Computational Design

Computational design extends beyond traditional CAD by enabling engineers to create custom design tools and automated workflows. While CAD operates as a digital drafting tool, computational approaches allow for the development of parametric systems and design algorithms tailored to specific challenges.

Generative design builds on this foundation by automating the exploration of design possibilities. Engineers define performance criteria and constraints, allowing algorithms to iterate through solutions - much like natural evolution. The process evaluates thousands of variants simultaneously, optimising for multiple factors such as structural efficiency, material usage, and manufacturing constraints.

This systematic approach enables:

Development of custom design tools and automation scripts
Discovery of solutions beyond conventional design patterns
Rapid adaptation to changing requirements
Integration of complex performance criteria into early design stages

The methodology has proven particularly valuable in aerospace, architecture, and industrial design, where complex requirements demand sophisticated optimisation approaches.

Evolution of an UAV Drone Wing

The aim is to replace the standard spars and ribs configuration in aircraft wings with an innovative optimized sandwich structure for future unmanned aerial systems.

In the chart on the left you see the evolution of the wing structure during the physics driven local parameter optimisation of beam-thickness and voronoi seed point distribution.

Progressing faster by teaching a surrogate model (FNN) the effect of the design parameters on the simulation results.

With only 1200 simulations the real cost function can be well represented (see right). The more simulations are done the more data can be fed back to the model for retraining. Read more on differential engineering: here

Evaluated 20+ optimisers whereas Differential Evolution is the best fit.

(Already ~8k functional evaluations lead to a global optimum in
> 90% of cases.)

Tools used:

Python 3.10 with Sklearn, Torch, Pandas
Hyperganic Voxel based Geometry (c#)
Hyperganic Immersed Boundary Method (IBM) structural simulation solver (c#)

Algorithmic Positioning of Pins in a Cold Plate

Modern HPC chips are designed to deliver ever-increasing performance. However, due to their intense computational demands, these chips generate significant heat and require efficient cooling to function properly.

Cooling of heated surfaces is typically achieved through conduction or a combination of conduction and convection.

Impinging cold plates primarily leverage the effects of convective heat transfer, while also utilising conduction. Read more to conduction vs convection: here

This dual-mechanism approach is highly effective for cooling high-heat-flux surfaces found in HPC chips.

In an impinging cold plate design, coolant jets are directed to hit the hottest spots on the chip surface that needs to be cooled. A branched inlet system is often employed in these designs (like above). Read more about Impinging Cold Plates: here

This system guides the coolant to multiple nozzles, where it is then sprayed onto the heated surface. The branched design ensures uniform coolant distribution across the chip, targeting multiple hot spots simultaneously.

Computational Design enables the engineer to rapidly explore positioning and options of pins to conduct heat effectively.

In the image to the left, the temperature distribution (shades of grey) on the heated surface is indicated, as well as the position, where the impinging jets hit the heated strips (red).

Adhering to manufacturing and functionality constraints, workflows can be implemented effectively positioning pins according to the highest heat load in all areas where the positioning is "allowed".

Tools used:

Rhino 8 + Grasshopper (Kangaroo2)

Mass-customized 3d-printed Insoles and Saddles

Traditionally the manufacturing of custom insoles has been time-consuming and material-intensive, as the process of milling thermoplastic or EVA materials often resulted in orthotics with uniform density and limited design flexibility.

However, with the advent of 3d-printing it is now possible to embed various mechanical properties directly into a material, offering customisation and precision that was previously challenging to achieve.

Hyperganic partnered with Elite Orthotics, which resulted in an application called Doc Sols, powered by Algorithmic Engineering and dedicated to designing custom insoles, being able to generate intricate lattice insoles in less than 10 seconds.

The use of lattices and TPMS (or in other words meta materials) really enable this field, as single material components can suddenly perform multiple tasks at once. Thus one encodes functionality into the components.

Why Metamaterials are Useful

Meta materials are engineered structures with unique properties that are not found in naturally occurring materials (bulk materials). They offer several advantages and are particularly well-suited for 3D printing applications.

Unique Properties

Metamaterials can exhibit extraordinary characteristics such as:

Negative refractive index
Acoustic manipulation
Mechanical properties like negative Poisson's ratio

Customizable Behavior

The properties of meta materials come from their structure rather than their chemical composition. This allows for fine-tuning of their behavior by adjusting the arrangement of their components.

This approach eliminates the need for complex computational design tools or even knowledge in the field as all relevant design parameters are exposed through an easy to use User Interface (UI). The knowledge of the orthopedist/ saddle maker as well as the knowledge on how to design these devices is encoded and implicitly available through the functions in the UI.

Tools used:

Hyperganic Voxel based Geometry (c#) including UI Framework (Qt)
Rhino 8 + Grasshopper
Qidi Studio + Qidi 4+
Python 3.10 with Pandas

Page updated

Google Sites

Report abuse