Case studies of AI in architecture
Case studies of Al in architecture
Combining Al and BIM in the design and construction
Combining AI and BIM in the design and construction of a Mars habitat
Naveen K. Muthumanickam, Jose P. Duarte, Shadi Nazarian, AH Memari, and Sven G. Bilen
Artificial intelligence (AI) as an academic discipline emerged in the 1950s. Over the years, it has been defined in many different but related ways. One particularly useful definition was put forth by Patrick Winston in his famous textbook Artificial Intelligence (Winston, 1992, p. 5), in which he states that “Artificial intelligence is the study of the computations that make it possible to perceive, reason, and act,” with the Turing Test used to verity whether a machine is capable of intelligent behavior (Turing, 1948). Since its inception, Al has encompassed different sub-fields and methods including semantic nets, search and optimization, rule-based systems, frames, constraints propagation, backtracking, planning, image and natural language, and learning. In recent years, “learning” has been referred to as “machine learning (ML)” and equated with Al itself, particularly among the general public. The application of AI to design has been present since the beginning, particularly because solving design problems poses interesting and complex challenges that go beyond simple calculations and require a more elaborate kind of intelligence. The application of AI to architecture had important developments in the 1960s with the work of researchers like Ivan Sutherland, who developed the first computer aided design (CAD) system called “Sketchpad” (Sutherland, 1963), and Herbert Simon’s influential book The Sciences of the Artificial (Simon, 1969). In the decades that followed, several authors used different AI techniques and paradigms to address architectural problems. One meaningful reference for the work described in this chapter was Jose Duarte’s discursive grammar tor Alvaro Siza’s Malagueira houses (Duarte, 2001). In his work, he used shape grammars (Stiny & Gips, 1972), a rule-based system, to encode Siza’s design rules and define the space of design solutions, a series of metrics to compare the evolving solution to the desired one, and best-first search to guide the generation of solutions towards the desired goal. The outcome was customized housing solutions in Siza’s style, who was unable to distinguish solutions generated by the system from his own designs, thereby validating the “intelligence” of the system through this modified version of the Turing Test.
The current work also aims to find solutions that best fit predefined conditions and uses a similar conceptual framework, encompassing generation, simulation, and optimization, as described in Duarte (2019). Its implementation has, nevertheless, some noteworthy differences: a parametric design system replaced shape grammars as the basis for the generative system; sophisticated software simulated the performance of candidate solutions from different viewpoints, including structural and environmental; and multicriteria optimization was used to find solutions that fitted different scenarios. ML was used in the optimizer to look at information about previous searches and utilize it to speed up the search process. Another important difference is that a new system incorporated information about the process used to materialize the solutions, which was also considered when searching for appropriate designs. In addition, the new system was developed on a building information model (BIM) platform, which permitted the control ot the various software used for the generation, simulation, and search operations and support the progressive development of the design. This platform was used not just to design the habitat, but also the process tor making it, including the printing system, its transportation, and set up. The use of such a platform was necessary to solve the complexities involved in the multifaceted design problem, which included the consideration ot two completely different locations, Mars and Earth, with very different environmental conditions with significant impact on design performance and the autonomous robotic construction process.
