CAD follows new scripts.
While the overall productivity obtained with traditional CAD/CAM systems may vary depending on the application, some of the highest levels of productivity have been achieved with the aid of programming languages that automate repetitive tasks. These languages allow engineers to create a program, called a script, directing the CAD system to perform a particular sequence of instructions for creating geometry; designers can then supply dimensions and the script generates the appropriate entity.
Most large CAD/CAM installations employ one or more programmers to create and modify these scripts, which are also called parametric programs. (For a discussion of parametric design systems, see the CIME section of ME, Jan. 1989, pp. 58-73.)
Eventually, parametric programming evolved from an optional feature to become the foundation for a number of CAD/CAM systems, including those developed by Automation Technology Products (Campbell, Calif.), Parametric Technology Corp. (Waltham, Mass.), the Patran division of PDA Engineering (Costa Mesa, Calif.), and Synthesis Inc. (Bellingham, Wash.). However, use of the term parametric can be confusing, because two additional approaches to programming - one employing variational geometry, the other, artificial intelligence - are frequently placed in the parametric category. The confusion may arise because all of the approaches appear to proceed according to the same principle: each allows engineers to create geometry by entering variables - typically dimensions - and invoking a script, instead of separately generating lines, arcs, and splines, as would be the case with a traditional CAD program.
The use of scripts is complemented by an ability to handle constraints, a feature shared by all of the approaches. A constraint governs the relationship among two or more geometric entities and holds even as the entities' dimensions or orientation changes. For example, users of parametric, variational, and knowledge-based engineering systems may establish a constraint indicating that two lines should remain parallel even when the angle of one of the lines is altered.
Despite the similarities, the approaches differ substantially in the kinds of problems that they are best suited to solve. Variational systems allow the designer to interactively sketch generic, 2-D geometry - for example, an undimensioned or partially dimensioned four-bar mechanism - and then to constrain that geometry by linking it to equations governing its behavior or by specifying its orientation. Next, the software determines the dimensions and other variables that satisfy the user's constraints. This allows the application of variational systems to engineering tasks that are not well understood or that are invetigative, such as the early stages of conceptual design for a new product.
Parametric systems usually provide generic geometric entities - for example, chamfers, holes, slots, and countersinks - that can be customized by entering exact dimensions. (These entities are often called features.) However, the designer must know which dimensions are dependent on others and in what ways. Therefore, parametic systems may be most useful when the design task at hand is well characterized: for instance, when engineers must develop the next generation of an existing product or vary current designs to create families of parts.
Knowledge-based systems operate much as parametric systems do, but their use of artificial intelligence and object-oriented programming allows knowledge-based systems to incorporate nongeometrical information, such as manufacturing, cost, and material criteria, with geometry. Accordingly, users of a knowledge-based system can constrain geometry so that it follows corporate engineering practices, for example, with the practices represented as rules within the system's knowledge base. With this capability, knowledge-based systems can inform users that manufacturing does not have the tools to cut a part as specified, for example, or they can suggest design alternatives that may be less expensive to produce. Therefore, knowledge-based systems may be most appropriate for the organization and dissemination of design and manufacturing information as well as the integration of complex design and manufacturing tasks.
The systems' basic differences can be conceptualized in this way. All of the systems allow engineers to create and iterate designs and then to observe changes from one iteration to another. However, the geometry-based approaches taken by variational and parametric systems attempt to accomplish this goal by linking conceptual design with first-order analysis, in the case of variational systems, and by automating the process of changing standard designs in the case of parametric ones. Knowledge-based systems do so by allowing design and manufacturing criteria to drive the generation and evaluation of designs.
The technology for variational geometry systems was developed in the early 1980s at M.I.T.'s CAD Lab. Later in the decade, commercial systems were introduced by Advanced Graphics Systems (Tulsa, Okla.), Cognition Inc. (now owned by Automatix and based in Billerica, Mass.), Iconnex (Pittsburgh, Pa.), MCAE Technologies Inc. (San Jose, Calif.), and Premise Inc. (Cambridge, Mass.). The programs may be roughly characterized as optimization tools for conceptual design and are commonly used for mechanism design, beam deflection analysis, tolerancing (with stack-ups), and other tasks.
