Approaches of Mechanical CAD

Direct Modeling

There are at least five types of approaches to creating a digital model in the domain of mechanical computer-aided design. The most popular, somewhat synonymous with CAD itself is direct modeling. It is a modern equivalent of drafting. The engineer uses visual "entities" such as line segments, arcs, text and a pointing device and monitor in the place of pencil, ruler, compass, protractor and drafting board to create a 2d or 3d model which is essentially a database of these entities, often still called a drawing, which can be queried and viewed visually as if it were layers of vellum on a drafting board. The entities do not all come from paper drafting but include data types which have evolved in the computer assisted drawing environment of CAD including Bezier and b-spline curves and surfaces, transformations, solids and object-based "features" which can represent holes, fillets, threading, knurl or teeth of a sprocket or gear for example.

Schematics

There is one problem with direct modeling that causes engineers to seek alternative approaches to computer aided design, and that is since it is purely a visual rendition of a particular design, even with notes, one cannot adequately convey the design intent or rationale behind a particular design. Conveying intent is a way for communicating understanding of a system. And this understanding needs to be transmitted from one generation of engineer to another from the source. There are other approaches, which can do this to varying degrees. The first, which is as old as drafting is schematic modeling, which uses diagrams, symbols and flowcharts to explain the purpose of a system being modeled. These visual representations, or schematics, often lack dimensions and angles of direct modeling, but essentially serve as a map of a process. For example, a motor might be represented with a symbol, but could in reality be any type of motor such as electric motor or internal combustion engine, but those details are intentionally left out for the viewer to get a high level understanding of the purpose of a system. A schematic is often, but not always, the starting place in a design of nontrivial complexity.

Parametric Modeling

The next approach to modeling is parametric modeling. Parametric modeling is simply a variation of direct modeling which uses geometric equations to both assist in the layout of a design and communicate the intent of a design. The equations, often called "constraints", define distances, angles, parallelism, orthogonality, incidence, tangency, concentricity and other geometric factors of a design. While parametric modeling is a step in the right direction, it cannot fully communicate the design intent of a model since the equations are often limited to those of a geometric nature, and not all intent is purely geometric.

KBE

The next approach to modeling is knowledge-based modeling, the practice of which is called knowledge-based engineering, or KBE. KBE systems in their full form are essentially Object Oriented programming languages which define the intent of the model in terms of their object hierarchy and attributes (or properties) of the objects. The attributes are defined by rules. Rules are logical expressions which define the relationship between an input and an output. Rules, which are often symbolic, can greatly express the design intent of a model or system. Between rule based design and parametric design, engineers can encapsulate almost all the information about a design. However, defining formal rules is often more difficult an undertaking than sketching or drafting which is the strength of direct modeling. For instance, knowledge-based modeling (or rule-based design) cannot express aesthetic principles behind a design. It is also more difficult to express ergonomic or psychological principles with rules. Since we live in a 3 dimension space world with a dimension of time, it is often the case that a drawing can convey subtleties of design intent where rules would become cumbersome or confusing. This type of design intent is often non-symbolic in nature, such as the spaciousness of an interior, or the smoothness of a control. Finally, rule based design, even with it's time saving in generating configurations or variants of a design, can actually be more time consuming than just drawing something in the first place.

A challenge

So one of the current challenges with computer aided design is to find a natural and effective marriage between rule based design and direct modeling. The solution for combining direct modeling with rule based coding is not as simple as parametric modeling might make it seem, as logical rules go beyond simple geometry. There may have to be some evolution of the approaches software vendors use to solve this problem. This could be a UI which is a sort-of combination of visual code editor, database management system, direct modeling system and version control system. The workflow must be aimed at capturing design intent, such as by recording what commands are used, what object snaps are used within commands and gently steer the user away from anonymous numerically-oriented fixed geometry and toward the establishment of a named-entity prototype drawing focusing on features and relationships rather than dimensions and angles. That way the model can be reconfigured, reused, subclassed and extended. Building any of these UI components: a visual code editor, a database management system and a intuitive direct modeling system are all non-trivial software engineering tasks. So far modern KBE systems have focused on browsers and text editors and have been mostly devoid of point-and-click oriented construction.

Generative AI/Machine Learning

The final and newest approach to computer aided design I will discuss is generative AI . It is also a meta-approach to design, where it has the ability to incorporate the four approaches of computer aided design modeling I have described. This meta-approach is based in machine learning which is essentially linear-algebra. Large Language models have demonstrated their ability to learn from and write code, so this type of AI is gaining the ability to generate rule based models. Also, AI has shown it's effectiveness in generating images based on descriptions or other images, and will likely be just as good at generating direct modeling drawings. So one type of marriage between rule based design and direct modeling would be to allow a generative AI to create all types of models for the same design, while being introspectable for engineers to glean full intent from an AI's generated model.