Achieving an Optimized Solution for Structural

Design of Single-Storey Steel Buildings using

Generative Design Methodology

MFD Project
Team No-4

Team Member

Prakhar Singh (CS21B1042)

Aniket Gupta (CS21B1060)

(1)
Abstract: The significant capabilities of emerging technologies need to be
studied to better understand how they can be used to enhance the
efficiency of the structural design process. Software already used in the
industry are evolving, and some applications are utilizing the power of
machine learning and artificial intelligence. Various companies are starting
to invest in these technologies and are searching for solutions to reduce
component mass, improve structural performance, and minimize
manufacturing process time. Currently, the Steel Centre at the University of
Alberta is researching these technologies' applications towards typical
structural designs. Industry consultation is being conducted to map out
current industry practices and logistics. A literature review of various
optimization algorithms and past studies on the application of generative
design (GD) is being performed. In addition, a single-storey case study is
being conducted that involves developing an automation tool in
Grasshopper that generates warehouse geometry according to user inputs.
S-Frame, an advanced structural analysis software, is being integrated into
the design tool. Wallacei, an evolutionary solver, is being used to input
design objectives and constraints, resulting in optimizing the key
parameters. This automation tool aims to assist in developing a deep
understanding of the possibilities of GD towards structural optimization, and
specifically towards single-storey structures in Canada, which would lead to
the creation of extremely efficient structures. Lastly, the case study
preliminary results are highlighted in this paper along with future
development and research.

1 INTRODUCTION
Single-storey buildings constructed using structural steel are commonly
used in Canada for shopping centres, recreation facilities, and industrial
buildings. During the design development phase of a project, engineers
evaluate multiple design parameters to achieve the owner’s objectives
within a limited time before the detailed design begins. The current single-
storey design workflow in design offices is mostly manual and tedious.
Moreover, miscommunication and human error may occur during design
due to overlapping tasks. Regardless of these disadvantages, the design
process has the capability not only to become automated, but also to seek
innovative new solutions (Rempling et al. 2019, Almusharaf and Elnimeiri
2010). GD is a specific application of artificial intelligence (AI) that can

(2)
quickly generate thousands of high-performing design scenarios (McKnight
2017). It has been applied to architectural designs in Canada (Nagy et al.
2018), but has not yet been realized in structural design. One important
factor that structural engineers prioritize during the design stage in order to
achieve an optimum design is the total weight of the structure. While
optimizing the weight cannot be said to not produce the lowest cost or
optimal structure, it is used in this case study as a proxy for labour cost,
material cost, and environmental effects, etc., caused by the manufacturing
and fabrication processes.

In this paper, an automation design tool is developed using Rhino3D

(Robert McNeel & Associates 1998), a 3D modeling software, and
Grasshopper (Rutten 2007), an algorithmic modelling plugin for Rhino3D.
The design tool takes the user inputs and feeds it to a metaheuristic single-
objective optimization algorithm. The application and performance of
various algorithms used in Grasshopper are studied and considered for
future implementation. In this case study, the algorithm's sole objective is to
reduce the structures steel tonnage. However, other essential objectives
and constraints in the design of steel structures are studied and
considered. For the structural analysis portion of this design tool, a link
between Grasshopper and S-Frame (Casoli 1981) is being developed to
incorporate an FEA solver into the tool. Enhancements to the geometry of
the single-storey structure generated in Rhino3D are proposed. After
discussing the preliminary results of this research, future steps to enhance
the automation tool is outlined.
The main objective of this research is to gain a deep understanding of GD's
possibilities towards structural optimization, which will lead to an
automation tool that can design safer and lighter single-storey structures in
Canada. The method has the potential to reduce material usage, minimize
construction waste, and improve productivity in the Canadian construction
industry. Furthermore, this research has strong potential to provide
Canadian practitioners in the steel construction industry with an automated
process to design single-storey buildings.