Robotic construction is beginning to make inroads into the architecture, engineering, and construction (AEC) field to assist in complex and repetitive tasks and to support construction in remote and challenging environments. Before deployment to such places, it is imperative to simulate the sequence of robotic movements and the subsequent behavior of material delivery and construction logistics to ensure accurate and sate realization of the building design. Robotic construction techniques such as additive concrete construction, in particular, are highly dependent on material-specific (concrete flowability, setting time, etc.) and robot-specific properties (robotic arm speed, toolpath direction, etc.). Simulating the construction process while also considering these properties early in the design phase is essential as they play a crucial role in determining the constructability of the building shape. The resulting building shape, in turn, impacts other performance factors such as structural integrity, indoor environmental quality, and construction cost. Hence, in addition to simulating the construction sequence, it is also essential to analyze the building design for these additional performance factors. For instance, Figure 13.1 illustrates an industrial robot used to 3D print a simple concrete cylinder by depositing material in a concentric circular tool- path layer by layer. Here, as time progresses from /, it can be noted that the cylinder deposited by the robot behaves nominally until t + 5 min, after which it starts failing. BIM-based techniques such as a digital twin (close-to-accurate digital representation of the robotic setup) can be used to simulate the construction sequence (i.e., a “4D” simulation). However, additional techniques like structural analysis using finite element modeling (FEM) tools and physical tests are needed to predict such unforeseen failures beforehand and address them in the design stages. More complex geometrical shapes at larger scales, such as that of a building, need to be analyzed and simulated for additional factors such as indoor environmental quality (IEQ) and building cost. Additionally, it is also possible to utilize advanced computer vision sensors to capture real-time data on structural deformation during the concrete 3D-printing process and adjust material- and robot-specific variables accordingly to rectify or correct such deformations. However, this requires sophisticated use of ML algorithms for predictive analytics and robotic process automation for timely implementation of the corrections. The former method, in which rigorous multidisciplinary analyses and simulations are used to ensure realization of the envisioned building design, can be considered analogous to the processes involved in the design of products in other engineering fields such as the automotive and airplane design and manufacturing sectors. On the other hand, the method in which real-time data collection, predictive analytics, and ML are used to prevent structural deformations can be considered analogous to the autonomous systems used in the landing of a reusable space
Figure 13.1 Conceptual overview of a failing printing test.
vehicle, in which the system automatically adjusts the various flight parameters to ensure the spent rocket follows a nominal landing trajectory.
This chapter focuses on the former approach, i.e., leveraging BIM as a connected software ecosystem for parametric modeling and generation of large sets of design alternatives, integrated multidisciplinary analysis, optimization, and construction simulation in a robotic construction project (specifically, additive concrete construction). An overview of the process underlying design for additive manufacturing (Df) along with challenges and technological knowledge gaps associated in implementing a similar framework to enable design for additive construction (DfAC) in the AEC field are discussed in the next section. This is followed by a section outlining the features of a novel BIM-based framework, which was developed and used for the design of a 3D-printed Mars habitat satisfying multidisciplinary requirements. Additionally, a brief overview of how computer vision techniques and ML could be leveraged for avoiding structural deformations and failure in robotic construction is also discussed along with potential avenues for future research.
Design for Additive Manufacturing (DfAM)
Additive manufacturing (AM) technologies have grown tremendously in the last two decades, enabling new strategies for rapid prototyping of designs and custom production within the manufacturing sector. The capability of AM technologies to place, bond, or transform raw materials layer-by-layer or voxel-by-voxel (elemental points in 3D space) enable the rapid production of a wide range of physical objects ranging from small artifacts to complex industrial components such as automobile and electronic parts. A wide range of AM processes such as powder bed fusion, material extrusion, and sheet lamination are used in the production of such parts and components (cfl, Thompson et al., 2016). With such wide ranging AM processes, it is crucial to note that the quality of the raw material at each voxel depends on multiple factors specific to raw material(s) type, manufacturing equipment (e.g., the build platform precision, nozzle geometry, and light or laser beam wavelength), and process parameters (e.g., the nozzle temperature, light or beam intensity, and traverse speed) (Thompson et ah, 2016). These variables affect the performance, quality, and cost of production of the parts and, in turn, impact design decisions. Therefore, it is necessary to analyze the design of the product for various performance factors with respect to these variables. Motivated by these, architecture and engineering design fields have worked to develop appropriate DfAM frameworks under the larger body ot work focusing on design for manufacturing and assembly (DfMA).
Typically, DfMA frameworks (generalized process shown in Figure 13.2) enable designers and engineers to design and optimize a product for various performance factors while concurrently considering its manufacturing constraints (imposed by the production system) in order to reduce production time and cost while increasing their performance and quality. Most of the product design processes (Figure 13.2) start with the generation ot 2D sketches and 3D models of a few conceptual ideas envisioned by the designers. Subsequently, these models are then iterated at an appropriate level ot detail and subject to multidisciplinary analyses specific to the product type and intended use case. Based on such multidisciplinary analyses and inputs from multiple stakeholders, the generated design alternatives are optimized to meet multiple performance factors, including manufacturing constraints ot the specified production system(s), while optimal design(s) are selected based on trade studies (trade-off exploration). In order to ensure efficient manufacturing and production of the product, sophisticated toolpath (robotic or 3D-printer nozzle motion) simulations are used.
