Masuda Naoya

Journal Entry For
Module 8 - Gen Des and ML
ACC Folder Link
Link to Student

Module 8 Questions:

  1. Importance of Setting Design Objectives

From my experience working on construction sites in Japan, setting clear design objectives is extremely important. The overall cost and construction schedule can vary significantly depending on these objectives. For example, targeting a Zero Energy Building (ZEB) requires completely different priorities and design strategies compared to building a low-cost business hotel that only aims to meet the minimum occupancy requirements.

  1. Importance of Defining Target Taxonomies

Defining target taxonomies is crucial when setting up Generative Design and Machine Learning workflows. It determines what types of datasets are necessary as inputs and clarifies the nature of the expected outputs. Proper classification helps guide the design exploration process and ensures meaningful and accurate results.

  1. Advantages and Disadvantages of Automated Workflow for Synthetic Datasets

One advantage of automating synthetic dataset generation is the ability to explore a vast range of input combinations and identify optimal design solutions based on given goals. However, a major disadvantage is the lack of flexibility. Unlike human designers, automated workflows tend to work only within predefined constraints and may fail to produce highly creative or original design ideas.

  1. Importance of Iterative Process for Design Optimization

An iterative process is essential for design optimization because it allows continuous refinement and improvement. Through repeated trials, the quality and validity of the design solutions can be significantly enhanced, leading to more reliable and higher-performing outcomes.

  1. Workflow for Setting Up a Generative Design Study in Revit

To set up a Generative Design study in Revit, we first create a Dynamo script. In this script, we choose design variables and set their ranges. We also create evaluation values that measure how good the designs are. The script must be saved in a special folder so Revit can find it. After saving, we can open the Generative Design tool in Revit, select the study type, and run the study to create many design options.

  1. How to Create a Good Synthetic Dataset

A good synthetic dataset must be big, have labels, and be parametric. It should include many different design examples to make sure the model learns well. Labels help the machine learning model know the connection between the input and output. A parametric setup allows easy control and changes in the dataset.

  1. The Four Solvers and Their Effect on Design Diversity

There are four solvers in Generative Design: Randomize, Cross-Product, Like-this, and Optimize. Randomize makes random designs, so we can see many different ideas. Cross-Product checks all possible combinations. Like-this makes designs similar to a design we like. Optimize searches for the best designs using the evaluation values. Each solver has a different way to explore the design space.

  1. Insights into Solver Differences and How to Use Them

Randomize and Cross-Product are good for the early stage because they show many different designs. Like-this is good when we want to change a design a little. Optimize is best when we want the highest performance based on goals. Choosing the right solver depends on the project stage and what we want to focus on: many ideas or the best idea.