Andrew Anagnost
Sr. Director CAD/CAE Products
Autodesk Inc.
San Raphael, Calif.

Edited by Leslie Gordon

A recent Aberdeen Group study showed that best-in-class manufacturers — those that get products to market faster with the lowest change costs — build about half the number of physical prototypes than other manufacturers. The reward: they get to market, on average, 58 days faster. A main factor in their success is digital prototyping. Here, information flows through a digital pipeline that remains unbroken throughout the product’s life-cycle.

Currently, most 3D CAD software does not make it easy to do fully integrated digital prototyping. Many users find the software is too complex, or they use CAD simply to document design problems they’ve already solved. For instance, a company might use 3D models to test form and fit and automate the creation of 2D drawings. What is needed, however, is the ability to quickly and easily build digital prototypes that let users test form, fit, and function without requiring them to be expert in generating geometry or managing geometric constraints.

Think about how you use word-processing software (like I am right now). Users need not worry about altering typefaces so documents will print, or what will happen after deleting a paragraph. They simply turn ideas into words and let the software take care of printing and text reordering.

Unfortunately, CAD functions that are a direct analogy to word processing’s Print and Delete still hinder many users. For example, unless 3D models are built by knowledgeable designers, the models will not necessarily let users answer ques t ions about how the final products will operate in working environments.

Also, the way parts are assembled in 3D CAD does not let users test the function of the parts. It only lets users put the geometry together. Thus, 3D programs force users to solve geometry problems, not engineering problems. And today’s software doesn’t make it easy to build digital prototypes. For one thing, users follow different paths to build digital prototypes than to build geometric models. Users must also deal with CAD functions such as face-to-face mates and edge constraints that are not real-world inputs — they are mathematical abstractions that were devised so the mathematics of the model could be solved on a computer.

These difficulties mean most manufacturers still rebuild models or hire outside experts to use geometric models to assess function. Or, more likely, companies do nothing at all. This leads to unpleasant surprises on the factory floor and lengthy physical prototyping cycles.

Under the hood of a typical 3D modeler
Many design methods have attempted to address the limitations of parametric feature-based modeling software. Among them are history-free modeling, rulesbased design, and even customized systems. However, the answer lies not with any of them alone, but with a combination of engineering and modeling called “functional design.” For a good grasp of functional design, it is helpful to first understand the attributes of typical 3D software and then take a more detailed look at the techniques that have tried to solve modeling problems.

First, 3D programs are built on top of a geometry kernel (either a broadly available or a proprietary one), i.e., a topological solver that does all the heavy mathematical lifting. The programs also create a history of features that let users define geometric components such as fillets, chamfers, and holes in a sequence of operations that mimic manufacturing processes. And the programs have constraint solvers that let users alter the size of sketches and parts, and place them relative to each other in space.

A parametric solver lets users vary the relationships of size and position with numbers and mathematical formulas. The solvers let users change one piece of geometry based on the size of another — such as the diameter of a hole that varies with the size of the bolt that must fit through it.

The kernel is generally invisible to users. The history tree and constraints must be actively managed by users and are the primary difficulty associated with creating, editing, and validating designs with 3D models.

In fact, designers and engineers know exactly how easy it is to get into trouble with 3D modeling. Features are created with a history — that is, they depend on features created before them — so deleting a feature early in the history can result in a model that either fails to regenerate (think of this as a document that won’t print) or no longer represents design intent (imagine your Times New Roman font printing out as Gothic). Users must thus pay strict attention to the order in which features are created (called the “feature tree”) to get the right result. Only then do they get intelligent models that can be reused by others and edited rapidly to respond to design changes.

Parametric constraints have similar problems. When users don’t pay attention to how and when they apply constraints, they build models that are either impossible to change or not usable by others, let alone being useful as digital prototypes.

Past techniques
When it comes to the techniques that tried to solve modeling problems, one mentioned above is history-free modeling. It delivers an intuitive interface that lets users quickly and easily create models without worrying about the order of modeling operations. The models are stable (meaning they almost always regenerate after a change), but they lack the design and manufacturing intelligence of parametric features. History-free modelers build models that cannot be used to validate form, fit, and function of designs. They are, in many ways, a move backward to modeling solutions that preceded parametrics and history, and remind us why those technologies evolved in the first place.

Another method, knowledgebased systems, tries to describe an entire product from engineering rules, and then use the rules to create 3D geometry that can be used to validate form, fit, and function, as well as generate 2D drawings. Essentially, rules-based programming is layered on top of the CAD system to program 3D models. This works well with highly repeatable designs (for example, conveyors, escalators, and elevators). But the cost and expertise needed to create intelligent models is prohibitive. The technology is currently limited to companies that can justify the consulting expenses.

Last is custom designs, prevalent in large automotive and aerospace companies. These systems consist of a series of applications (usually based on a core 3D CAD package) purpose-built to automate or simplify specific design and digital prototyping tasks that are done over and over again each time a car or aircraft is designed. The highly regulated nature of these industries, along with the per-unit costs and product volumes, justify investing in these tools, but few of them are in widespread use.

The functional fix
As mentioned previously, 3D CAD users are largely unaware of the mathematics needed to create computer representations of 3D parts. The same should be true for history and nonhistorybased features, constraints, and parametrics. A functional design approach with Inventor software, for instance, pushes these capabilities back to the level of kernels operating behind the scenes. The software exploits the feature history, delivers the ease of use of local history-free operations, and leverages rules to generate parametrically constrained geometry, all from simple functional representations of a product design.Functional design does not eliminate the need to solve geometricmodeling problems, but it minimizes the amount of time users spend on such problems. And it ensures the generated geometry is usable as a digital prototype.

To do this, functional design tools have what is called a functional engine. It sits on top of the various modeling kernels and technologies and uses 2D and 3D schematics to create 3D geometry that is editable and also a digital prototype, i.e., a model that can be validated for form, fit, and function.

Complete automation of designs will probably never happen, but the number of cases that are targets for automation far exceeds those that are not. Simple examples come from shafts, pulleys, and transmissions. More complex examples come from conveyors, drivetrains, and wiring and piping systems. Their functional requirements can be captured and the rules that govern them are well understood or easily customized.

Thus, the functional approach captures real-world design intelligence earlier in the cycle because design elements are active, not static. The elements reflect the behavior of parts they represent, including relationships to other parts. For example, when tools in the software build torque and speed characteristics into a gear, the component knows it must interact with another gear — it is not just a disk with a certain number of teeth and a set of dimensions.

Functional design lets users generate parts from real-world inputs such as speed, power, and material properties. Simulation in the software continually computes dynamic operating conditions. This lets users, at any stage of the design, size motors, select actuators to sustain certain loads, and analyze positions, velocities, accelerations, and loads that affect each component in a mechanism. The software also replaces abstract entities such as edge constraints with functional, real-world constraints in the form of sliders, ball joints, and hinges.

Engineers, thus, can solve engineering problems early on, with the simplest representation — sometimes a 3D model, sometimes a sketch. Once the problem is solved, the software generates the correct geometry.

Functional design brings exciting changes to the CAD industry. The next 10 years will bring even more change. By then, 3D design tools will be used in all aspects of manufacturing by tens of millions of users worldwide to rapidly validate form, fit and function using digital prototypes. This future vision requires that design-software developers introduce functional design tools that let users focus on problems they’re trying to solve.

Make Contact Autodesk Inc., autodesk.com