Considering the more nuanced outline ot the process underlying DfAM shown in Figure 13.3, the process begins with the generation of a 3D model, which is then sliced/ contoured (typically along the vertical axis) with each layer containing a toolpath (route map) along which the nozzle traverses. This is accomplished using a variety of specialized slicing software as listed in Figure 13.3. Based on the geometry of the part and the type of
Figure 13.2 Design for manufacturing and assembly (DfMA) framework. Adapted from www. autodesk.com/product/fusion-360.
Figure 13.3 Design for additive manufacturing (DfAM) framework.
Figure 13.4 (Left) In-house prototype of offline robotic programming software (Credits: Naveen); (Right) Digital twin of two robots performing a synchronous welding operation (Can et al., 2013).
AM process used (e.g., powder bed fusion or material extrusion), the slicing software also automatically generates structures to act as temporary supports for cantilevered elements to be printed. These are usually printed at a lower resolution (i.e., much coarser) as they are removed after the printing. Further, these contours, toolpaths per contour, and transitions between each contour have to be converted into a machine-readable format called G-code. Additionally, a machine-code (M-code) is also generated, with which the manufacturing equipment is issued specific instructions such as the velocity of the nozzle, deposition temperature, material deposition cut-off times, bed positions, and robot arm positions. Both the G-code and the M-code together act as a machine-readable set of instructions for the manufacturing equipment (or a robot in terms of large-scale industrial manufacturing). When large-scale industrial robots are used for manufacturing, these codes are then subject to an inverse kinematic solver (or robotic motion solver), which simulates the robotic motions per the instructions to identify any self-collisions and singularities (a state in which the robot cannot execute a specified motion as it violates the degree of freedom of one or more of its axes). When the production system comprises multiple robots and associated manufacturing equipment, state-of-the-art industrial practices use offline robotic programming to synchronously simulate the sequence of robotic motions. For example, Figure 13.4 shows a sample of offline robot programming software in which a model is used to simulate the toolpath of an industrial robot executing a specific type of manufacturing or production activity.
Design for Additive Construction (DfAC)
Types of additive construction systems
The state-of-the-art developments in additive construction technologies leverage those in materials research and printing systems research. On the materials front, a growing body of work can be found on concrete 3D printing (additive concrete construction), clay 3D printing, and thermoplastic-based geo-polymer 3D printing, among other related developments (Rael & San Fratello, 2011). On the printing systems front, a variety of printing systems have been developed such as gantry, cable-suspended, small scale swarm robots, industrial robots, and foldable (in which 3D structures are designed as foldable 2D shell structures) (Labonnote et al., 2016) (Figure 13.5). The scope of this chapter is limited to additive concrete construction using industrial robots.
Figure 13.5 Types of additive construction systems. Adapted from Labonnote et al. (2016).
Figure 13.6 Conceptual framework for design for additive concrete construction (DfACC).
As a subset of the broader DfAM framework, a framework for design for additive concrete construction (DfACC) can be outlined (Figure 13.6). The process starts with the generation of a 3D model, followed by slicing the geometry and translating the contours, toolpath per contour, and transition between contours into a machine-readable code, which in turn is subject to an inverse kinematic solver for robot motion planning. However, the crucial difference here is with the material in use (concrete) and the slicing procedure. The material properties and behavior of concrete are starkly different from the materials traditionally used in AM such as polymers and metals. Such a scenario warrants specific treatment in the slicing of the geometry to be additively constructed to enable close-to-accurate realization of the envisioned geometry. For example, the diamond shape in Figure 13.6 has tapered (partially cantilevered) facets near the bottom of the geometry. Such tapered facets can be 3D printed successfully in the case of plastic or metal extrusions with additional temporary support structures generated during the slicing procedure. However, the same might not work tor a concrete structure and might lead to potential failure of the part during printing (far right illustration in Figure 13.6 (Sce- nariol)—the left overhang exceeds the maximum inclination angle of a structure that permits printing without supports). Another instance that might cause similar type of failure in additive concrete construction is a scenario in which the interior portion of the diamond-shaped geometry might be hollow (Scenario 2 as indicated in Figure 13.6). To address such challenges, most of the additive concrete construction techniques developed to date discretize the overall structure into a kit of parts that can be printed in a convenient orientation to minimize the effects of gravity; allow the printed part to cure (hardening and setting of concrete); and then reorient them during assembly (van der Zee & Marijnissen, 2017). The scope of this study is to explore and demonstrate additive concrete construction techniques developed by a multidisciplinary team at Penn State as part of NASA’s 3D-Printed Mars Habitat Centennial Challenge, which enables architectural scale concrete structures that do not require support structures or reorientation and assembly of printed components.
