🎯 Optimisation
Two ways to find a good design, both by moving the sliders of a definition and solving it again:
- Explore a design space with the Explorer components: every combination of a few sliders, one solution each, with the results, a picture and the 3D model of each design saved to a folder. The Alpaca4d Explorer opens that folder in the browser, to filter the designs on parallel axes, compare trade-offs and look at any of them in 2D and 3D.
- Search for the best design with Non-Linear Optimisation: an algorithm moves the sliders towards the smallest or largest value of one number, such as the mass, and can keep other numbers, such as the utilisation, within limits.
They do what Colibri with Design Explorer, and DoDo's NL-Opt, did. Both ship with Alpaca4d, so there is nothing else to install.
A third works on the material itself rather than on sliders: Topology Optimisation finds which part of a block of bricks or tetrahedra a structure needs. BESO Topology finds it by carving the block down, solid or void all along, and Multi-Material Topology also chooses which of several materials goes where.
Examples to start from:
- the Footbridge Study, a study of 108 designs for the Explorer;
- the Non-Linear Optimisation cases, eight problems with known answers;
- the Topology Bridge, where a bridge needs its steel;
- the Multi-Material Bridge, the same block in steel and concrete;
- the BESO Bridge, the same block carved out by BESO;
- the Truss Node and the Beam-Column Joint, nodes where members meet in space, each found by all three topology optimisations side by side;
- the Topology Benchmarks, two problems checked against top3d.
Components in this tab
| Component | Nickname | Does |
|---|---|---|
| Explorer Inputs | Explorer Inputs | Steps through every combination of its sliders, value lists and panels. Press Fly. |
| Explorer Outputs | Explorer Outputs | Names the results a design is judged by. |
| Explorer Record | Explorer Record | Saves each design: data.csv, a picture, a 3D model. |
| Explorer Image | Explorer Image | Which views to take a picture of, and how big. |
| Explorer Subset | Explorer Subset | Runs part of a study, to try it out or split it between machines. |
| Non-Linear Optimisation | NL-Opt | Minimises or maximises a number by moving sliders, with NLopt's algorithms. |
| Section Optimiser | Section Optimiser | Sizes steel beams from lists of sections, to EN 1993-1-1. |
| Design Group | Design Group | Beams the Section Optimiser sizes from one list of sections. |
| Topology Optimisation | Topology | Finds where a solid of bricks or tetrahedra needs its material. |
| Multi-Material Topology | MultiTopology | Finds where each of several materials goes in a solid, and where none is needed. |
| BESO Topology | BESO | Finds where a solid needs its material by carving it down, every element solid or void. |
Which one
| Explorer | Non-Linear Optimisation | |
|---|---|---|
| Answers | What are the options, and what trades off against what? | Which design is best by this measure? |
| Variables | A few, each in a few steps | Any number, each as finely as its slider allows |
| Results | As many as you like, weighed afterwards | One objective, and constraints it must meet |
| Designs solved | Every combination, once | As many as the algorithm needs, one after another |
| Leaves you with | A study folder to explore and share | The sliders on the best design, and the history of the run |
They work well together. A study shows where the good designs are and what matters; narrow the sliders' bounds to that region, then let the optimiser find the best design inside it.
Exploring a design space
- Wire the sliders, value lists and panels that make a design into Explorer Inputs. Name them: each name becomes an axis.
- Wire the results that matter - mass, maximum displacement, utilisation, first period - into Explorer Outputs, and rename its inputs to name the columns, units included:
Mass [t]. - Connect Inputs and Outputs to Explorer Record, and the AlpacaModel to its Model input to get a 3D view. Set Write to
true. - Press Fly on Explorer Inputs. Rhino stays usable while it runs; Fly again or Esc stops it.
- Open the Alpaca4d Explorer and drop the study folder on it. Right-click Explorer Record ▸ Open the Alpaca4d Explorer opens both.
Note
A stopped study picks up where it left off. Explorer Record never saves a design twice, and Explorer Inputs skips the designs already saved without solving them. Flying again after a stop, a crash or an extra slider step only runs what is missing.
What the explorer does with it
- Filter by dragging along any axis; combine as many as needed.
- Goals: click a result's name and pick Minimise or Maximise. Pareto front then keeps the designs no other design beats on every goal at once.
- Gallery, table, scatter of the designs that pass, sorted and coloured by any column.
- Each design: its picture, its 3D model, every value with where it sits in its range, and arrows on each input to step to the design next door, among those that pass the filters.
- Rate designs with stars and export the ones that pass as CSV.
A study folder on your computer is read by the browser and never uploaded. To share a study, put the folder online - a GitHub repository works - and open it by its address: the address bar then holds a link to the study, and to the design picked.
Searching for the best design
- Drag from the Variables grip of Non-Linear Optimisation to the sliders the design depends on.
- Drag from the Objective grip to the number to minimise or maximise, such as the total mass.
- To keep other numbers within limits, drag from the Constraints grip to numbers that must end up at or below zero, such as
utilisation - 1. - Double-click the optimiser, pick an algorithm and press Start. When it ends, the sliders are on the best design it found, and one Undo puts them back.