workpiece geometry, knowledge-based programming is driven by generically defined workpiece features that are tied to generalized machining processes. Consider, for example, programming a counterbored and threaded bolt hole for a machining center. In an interactive process, the programmer selects a drilling operation and the tool to be used, then specifies the hole centerpoint and depth as well as the feed and speed. Then he decides that the chamfering operation comes next, and repeats the same programming process. And then on to the tap.

In the knowledge-based process, however, the entire machining process is already determined, and triggered as soon as the "threaded hole" feature is recognized. Operation sequences, tool selections, feeds and speeds are all selected automatically based on logic and/or information tables built into the CAM system database. To make the generic process specific, variables such as hole size and location, thread specification and workpiece material are plugged in. (This may be a manual data entry task, or the system may be programmed to automatically recognize features and their attributes--such as materials, tolerances, and so on--in an appropriately structured CAD file.) In any case, once the variable questions are answered, the generic process is "sized" accordingly. The thread specs drive the initial hole size; which drives the drill and chamfer tool selections; which, in combination with the material, drives the feeds and speeds, and so on.

This methodology is so wide open in its potential that creative process planners can automate the programming of families of workpieces far more complex than one might consider practical, which brings us back to Hunter Innovations. The hip implant piece is a good example of what can be done. While the general configurations of all pieces are essentially similar, each is unique in terms of its specific dimensions. It's not just a matter of scaling the entire piece up or down. Various geometric proportions of different piece features change relative to each other, some of which can be described only with full surface definitions. Even so, Hunter has devised a way to generically define the piece with just seven features, each of which is tied to a machining process. Three of the programs are two-axis turning operations while the other four are three- or four-axis milling operations executed on a vertical machining center.

The programming process is still not entirely automatic, but it takes Hunter less than half the time than the previous method of programming each piece conventionally. When the scan file comes in, it is still in the form of a series of cross sections. First, the file must be converted to a composite surface definition, which is itself a proprietary trick that Hunter has perfected. Then they go about the task of marrying that specific geometry to the predetermined features and their corresponding processes.

With some features, it's a relatively simple matter of parametrics. The distal, for example--the long shaft that extends down into the bone--is just a cylinder. Length and diameter values are all that are required to set the specific feature geometry and then automatically create the proper turning program. On the other end of the workpiece, programs to generate the femoral neck and taper are similarly created.

But the middle of the workpiece is a different story. Here, geometry can only be described with complex surfaces that must blend smoothly into the cylinders at either end. These features cannot be defined parametrically, so Hunter has devised a method that essentially replaces the generic surface geometry with that of the specific workpiece geometry. Then the 3D tool path is automatically altered accordingly.

So even though every hip implant that Hunter manufactures is different, each is made with exactly the same process. In the shop, the seven feature programs are spread across four different setups, which do not change. This provides for efficient part processing as well as consistent quality. From the time Hunter receives a CT scan, it takes just three days to produce the implant complete. And that includes a special surgical broach (that exactly matches the implant geometry), and perhaps a reamer too, that must also be manufactured for the implant surgery.

Capturing Methods

According to DP Technology's Paul Ricard, while Hunter's application may be exotic, similar uses of knowledge-based machining technology are growing increasingly common. "Families of parts are the most obvious application," he says, but lower level automation of recurring features is quite common as well.

One reason the use of these automation tools is growing is simply because it is getting much easier to do. Yet the systems can still yield surprisingly powerful results. In the case of Esprit, for instance, the system comes with an expert system shell into which rules can be entered to automatically manipulate input data as well as define multiple step processes. Creating such capabilities can be a matter of establishing decision trees with multiple branches. For example: If hole diameter is < 1 inch, insert reaming operation. If hole diameter is > 1 inch, insert boring operation.

But that's not the only way to capture and express methods know-how. A generic feature-driven process can also be constructed essentially by a teach method. In this case, the base program is created in a conventional interactive manner with hard numbers for the workpiece geometry. Then the programmer goes back into the program and substitutes variables for the fixed geometry. When the variables are replaced with hard numbers for a specific workpiece, the program is modified accordingly. Or, instead of simple variables, the programmer may insert some sort of logic that can be quite sophisticated. Says Mr. Ricard, "You can do much more than parametric programs. You can write a global routine that will automatically order and sort a set of cross sections and then create a lofted surface from them. You can compare volumes with Boolean equations. You can assemble a series of surface patches into a solid or composite surface that gets turned into a feature." And once you have a feature, you can associate a process with it and automatically generate the program.

The ability to capitalize on these capabilities has become one of Hunter Innovations' key competitive advantages. They've successfully automated other complex families of parts, and now are looking for additional applications outside the medical and waterworks industries. Says Mr. Powell, "Our goal is to find lower volume, complex components that can be divided into families, and where customization is needed."

When they find those jobs, Hunter will to be tough to beat. They have the skills and the tools. And for the right applications, they can make unique "one-offs" seem like running the very same part over and over again.