Another important factor is the slicing procedure to develop the toolpath for additive concrete construction. Apart from the contours in the vertical direction, there are multiple possibilities to devise the toolpath per layer. Typically, in AM, the toolpath is divided into an outer wall (perimeter ot the printed geometry) and an infill, which is the portion surrounded by the outer walls (Figure 13.7). The number of outer wall(s), infill pattern, infill pattern density (number of layers in horizontal direction), and bead shape and dimensions (dictated by the nozzle shape and size) can be customized pertaining to the design specifications. Beads are also represented as filaments elsewhere in the literature. We tend to use the term bead to represent individual concrete filaments due to the circular-shaped nozzle (Right side illustration in Figure 13.7) we use in our robotic setup.
To better understand the impact of these toolpath design variables on the overall performance of the 3D-printed geometry, let us assume the case ot a simple concrete slab as an example (Figure 13.8). Flere, as usual, the design process starts with a 3D model, which is then subject to the slicing procedure. As shown in Figure 13.8, two types of toolpath—A (continuous spiraling) and В (alternating zigzag)—were generated. Once these toolpaths were converted into machine-readable codes, a robotic setup was used to print the concrete slabs using both toolpath A and B. Based on physical testing, it was identified that a concrete slab
Figure 13.7 (Left) Toolpath simulation of a simple slab; (Right) Sectional view of concrete beads deforming under compressive load.
Figure 13.8 Tradeoffs in designing toolpath.
printed with toolpath В had better structural performance in terms of layer-to-layer adhesion than toolpath A (Ashrafi et al., 2019). However, it should also be noted that, due to the zigzag nature ot the toolpath, the robot had to perform complex maneuvers, which resulted in vibrations at each turn of the toolpath. Such vibrations, in turn, led to concrete deformations and hence a loss in shape accuracy. Trying to reduce the velocity of the robotic arm would be a straightforward answer to solve this issue. However, doing so affected the consistency ot the concrete deposition (flowability), which caused concrete clogging at the nozzle head. Also, reducing the velocity increases the overall print time. Hence, there exists multiple such tradeoffs between various aspects ot the toolpath design, which, in turn, affects the overall design performance. Moreover, it is essential to note that, as the scale and complexity of the geometry increases, the toolpath complexity also increases, thereby leading to multiple such tradeoffs.
DESIGN OF MATERIAL PROPERTIES
In addition to the toolpath design, it is equally important to develop concrete mixture(s) with the appropriate composition (cementitious material, aggregates, binders, and additives) to ensure successful 3D printing of concrete structures. While developing the material composition, it is essential to consider multiple rheological aspects such as flowability and setting time, among many other factors. In simple terms, a particular type of concrete mix might be plastic and malleable when wet, ensuring that the concrete mix does not harden within the delivery hose causing a clog. However, the same mix might take a longer time to harden once printed resulting in inadequate lower layers (beads) that deform under the weight of the upper layers. This potentially might result in anomalies in the flatness of the surface and, in turn, affects the shape accuracy. Apart from flowability and setting time, concrete mix can also be modified to fit other needs such as structural, thermal, and aesthetic properties ot the additively constructed part. For example, functionally graded concrete, in which a specific percentage of aggregates in the concrete is replaced with materials such as cork granules can be used to make the 3D-printed concrete structure lightweight and have better thermal insulation properties. However, modifications to the material composition might also have adverse effects on other performance aspects. For example, toolpath A in Figure 13.8 had better shape accuracy when printed with a normal concrete mix but showed inconsistent results in terms of flatness of surface and shape accuracy when tested using initial trials of a functionally graded concrete mix (Figure 13.9). The functionally graded concrete—cork composition was perfected after many such trial 3D-printing tests and is covered in detail in Craveiro et al. (2017).