(3)
2 OPTIMIZATIONS IN STRUCTURAL DESIGN

2.1 Optimization Algorithms

An evolutionary solver used in this project's optimization tool is responsible
for finding the optimal solution for the design problem by implementing a
metaheuristic single-objective algorithm. Various evolutionary solvers are
discussed in the literature. Most of these algorithms share the same
concepts, as they are developed based on the group behaviour of different
creatures in nature and how they evolve. The main advantage of these
algorithms is that they are derivative-free. Whereas other mathematical
approaches require a well-defined and differentiable objective function and
attempt to find the optimal solution by computing the derivative of the
objective function, metaheuristic algorithms search the domain just by
assessing the objective function's value. Since optimization of engineering
systems requires evaluating sophisticated objective functions that are not
usually differentiable, metaheuristic algorithms have gained popularity
among researchers. The key stages of the optimization process are
illustrated in Figure 1 and summarized below. In this figure, N is the
number of solutions considered for the first generation, and m is the total
number of generations considered for limiting the loop of updating the
generations. It is worth noting that the design variable considered here is
only the spacing between the columns, and the objective function is the
total weight of the structure. The penalty function increases the total weight
of structure when the results obtained from structural analysis software do
not pass the design codes' requirements.
1) Stage 1: the algorithm creates a set of random solutions by varying
design variables associated with the problem. This set is also referred
to as the first generation. Each solution returns a specific value for
the objective function defined for the optimization problem and by
inspecting these values, different solutions can be ranked against
each other.
2) Stage 2: in every optimization problem, solutions are subject to
different constraints with a feasible space defined. The algorithm
applies these constraints to the solutions by a penalty function. If a
solution meets all constraints, the value of its objective function
remains the same. However, if the solution's design variables violate
these constraints, it would get penalized by a multiplier in its objective
(4)
 function so that it would not be able to compete with feasible
solutions of the generation when it comes to rank them based on their
objective function.
3) Stage 3: once the set is sorted based on the objective function of the
solutions, the algorithm applies certain mathematical functions to the
generation and adjusts their design variables, leading to a new set of
solutions (i.e., the next generation). The mathematical functions vary
for different algorithms. For instance, particle swarm optimization
(PSO) algorithm updates the solutions with the velocity function
(Kennedy and Eberhart 1995) and genetic algorithm generates new
solutions with mutation and crossover functions (Goldberg 1989). The
algorithm's main goal is to modify the solutions by generating and
guiding the solutions toward the global optimum of the problem. While
these functions focus on obtaining the best solutions of each
generation and improving them in subsequent generations,
randomness is also formulated within them, which helps the algorithm
search the entire feasible domain of the problem and prevent getting
trapped in zones where local optima are located.

4) Stage 4: it has been proven mathematically that the functions

responsible for generating new solutions help the optimization
algorithm converge at the end if it undergoes a sufficient number of
iterations. There are two ways to specify when an algorithm should
terminate the loop of creating new generations and bypass
performing the second and third stages. The preferred approach is to
consider a total number of generations for the algorithm before it has
stared generating solutions. The second approach involves checking
the convergence at each iteration by comparing the best solutions of
the last two generations with each other. If the difference between the
value of the objective function of these two solutions is less than the
specified tolerance, it is assumed that the algorithm is no longer
capable of finding a better solution, so it is allowed to stop generating
new ones. The latter approach might not be appropriate because
sometimes the algorithm might get stuck around a local optimum.

(5)
 Figure 1: Optimization process stages
2.2 Grasshopper Algorithms
The efficiency of optimization algorithms strongly depends on the number
of variables, constraints, and objective functions. Past studies showed that
Optimus, a tool based on the jEDE algorithm, outperforms several other
single-objective optimization tools of Grasshopper in the optimization of a
frame structure (Cubukcuoglue et al. 2019). The following is a list of the
tools that Optimus was compared against, and the algorithms that they use:
Galapagos (Rutten 2013), based on the genetic algorithm; SilverEye
(Cichocka et al. 2017), using the PSO algorithm; and Opossum (Wortmann
2017), using an RBFOpt algorithm (Costa and Nannicini 2018). In this
study, the performance of Wallacei (Makki and Showkatbakhsh 2018), a
tool