Normally, the programs provide a 2-D sketchpad, a spreadsheet for entering variables, and a means for linking engineering equations to geometry. They may also provide links to analysis programs, as Cognition's Mechanical Advantage does to the Beasy boundary element analysis program developed by Computational Mechanics Ltd. (Billerica, Mass.). Users of the systems typically generate and evaluate designs by creating rough sketches, appropriately constraining the geometry, entering several sets of parameter values, and comparing the resulting design interations. They can also determine optimal values for a parameter by creating sensitivity plots, which show how a performance parameter by creating with respect to a design parameter.
Variational systems are frequently labeled as parametric tools partly because they allow engineers to iterate designs by manipulating parameters. However, variational and parametric systems computer parameter values differently. Parametric systems use a sequential equation solver, which normally requires users to know in advance the steps that must be carried out to create and iterate geometry and to indicate what comes first, then second, . . ., then last. On the other hand, variational systems employ a simultaneous equation solver and accept somewhat less structured input. For instance, variational systems can generate fully dimensioned sketches even when some dimensions are not initially known. Changing Course. These differences extend to constraint propagation. In parametric systems, constraints are unidirectional. For instance, the user of a parametric system could establish a parallel constraint governing two lines A and B by indicating that the lines should be separated by a certain distance and that B should be some angle from the datum. Accordingly, if the angle B changes, the parametric program will adjust A so that it remains parallel to B. However, this normally does not hold for the opposite case: should the user change the angle of A, the system probably will not be able to change the angle of B because the parallel constraint goes from B to A. In this way, constraints created with parametric systems typically imply causality.
Reversing the causality of constraints can be a complicated and perhaps even impossible procedure for a parametric system, depending on the situation and the program's capabilities. Users may be required to rewrite the script or to rebuild the model. Even so, some parametric programs allow users to manipulate the script with relative ease. (For this example, users of some parametric systems could reverse the causality simply by indicating that A instead of B should be some angle from the datum.)
Because variational systems use a simultaneous equation solver, they can more easily accommodate changes affecting constraints, according to Robert Light, who participated in the development of Cognition's Mechanical Advantage program. For instance, if a designer specifies that three lines should be parallel (A to B and B to C) and later wishes to make A and B perpendicular to C, he or she would do so by deleting the parallel constraints and adding two new constraints (A is perpendicular to C; B is perpendicular to C). The program would then automatically rewrite the script to reflect the engineer's changed intention.
The ability to modify constraints in this way, Light suggests, makes it easier for engineers to change strategy, as they often do during conceptual design. Moreover, while engineers may not know at the outset of design which strategy they are likely to pursue, they also may not know for which variables they wish to solve. As noted earlier, variational systems are able to generate geometry even when some parameters are not known, extending the systems' ability to handle unstructured design tasks from preliminary design tasks to the process of design iteration.
For example, consider the simply supported beam in Figure 1, where the load is assumed to be in the middle of the span. Here, the engineer might want to solve for the deflection delta first, given the values of F, tau, W, L, and E. The designer could also calculate the principal frequency of vibration, omega, for the beam given E, rho, tau, and L. Later in the design process, the engineer might wish to calculate the necessary thickness tau and Young's modulus E by specifying the values for the deflection, delta, and natural frequency omega. With a variational system, the user specifies that omega and delta are independent, and the system will calculate the values for E and tau. The variational system then reverses the causality to solve for the desired values.
The ability to reverse the order of solution is important even when the user addresses a purely graphical problem. For instance, an engineer designing a back hoe might wish to study the cut envelope, the total area that can be reached without moving the vehicle. With a variational system, the engineer would define the lengths of the hydraulic cylinders as independent parameters so that he or she could then simulate the back hoe's motion using each cylinder's length as input, as shown in Figure 2. The software would then sweep each cylinder length from its maximum to its minimum cylinder extension to drive the mechanism through its maximal motion, thereby defining the cut envelope.
Later, the engineer might wish to study the resultant cylinder lengths as the back hoe makes a horizontal cut at a depth of 10 feet. In the previous example, the variational system would calculate the bucket position given the lengths of each of the hydraulic cylinders; in this example, the engineer wishes to specify the bucket motion and solve for the lengths of the hydraulic cylinders. With a unidirectional constraint system, the engineer would probably have to redefine much of the design to drive the causality in the opposite direction. But with a variational system, the engineer specifies that the lengths of the hydraulic cylinders are now dependent parameters and adds the constraints that move the teeth of the bucket along a straight line at a depth of 10 feet. Then, the engineer can study the back hoe's motion as it makes a horizontal cut at the specified depth (Figure 2).
