This article focuses on the application of Generative Design to mechanical engineering. Note that Generative Design also refers to the autonomous generation of electrical schematics based on diagrams, as well as placing components and routing traces through circuit boards that are based on diagrams.
Generative Design Terms and Definitions
Generative Design is the capability of Mechanical Computer Aided Design (MCAD) applications to autonomously generate one or more geometric designs to satisfy specific objectives and constraints. It is characterized by a number of traits:
- Component Focused: Currently, this capability is applied to the design of individual components. In the future, the scope of its application may expand.
- Autonomous Execution: Once initiated, it works autonomously of additional user input. In this way, it operates similarly to a gradient-based structural optimization capability.
- Goal Driven: The decision-making algorithm that MCAD applications use is driven to achieve an explicitly defined goal or implicit objective. Examples might include minimizing deflection under load or total weight.
- Constraint Bound: This decision-making algorithm is also bound by constraints defined by the user. The breadth and depth of constraints can vary from solution to solution. This can include design constraints related to engineering physics, such as peak stress. It can include manufacturing constraints, such as producing geometry that can be pulled from a plastic injection mold.
Generative Design Decision-Making Algorithms
While running, Generative Design capabilities leverage a decision-making algorithm to determine the shape of the design’s geometry. Today, there is not one such algorithm in use. Instead there are many that vary from solution to solution. These types of algorithms include:
- Topology Optimization: Developed in the mid-1980s, this engineering structure-based algorithm conducts an analysis of an existing piece of user-defined geometry and then removes material not carrying significant loads. This procedure is run progressively, removing material repeatedly over time, ultimately producing a final design.
- Biomimicry: Established more recently, this algorithm mimics behaviors seen in nature, such as replicating growth of bacteria colonies, the growth of roots and branches in trees, or the evolution of bone structures, to optimize weight-to-strength ratios.
- Morphogenesis: This algorithm leverages research on how groups of cells respond to their environment. Cells actively loaded grow stronger. Cells that are not loaded are discarded.
Topology Optimization is considered a Subtractive Generative Design method because it progressively removes material. Biomimicry and Morphogenesis are considered Additive Generative Design methods because they grow or add material to the design.
Compute Platforms for Generative Design
When first introduced, Generative Design capabilities were closely associated with Cloud-Based MCAD applications. However, this technology is now also available with desktop offerings. Note that compute power is an important consideration for this technology.
Cloud platforms offer access to elastic compute resources. Additional cores and storage space can be allocated to Generative Design capabilities. An alternative to such cloud-based resources, however, can be found in high-performance computing on local networks. Generative Design runs on local desktop resources as well. However, as a user requests more designs to be generated, more of a desktop’s compute resources will be required.
Generating One Design, or Many
A difference in the Generative Design capabilities of different MCAD applications lies in the number of geometric designs that are produced. Some generate one. Others generate many.
Solutions that generate one design often leverage topology optimization algorithms, as the application starts with existing geometry and removes material. Solutions that generate numerous designs often leverage biomimicry and morphogenesis algorithms. Such algorithms provide variability against the same input, which results in different design geometry.
As Generative Design develops the geometry of a design, finite elements or voxels are often added or removed. These discrete geometry elements have flat faces. Removing or adding them means that the outer surfaces of the resulting geometry are also composed of planar surfaces. This is Faceted Geometry.
There are limited geometry modeling tools that can manipulate Faceted Geometry. Parametric Modeling capabilities cannot be used, because there are no discrete features or parameters. Direct Modeling is not effective, because this geometry is not in the traditional prismatic shape. Mesh Modeling, however, provides a specialized set of tools to change Facet Geometry. These capabilities are important for those who want to make changes to Facet Geometry for its use with 3D Printing, which can use such models as they are.
Some solutions have developed new post-processing capabilities that smooth Facet Geometry, once Generative Design is complete. This geometry cannot easily be edited by Parametric Modeling or Direct Modeling. However, it can be manipulated with surface modeling tools. This is important for those who wish to use traditional manufacturing methods for production, such as machining or tooling.