(6)
based on the NSGA-II algorithm, which is primarily developed for solving
multi-objective optimization problems, with the four Grasshopper
optimization tools introduced above to determine which tool yields the best
result for optimizing single-storey buildings. By introducing the new
nondominated sorting concept, the nondominated sorting genetic algorithm
II (NSGA-II) allows us to solve optimization problems with more than one
objective function with the help of fundamental components of the genetic
algorithm, which can only be used for solving single-objective problems
(Deb et al. 2002).
2.3 Wallacei
Wallacei is an evolutionary multi-objective optimization and analytic engine.
This evolutionary solver can consider several objective functions
simultaneously to determine the optimum solution. In this case study, there
is only one objective function, reducing steel tonnage. However, the ability
to run several objective functions is a highly valuable property considering
the automation tool requires more objectives, as mentioned in Section 3.1.
In addition, the solver allows the user to store and save arbitrary data for
each iteration of the design. Compared to other Grasshopper components
such as Galapagos, Wallacei has specific features that give the user better
control over the optimization, graphs, and plots to follow the optimization
(Granberg and Wahlstein 2020). The basic interface for the Wallacei
component in Grasshopper is shown in Figure 2.

Figure 2: Wallacei interface in Grasshopper

(7)
3 PROPOSED AUTOMATION TOOL FOR STRUCTURAL DESIGN
3.1 Description of Automation Tool
GD methodology developed for the purpose of single-storey building design
can generate a large number of layout options according to the designer's
specific requirements. The project's GD workflow has three main
components: generate, evaluate, and evolve (shown in Figure 3). The
designer can specify the length, width, and height of the building inside the
Grasshopper script, shown in Figure 4a. This generates the first design
option in real-time within Rhino 3D, as shown in Figure 4b. The
evolutionary solver Wallacei is then used to produce a large number of
options by varying the equal column spacing used in each direction to
obtain the most cost-effective option, taking into account the weight of
structural steel only, a standard method implemented by fabricators in
approximating the cost. This process leads to various plausible

design options with respective design data, aiding the designer to make a
judgment call on which options to proceed with.

(8)
 Figure 3: Project’s GD workflow

Figure 4: Single-storey structure automation tool

3.2 Structural Analysis Component
The current analysis method in the automation tool mentioned in Section
3.1 uses Excel. After Excel performs a simplified calculation to select the
structure’s members, a summary of the structure's weight is created. From
this summary, the weight of the beams, struts, and bracing is totalled and
represents the total weight of steel for the structure. This total steel weight
is the driving factor for comparing various layouts that the script produces.
However, using Excel is a very simplified method of structural analysis and
needs to be replaced with a more advanced means of analysis.

(9)
3.3 Implementation of Finite Element Method for Structural Analysis
To analyze the structure using the structural analysis program S-Frame, a
link between the scripts in Grasshopper and S-Frame is needed to transfer
the model’s data from Rhino3D to S-Frame. Since there is no current API
that exists to connect Grasshopper and S-Frame, a middleware text file is
required to create this link between software. This text file will export the
necessary information from Grasshopper and import it into S-Frame to
create the structure model. To create such a text file, a template text file is
made that holds all the semi-constant information that can later be filled out
with the remaining data to match the desired model. Filling out the
remaining data can be performed with a C# script that grabs all the needed
geometry data from Grasshopper and inserts it into the text file. Once this
is done, the text file can be opened with S-Frame, creating the S-Frame
model and allowing the FEA solver to analyze the structure. This creation
of this link is almost complete, as the only data left to transfer is the bracing
geometry. The transfer of the model’s data between software is shown in
Figure 5.

Figure 5: Transfer of model’s data between Rhino3D and S-Frame

(10)
To create the C# script that fills out the needed text file, a custom
Grasshopper component was created and shown in Figure 6. This custom
component takes the user inputs such as the number of bays in each
direction, the wall height, and bay spacing, which are used to generate all
the remaining data required for the text file. This data consists of design
codes, geometry, loads, section and material properties, etc. This is
intended to create a fully-parametric tool that can update the text file for S-
Frame as soon as one of the model’s parameters changes inside
Grasshopper.