While variational systems' use of omnidirectional constraints may give the designer more freedom when performing engineering tasks that are not well understood, it may also impose a heavy computational burden. As the number of parts grows, the number of constraints can grow faster. And for every constraint, the solution time increases as the square of the number of constraints. Therefore, a part with 1000 constraints can take 100 times longer to generate than a part with 100 constraints. Visualization. Another potential disadvantage of variational systems is their restriction to two dimensions. According to Cliff Brown, manager of design applications for the Patran Division of PDA Engineering (Costa Mesa, Calif.), no algorithms have been developed to extend the geometry generated by variational systems from 2- to 3-D. This restriction, in turn, may limit the kinds of analyses that can be performed on the geometry. Brown also notes that some software vendors have extended parametric modeling techniques to finite element modeling; for instance, an extension of his company's P/Concept program can parametrically couple 3-D solid models to 3-D finite element models.
Finally, Brown suggests that the restriction to 2-D may limit the engineer's ability to visualize a design, especially since variational systems normally cannot generate complex curves and surfaces. (On the other hand, he says that variational systems are good tools for visualizing in-plane motion.) Brown contrasts the capabilities of variational and parametric systems by noting that variational systems may be better for situations where "you don't know how a mechanism looks but you know where you want it to go," while parametric systems may be better for situations where "you know how a mechanism looks but don't know yet where you want it to go."
The Third Dimension
Ironically, the primary criticism leveled against parametric systems also serves as their chief defense. CAD/CAM vendors introduced parametric programming in the 1970s as a means of automating tasks associated with the generation of geometry. Since these tasks were typically repetitive, users generally knew the sequence of operations required to create a frequently used part or assembly, and they usually had a good understanding of the relationships governing the geometry before they invoked the script.
Consequently, users of parametric systems that have been developed for these situations typically must know in advance what parameters serve as inputs to the script and what relationships govern the geometry. Some charge that this makes parametric systems unsuited for design tasks that are not well defined. On the other hand, Brown argues that fairly well-characterized design situations are the ones most likely to be encountered by the majority of designers. "Optimizing the design of current products is the bulk of industry's work today," Brown notes. "The stress is on innovation instead of invention."
Another advantage potentially held by parametric systems is their ability to accommodate both 2-D sketching and 3-D solid modeling. With a parametric system, the designer normally sketches geometry in 2-D and assigns constraints. After the 2-D sketch has been rectified - that is, after the process of solving the constraints has been completed - the designer may choose to project or rotate the profile into 3-D. Part features, such as holes, slots, and pockets, can be positioned relative to each other, and more complex structures can be built by combining simpler structures in a building-block fashion. Consequently, engineers can study not only the relationships governing part geometry but also the interaction of parts making up an assembly. For instance, a designer could direct the parametric system to automatically generate three versions of the same model, each including a different standard clamp design drawn from the engineering department's central data base.
Although parametric systems' use of unidirectional constraints can relieve the computational load to some degree in comparison to that required for variational systems, complex assemblies may involve hundreds or thousands of equations and parameters; since each change requested by the user normally propagates through all of these, response times may be slow. Further, depending on the program and its modeling technology, parametric systems may not be able to generate models with complex shapes, like blends, and some of the most sophisticated free-form surfaces. And although many users like parametric methods of constructing parts, others say that the modeling of very complex assemblies can be tedious.
An example of a part that may be well suited for parametric modeling would be the housing for a gate or ball valve. The independent parameter in this instance would be the inside diameter (ID) of the pipe on which the valve operates. The flanges on each end of the valve have an outside diameter (OD) equal to twice the ID of the pipe. The valve stem hole has an OD of three-quarters of the OD of the pipe and a wall thickness equal to the pipe's wall thickness. The wall thickness is one-eighth of the pipe's ID. If the designer changes the pipe's ID, all the dimensions change automatically. However, the only thing that would change if the OD is increased would be the OD and ID of the valve stem housing, since the constraints go only one way. As this example suggests, parametric programs can be effective tools for establishing relationships like these relatively quickly and easily, giving the user helpful feedback throughout the design process.
Follow the Rules
Knowledge-based engineering systems were originally created to specify very large and complex models or structures having tens and even hundreds of thousands of graphical and nongraphical entities. Consequently, the systems have usually been installed at many of the largest manufacturing concerns to integrate corporate- or division-wide design and manufacturing activities.
Knowledge-based systems are sometimes mistaken for parametric ones because they provide most, if not all, of the features that parametric programs do. However, knowledge-based systems, which are sold by ICAD (Cambridge, Mass.) and Wisdom Systems (Chagrin Falls, Ohio), use artificial intelligence techniques to accommodate more sophisticated constraints.
