Empower the architects for analysing the relationship between structure & embodied carbon in early design stages!


Download Plugin

Problem Statement

THE BIG PICTURE

Looking at all the new construction that is projected to take place between now and 2040, we see the critical role embodied carbon plays.

THE MAIN ISSUE


The opportunity to change the embodied material decreases as the project progresses, since each change has a direct impact on the project cost.

Solution

Start Early by empowering the architects to take sustainable decisions!


LearnCarbon is a Rhino plugin that integrates two machine learning models :

  • Model A:

inputs a conceptual massing model by just a click and gathers data on the area, total built-up, structure type and predicts the Global Warming Potential.

  • Model B:

inputs area, total built-up, target GWP value and predicts the suitable structure.

METHODOLOGY

STEP 1 - Cleaning and augmenting the dataset
  • Carbon Leadership Forum - Embodied Carbon Benchmark study
  • Synthetic data based on structural principles (numpy, pandas)
STEP 2 - Training the ML Model
  • Training on Google Collab with Tensorflow and Keras
  • Validating the model
STEP 3 - Link with Grasshopper
  • Calling the ML model to predict for efficient workflow using Hops
STEP 4 - Rhino plug-in
  • Developing a UI in Rhino for ease of use of the architects running on Flask with a GH Hops server (C#, WPF)
STEP 5 - Mapping the workflow
  • Developing a website with detailed documentation on the data developed and analysis (HTML, CSS, JavaScript, jQuery, Bootstrap)
  • Open source project accessible via GitHub and webpage.
Documentation

Demo

Model A

Model B

Benefits

1

Improve the Whole Building Life Cycle Impact significantly, with the structure being the maximum contributor

2

Enabling the roadmap to achieve ambitious certifications like LEED, DGNB, GreenStar etc. for materials & resouces

3

Prediction for stages beyond Cradle-to-Gate in the early stages of the project by Machine Learning

4

Possibility to create an office-specific dataset to adapt & document the office practice


Limitations

  1. Limited Data - LCA quantities against the Bill of Materials & quantities.

  2. Structural Principles - Guidelines for structural layout

  3. Disparity in LCA methods - Calculation methodology, reporting of EPDs and bylaws differ in various regions

  4. Dependencies - The plugin will require specific dependencies to be installed on the client and also a server running on the machine.

Future Work

  1. Integrate structural design & basic layout for the geometry input.

  2. Predicting percentage of material quantities for the structure.

  3. Collect document a dataset for the BOQ and LCA values from participating architectural offices.

  4. Provide a web based platform having similar functionalities to the plugin.

Team

Acknowledgement

  1. McNeel
  2. Carbon Leadership Forum
  3. Open source community for open-source tools (for Machine Learning & Grasshopper plug-ins)
  4. Duncan Horswill, Henning Larsen Architects
  5. Martha Lewis, Henning Larsen Architects
  6. BIGIDEAS, Bjarke Ingels Group
  7. Gabriella Rossi, PhD Fellow, CITA