Module 8 Questions:
- Defining objectives determines which performance metrics drive the solution space and thus shapes the geometry bias. For example, minimizing construction cost favors compact, low surface area massing, whereas maximizing daylight yields more elongated or perforated forms.
- Taxonomies partition the dataset into algorithmically distinct families. This divide-and-conquer approach keeps workflows modular and ensures each subset is parameterized uniquely, which enhances diversity control and prevents overlap or repetition.
- Automated workflows
- Benefits: automation enables large-scale, reproducible sampling, it embeds ground-truth labeling, and modular scripts support easy expansion and variant generation
- Challenges: ensuring unbiased sampling, managing computational load of simulations; maintaining workflow flexibility
- Iteration lets you refine input ranges, correct sampling gaps, and improve surrogate accuracy. By tweaking objectives or constraints and rerunning studies, you progressively converge on better-performing designs and more robust datasets.
- Workflow:
- Define design variables (inputs)
- Define objectives (outputs)
- Map inputs to outputs
- Select a solver
- Generate design options
- Explore results
- Export preferred options or tweak and repeat
- Approach in Dynamo:
- Large and parametric: script parametric building blocks
- Expandable: encapsulate taxonomies in reusable Dynamo nodes so new variables can be added without reworking the core graph
- Randomize, cross-product, like-this, optimize
- I would combine random sampling for initial exploration with optimization for target refinement.