"The advantage of the [knowledge-based] approach is that anything that can be described in words by the engineer can be written as rules," observes ICAD's Lawrence W. Rosenfeld. For instance, governmental regulations and corporate engineering standards can be expressed as rules and entered in the knowledge-based system's rule base; these rules can then act as constraints while the engineer creates a design.
"This added flexibility enables the integration of engineering and manufacturing rules, but comes at the price of having a more language-based method of [human-computer] interaction," Rosenfeld notes. However, because knowledge-based systems typically incorporate object-oriented programming techniques, they may also simplify the integration of the many programs used by different engineering specialists participating in a development project. Rosenfeld says that this approach allows for a tight integration of many kinds of programs, ranging from commercial codes to those developed in house, as well as the input and output that they produce. Therefore, both the expertise and the tools of many engineering specialists can be linked, promoting the goal of concurrent engineering.
Since knowledge-based systems are usually customized to include the regulations, manufacturing methods, and engineering practices observed by the customer, engineering managers must be prepared to devote time and energy to the creation of a rule base. "It takes commitment and focused effort to use [knowledge-based systems] to their full potential," Rosenfeld says. But as the following example suggests, the effort can pay handsome dividends.
AT GE's Aircraft Engine Group (Lynn, Mass.), engineers have switched from a sequential design process to a concurrent one using the ICAD program. They first incorporated knowledge from all of the location's engineering groups - including those involved with aeromechanics, aerodynamics, manufacturing, structural, creep life, and thermal analysis - in the ICAD program's rule base. With this information on-line, the GE engineers were able to automatically generate a 3-D model of the airfoil, internal cooling channels, and structural blade attachments for the turbine blade.
The cross-sectional profiles for the outer airfoil were imported from their CAD system and used as input to the ICAD knowledge base. Then, drawing on its rule base, the ICAD system generated a lofted surface model of the airfoil and the internal cooling channel geometry from the imported profiles. Rules in the product model retrieved existing designs of the structural blade components automatically from on-line catalogs and incorporated them into the airfoil design. This process allowed a comprehensive 3-D model to be completed early in the design cycle, before analysis was performed.
This size and shape of the dovetail in Figure 3(a) were determined from inputs including the size of the blade and disk, the stress from aerodynamic pressures, and the rpm of the engine. The specification of a higher rpm resulted in the design of a more substantial dovetail to support the higher stress as shown in Figure 3(b). The blade's size was determined by a first-order analysis of the stress, life, and vibrational considerations; the exact representation of the surface blend between the blade and dovetail was driven by the engineering rules used to create a blend with minimum stress gradient. Moreover, the ICAD system's rules automatically triggered the creation of the several design views needed for each discipline involved in the project, including views for machining, casting, and mechanical design, as well as cost and engineering analysis reports.
GE managers contend that their approach to design automation changed the competitive balance in their industry and say that one result has been that their competitors are now adopting knowledge-based, concurrent engineering to keep up.
Although parametric, variational, and knowledge-based systems differ in important ways, the decision to purchase a design automation tool may not necessarily dictate a choice of one system instead of another. Because the tasks best performed by each are complementary, engineers may find over time that a combination of the tools yields the greatest advantage.
In general, ICAD's Rosenfeld summarizes the capabilities of the systems by suggesting that variational systems are best suited for first-order analysis for conceptual design; parametric systems are appropriate for the design of moderately complex parts and assemblies; knowledge-based systems are best applied to the design automation of large and/or complex products and processes. In any case, engineers should be sure to carefully evaluate the kinds of tasks that are to be automated and then select the tool that best fits the job.
PHOTO : Figure 1. Linking geometry to equations.
PHOTO : Figure 2. Comparing parameter values: (a) cylinder length; (b) cut envelope.
PHOTO : Figure 3. Turbine blade. The dovetail (bottom) connects the GE turbine blade to the rotary disk in the jet engine: (a) standard dovetail; (b) enhanced dovetail with blending done by ICAD's rule-based surface modeler.
|Printer friendly Cite/link Email Feedback|
|Title Annotation:||new programming tools for computer-aided design systems; includes related article on spreadsheets|
|Date:||Nov 1, 1989|
|Previous Article:||Repackaging the supercomputer.|
|Next Article:||Buckling flows; exploring the origins and structure of turbulence.|