Site Choice

Our site selection intentionally deals with different volumetric constraints in order to see how neural network will perform in the creation of acoustic forms in unusual or difficult spaces.

 

New York City

Chicago

211 E 26th St

New York, NY

320 E North Water St

Chicago, IL

 

Site Dimension

Small Site

Large Site

Site Dimension: 

100 ft (L) x 49.5 ft (W) x 47.5 (H)

30.5 m (L) x  15 m (W) x 14.5 m (H)

Volume:

233,665 cubic foot

6,616 cubic meter

Site Dimension: 

328 ft (L) x 147.6 ft (W) x 98.4 (H)

68.6 m (L) x  45 m (W) x 30 m (H)

Volume:

3,272,080 cubic foot

92,655 cubic meter

 

Input for Graph Convolutional Neural Network

You will need the following three inputs to generate the outcome concert hall. Here is how you do it.

+

+

Input Volume

Input site constrain parameter, or any original volume.

Acoustic Parameter

Seat number, Volume, Reverberation Time, etc.

Silhouette

Input the silhouette image of your wanted concert hall.

 

Result Matrix

44 selected result models from the Graph CNN. 22 of which are generated from the small site, and the other half from the large site. We vary the input mesh shape, mesh counts, acoustic parameters, etc.

Small Site

Large Site

Simulation Matrix

 

44 acoustic simulation results from Ecotect. The performance of sound waves is visualized through simulated particles. A range of results can be seen demonstrating how sound travels within our forms.

 

Parameter Diagram

14 influencing factor for our Graph CNN. This diagram shows different influences and experiment results.

Section Style Transfer

 

Section drawings generated through another neural network: Style Transfer. Using existing concert hall section drawings to project an additional layer of information onto sections drawn from 4 selected models generated by our Graph CNN.

section_4.jpg
 

Acoustic Simulation

Selected examples of acoustic simulation.

S - 2

Input Values -  Seats: 719  Volume: 6615  RT: 1.91  Sa: 620  H: 24  W: 15  L/W: 2.03  H/W: 1.60  V/N: 9.2  a: 0.9

Output Values -  Seats: ​167  Volume: 1212  RT: 1.97  Sa: 110  H: 9  W: 12.6  L/W: 2.13  H/W: 0.71  V/N: 7.25  a: 0.9

S - 12

Input Values -  Seats: 416  Volume: 3740  RT: 1.9  Sa: 352  H: 6.2  W: 10.4  L/W: 0.5  H/W: 0.6  V/N: 9  a: 0.9

Output Values -  Seats: ​171  Volume: 1542  RT: 1.56  Sa: 177  H: 15.6  W: 13.3  L/W: 2.21  H/W: 1.17  V/N: 9  a: 0.9

L - 2

Input Values -  Seats: 2667  Volume: 24000  RT: 2.08  Sa: 2066  H: 30  W: 44  L/W: 1.48  H/W: 0.68  V/N: 9  a: 0.9

Output Values -  Seats: ​2433  Volume: 19351  RT: 2.16  Sa: 1600  H: 41.9  W: 43.5  L/W: 1.51  H/W: 0.96  V/N: 7.95  a: 0.9

L - 21

Input Values -  Seats: 650  Volume: 5500  RT: 1.97  Sa: 500  H: 24  W: 13  L/W: 2.23  H/W: 1.85  V/N: 8.46  a: 0.9

Output Values -  Seats: 247  Volume: 1940  RT: 1.93  Sa: 180  H: 19.3  W: 19  L/W: 1.58  H/W: 1.02  V/N: 7.09  a: 0.9

 

Interior Rendering

A sense of the interior space within mesh models generated by the Graph CNN.

s1-f.jpg
S-6 balcony.jpg

 © 2020 by Taubman College, Thesis Team

  • Grey Facebook Icon
  • Grey Instagram Icon
  • Grey Vimeo Icon
L_22