Figure 6: Custom Grasshopper component created from C# script

(11)
4 PRELIMINARY RESULTS AND FUTURE RESEARCH
4.1 Cost Estimate
In practice, designers estimate the steel structure cost in the design
development stage based on steel tonnage. The steel tonnage can be
considered a key metric to evaluate and compare design options. However,
the total weight of steel can only provide an approximate estimate of the
project's total cost. Other important variables are also required to improve
the estimate’s accuracy, since the total construction cost of structural steel
framing is not necessarily a function of its weight (Ashworth and Skitmore
1983). The three primary components of the total cost are a function of the
material, fabrication, and erection costs, as shown in Figure 7 (Barg et al.
2018). The material category is defined as structural shapes, plates, steel
joists, steel deck, bolting products, welding products, painting products, and
any other products purchased and incorporated into the project. The
fabrication category includes the detailing and fabrication labour required to
prepare and assemble the shop assemblies of structural shapes, plates,
bolts, welds, and other materials. The erection category includes the
erection labour needed to unload, lift, place, and connect the structural
steel frame components. Lastly, other costs are defined as all cost items
not specifically included in the three previous categories (Carter and
Schlafly 2008). As shown in Figure 7, the material costs only constitute
one-quarter of the total cost, and the majority of costs are associated with
the fabrication and erection of structural steel framing. From the results
shown in the mentioned studies, it is evident that the proposed automation
tool needs to incorporate other cost estimating factors to provide a more
realistic estimate of the project’s total cost.

(12)
 Figure 7: Distribution of the total cost of structural steel framing
(Carter and Schlafly 2008)
4.2 Building Geometry
The current lateral load resisting system of the single-storey building
consists of cross-bracing placed at the corners. However, other bracing
configurations such as chevron bracing, V-bracing or single diagonal
bracing can also be used in such low-rise buildings. The possibility of
different bracing configurations (Figure 8), number of braced frames and
bracing locations will be studied in the future. Note that the application of X-
bracing, inverted V-bracing, and diagonal bracing options for a simple
portal frame structure were evaluated in the past using a python script built
in Grasshopper (Vasilev 2020). Furthermore, other steel frame geometric
parameters such as height to eaves, the pitch of frame, and haunch length
are so far missing in the design. Other research has included such
parameters in optimizing steel frame buildings (Phan et al. 2013,
Hernández et al. 2012). Adding different bracing configurations and other
frame geometric parameters will widen the design space's scope and
create a more realistic structure.

(13)
Figure 8: Typical bracing configurations for portal frames (“Portal Frames”
2021)

(14)
4.3 Future Direction
While the current study shows promising results obtained using the
proposed automation tool, it is still rudimentary and requires further
refinement and development before it could be implemented in practice.
Further improvements to the proposed automation tool are as follows:
• Incorporate finite element analysis into the project's workflow by
completing the link between Grasshopper and S-Frame.
• Incorporate additional factors in the overall cost estimation process to
obtain an accurate value.
• Consult with fabricators and erectors to better understand their
preferences in the construction of single-storey steel buildings and
implement this industry knowledge into the optimization process.
• Expand the scope of the generated frame geometry to produce a
more realistic structure and improve optimization.
• Implement multi-objective optimization algorithms by adding more
design variables and objectives.
• Explore and compare different optimization algorithms used in
structural applications and determine if better results can be obtained.
• Combine a connection design optimization tool with the current
automation tool.

5 CONCLUSIONS
An automated optimization tool is proposed here for the structural design of
single-storey steel structures by reducing steel tonnage in the design stage.
The preliminary results show the evolutionary algorithm adopted can result
in optimized design options in the design development stage to help the
designer select an efficient, better design option. Future studies will
incorporate other key design objectives including member availability,
connection type, and different frame geometries into the proposed
automated design tool to achieve a more accurate estimate of construction
costs.

