Imaginative Canopy
Haotong Xia, Jiaying Qu, Lichong Tong, Ningyuan Deng, Shikang Ding| Generative Design | Fall 2021 | GSAPP
Introduction
Under the influence of the epidemic, Columbia University students and faculties generally chose to have meals outdoors to reduce the risk of contracting COVID-19. For the safety and health of the community, Columbia University is opening some outdoor spaces on campus for use without reservation. Sheds were built on these outdoor open spaces to provide shade from the sun and rain for the users. However, these sheds are too large and located at some of the most important traffic nodes on campus, causing disruptions to the paths that people walk on a daily basis. In addition, when entering the interior of the tent, it is obvious that the space is not sufficiently lighted. At the same time, the overly large round table and the casual arrangement make the space underutilized and cannot meet the dining needs of a large number of people.
Considering the existing condition, for this design, we chose the shed in front of the Alma Mater statue which is located on one of the most important traffic spaces on campus, the Low Plaza, as the object of transformation. Our goal was to redesign the canopy using generative design ideas to reduce disruption to the flow of people, increase light, and improve interior space utilization.
Methodology
Design Space Model
To obtain more reasonable design results, we first applied Rhino and grasshopper to build a parametric model. Unlike the traditional design approach, we use genetic algorithms to automatically search for high performance solutions in our design model. And use Discover, a grasshopper plugin, to evaluate the search process.
The design process is divided into two stages. In the first stage, the table space and light/circulation space are divided based on the flow of the people, and the total walking distance, usage area and light area are used as optimization targets to screen the results. In the second stage, the table and chair arrangement is further screened for the plan resulting from the first stage screening, and the one with the largest number of tables is derived. The final canopy form and table and chair arrangement planes are obtained.
Step 1: Space Division Optimization
- Random Flow Simulation
First we identified several major pedestrian flow entrances around the base range on a plane, with points and points connecting each other to represent the paths of the crowd. Since the crowd flow of different paths is in a state of change, a random flow value is given to each path. The number of lines is used to represent the crowd flow for that path, and the path with more people flow has a higher number of lines superimposed on it.
- Paths and Wool Experiment
The wool experiment was first carried out by Frei Otto, who immersed dry and slack wool in water and raised it slowly. The wet wool was held together by the tension between them. The simulation model finds the minimal path system. Each distributed point is reachable, but there are considerable forced detours between some pairs of points. The system is a branching system without any redundant connections.
In the design, we optimize the paths in the model by simulating the effect of the wool experiment with the grasshopper plugin Kangaroo. Adjacent paths are attracted to each other under certain constraints and given attractive forces, and the original straight paths are bent and partially overlapped to form a new surface. The total length of the optimal paths will be shorter than the total length of the initial paths, effectively integrating the flow and improving the overall efficiency of crowd circulation.
- Space Division Evaluation
We add up all the new path lengths processed by the wool algorithm and use the shortest sum of path lengths as the first objective value. The new paths divide the whole surface into several areas. We consider the circulation space as the main lighting area, and the area which is cut by the path is the space where the tables and chairs are placed. Therefore, in order to optimize the lighting conditions under the roof, we calculated the distance from the center point of each area to the nearest lighting point, took the average, and set the smallest lighting distance as the second optimization target. Meanwhile, the maximization of the total area of the table and chair areas was taken as the third optimization objective. These three variables will be evaluated in Discover and the situation that maximizes light, minimizes paths and uses the space most efficiently will be selected for the table and chair placement.
Step 2: Table Layout Optimization
After filtering the best space division results in Discover, we proceeded to optimize the table layout for the plan. Firstly, we set the size of individual table units, and the radius of one unit is 1.5 meters. It is ideal that the units are tangential to each other without overlap or excessive gaps. Place an array of units covering the base area. Given random numbers to move and rotate the array, and calculate the number of tables for each method separately, and perform a second evaluation in Discover to select the layout with the largest number of tables.
Input Parameters
Based on the needs of the model, two main variables are introduced in the model for calculation.
- People Flow in Different Paths (Discover Continuous Input)：
Different flow values are randomly given to each population path at a preselected starting point. By changing this variable to simulate different kinds of possible changes in the flow of people passing through the Low Plaza at different times and under different circumstances.
- Table and Chair Unit Arrangement (Discover Continuous Input)：
Change the position and rotation angle of the table unit array. By randomly varying the position of the table unit array, the variety of input cases is increased and more accurate cell arrangement results can be obtained by multiple calculations.
Performance Metrics
In order to optimize the process of calculation, we achieve the desired metric in two steps. To balance the usage of the space under the tent with the lighting conditions, we defined in the first Discover simulation three main matrices, the minimum lengths of the circulation paths, and the minimum of average lighting distance and the maximum of table areas. In order to make the results presented by Discover clearer and easier to filter, we combine the three objective values into two objectives. A function is used to combine the path length and the average lighting distance, and the minimization of the combined results is entered into Discover as the optimization objective. In the second simulation, maximizing the number of tables will be the most important metric to guide our choice of results.
For the whole selection process, in the first step, in order to balance the optionality of the three metrics and to quantify them at the dipolar level, we artificially integrate complex objectives by setting weights by means of functions, and among the selected results that fit the range, we artificially filter the unquantifiable objectives such as space separation and usage patterns to obtain more feasible results.
Results
By quantifying the performance of each two steps, we create evolutionary processes that allow computer to search through our design systems to find the canopy design and table layout with the least disturbance to the pedestrian path, the best lighting conditions and the highest space utilization.
Conclusions
In generative design it is important to translate design goals into easily quantifiable parameters. In the design, we quantify the efficiency of the overall floor plan as the sum of path lengths, the light level as the distance from the center of the area to the light point, and the space utilization as the number of tables. We also used genetic algorithms to allow the computer to assist us in the evolutionary processes to obtain the best results from a large number of possible outcomes. This systematic approach to tent optimization can also be quickly and easily applied to other similar tent designs.
During the optimization process, we initially synchronized steps 1 and 2 in Discover, but this caused the software to run too slowly. After some adjustments, we split the optimization into two steps, which made the overall running speed smoother. However, by selecting the best results from the first optimization and then filtering them in the second step, the final results obtained may not be as accurate as if they were optimized at the same time. In the subsequent design process, in order to ensure the accuracy of the results, when we optimize in steps, we can select multiple best results for each step, and then perform the second optimization separately.