(15)
6 REFERENCES

Almusharaf, A. and Elnimeiri, M. 2010. A Performance-based Design

Approach for Early Tall Building Form Development. 5th International
Conference Proceedings of The Arab Society for Computer Aided
Architectural Design, Illinois Institute of Technology, Chicago, IL, USA, 1:
39-50.
Ashworth, A. and Skitmore, M. 1983. The Effectiveness of Estimating
in the Construction Industry, Master’s Degree, The Charted Institute
of Building, Englemere, Berkshire, England.
Barg, S., Flager, F., and Fischer, M. 2018. An Analytical Method to
Estimate the Total Installed Cost of Structural Steel Building Frames
During Early Design. Journal of Building Engineering, 15: 41-50.
Carter, C., and Schlafly, T. 2008. “$ave More Money”. Modern Steel
Construction, American Institute of Steel Construction.
Casoli, G. S-Frame. V. Enterprise. S-Frame Software Inc. Windows. 1998
Cichocka, J.M.; Migalska, A.; Browne, W.N.; Rodriguez, E. SILVEREYE—
The Implementation of Particle Swarm Optimization Algorithm in a
Design Optimization Tool. In Proceedings of the International
Conference on Computer-Aided Architectural Design Futures, Istanbul,
Turkey, 10–14 July 2017; pp. 151–169.
Cichocka J.M., Migalska A., Browne W.N., Rodriguez E. (2017)
SILVEREYE – The Implementation of Particle Swarm Optimization
Algorithm in a Design Optimization Tool. In: Çağdaş G., Özkar M., Gül
L., Gürer E. (eds) Computer-Aided Architectural Design. Future
Trajectories. CAADFutures 2017. Communications in Computer and
Information Science, vol 724. Springer, Singapore

(16)
Costa, A., Nannicini, G. 2018. RBFOpt: An Open-Source Library for Black-
box Optimization With Costly
Function Evaluations. Mathematical Programming Computation, 10(4):
597–629
Cubukcuoglu, C., Ekici, B., Tasgetiren, M.F., and Sariyildiz, S. 2019.
OPTIMUS: Self-Adaptive Differential Evolution with Ensemble of Mutation
Strategies for Grasshopper Algorithmic Modeling. Algorithms, 12(7): 141.
Deb, K., Pratap, A., Agarwal, S. and Meyarivan, T. 2002. A fast and
elitist multi-objective genetic algorithm: NSGA-II. IEEE
Transactions on Evolutionary Computation, 6(2): 182-197.
Goldberg, D.E. 1989. Genetic Algorithms in Search, Optimization and
Machine Learning. 1st. ed., Addison-Wesley Longman Publishing
Co., Boston, MA, USA.
Granberg, A. and Wahlstein, J. 2020. Parametric Design and
Optimization of Pipe Bridges, Master’s Degree, KTH Royal Institute
of Technology
Hernández, S., Brebbia, C.A., and De Wilde, W.P. 2012.
Computer Aided Optimum Design in Engineering XII, WIT Press,
Southampton, UK.
Kennedy, J. and Eberhart, R. 1995. Particle swarm optimization.
Proceedings of ICNN’95 - International Conference on Neural Networks,
IEEE, Perth, WA, Australia, 4: 1942-1948
Makki, M. and Showkatbakhsh M. 2018. Wallacei. Wallacei.
McKnight, M. 2017. Generative Design: What It Is? How Is It Being Used?
Why It’s A Game Changer. The International Conference on Design and
Technology, KnE Engineering, Geelong, Australia, 2: 176-181.
Microsoft. C#. V. 7.3. Microsoft. Windows. 2000.
Nagy, D., Villaggi, V., and Benjamin, D. 2018. Generative Urban Design:
Integrating Financial and Energy
Goals for Automated Neighborhood Layout. SpringSim: Spring Simulation
Multiconference, Society of
Computer Simulation International, San Diego, CA, USA, 25: 1-8.

(17)