Apart from thermal insulation and structural performance, functionally graded concrete mixtures might also be used for aesthetic and architectural purposes. For example, functionally graded concrete mix with silica and other similar aggregate replacements have been explored to create a seamless transition from concrete to glass without the use of any
Figure 13.9 Tradeoffs in designing the material composition.
Figure 13.10 (Left) Conceptual render of shelter 3D printed using functionally graded material; (Right) Sample showing a block of functionally graded concrete seamlessly transitioning from 100% geopolymer concrete to 100% glass.
mechanical frame or bonding agent (Nazarian et al., 2015), and later incorporated into the design proposal developed by the Penn State team in the virtual design level of the final phase in NASA’s 3D-Printed Mars Habitat Centennial Challenge Competition (Figure 13.10, Craveiro et al., 2020). However, this chapter does not delve into the details of functionally graded concrete—glass mixture.
DESIGN OF PRODUCTION SYSTEM AND LOGISTICS
Additive concrete construction is enabled through the synchronous sequence of activities of multiple pieces of equipment such as the material mixing, feeding, and deposition systems (Figure 13.11). Specifically, these systems include silos for raw material storage, concrete mixers, industrial robot(s), and robotic end effectors (nozzle, grippers, etc.), which are fitted to the end of the robot arm(s) to perform activities such as extruding material, picking and placing objects, feeding computer commands to the robot, and holding sensors for real-time feedback. Such equipment and systems, which must consider the conditions of the physical site on which they operate, impose certain production and logistical constraints. It is essential to consider these constraints during the design of the building to ensure its constructability. Particular to the 6-axis industrial robot (ABB IRB 6640) that was used by our team for NASA’s 3D-Printed Mars Habitat Centennial Challenge, the robot’s motion is constrained by the degrees of freedom of each axis of rotation and also has a limit on the maximum distance it can reach (2.8 m). In simple terms, an imaginary sphere of ~2.8-m radius is the maximum range that the robot arm can reach, limiting the size of the structure that can be additively constructed. Such constraints warrant the need for designing and installing end effectors (or extensions), which help to extend the capability of robots to exceed their limited reach and, in turn, allow the additive construction of larger structures (Watson et al., 2019). Additionally, raw materials (cement, water) are stored in silos and tanks and are connected to a mixer using hoses. Further, another hose delivers mixed concrete paste from the mixer to the nozzle attached to the robot end effector. It is essential to highlight the fact that even minor aspects such as the length of the hose connecting the material mixer to the robot end effector needs to be decided strategically in order to avoid any unnecessary slack in the hose, which might interfere with the toolpath and, in turn, affect the printing process. Hence, the positioning of these robots and other equipment requires strategic spatial and logistical planning and optimization in order to ensure efficient site preparation tor safe printing of the envisioned structure without any hindrance.
Figure 13.11 Rendered representation of the robotic construction setup: (1) Truck, (2) Conestoga trailer, (3) Water tank, (4) Large silo, (5) Small silo, (6) Mixer and pump, (7) Computers, (8) Robot controllers, (9) Safety fence, (10) Safety laser, (11) Opening-placement robot, (12) Printing robot, (13) Hose, (14) Nozzle, (15) Printing area, and (16) Scissor jack for monitoring.
DESIGN OF OVERALL BUILDING
Apart from the design of toolpath, material, production, and logistics systems involved in additively constructing the enclosure of the building (floor plates, walls, and roofs), the architectural design of the interior spatial layout and associated mechanical, electrical, and plumbing (МЕР) and safety systems adds another layer of complexity to the design. This impacts many aspects of the building performance such as overall structural integrity, indoor spatial quality (accessibility, functionality, and aesthetics), indoor environmental quality (1EQ) (heating, cooling, ventilation, and lighting), and sewage and sanitation. It also plays a crucial role in dictating the overall geometry of the building, thereby impacting the sequence of construction (and assembly), construction time, and cost. Traditionally, in building design and construction projects, construction simulation is done only at later stages when the design is much more detailed and specifically for construction management purposes. Conversely, DfACC needs designers and engineers to account for constructability (in addition to other performance factors) from the early design stages (Figure 13.12). Early stage construction simulations are necessary to identify any potential robot—structure or robot—robot (if multiple robots are used) collisions downstream during the construction process, which might lead to costly revisions. Additionally, there might be tradeoffs in deciding the geometry that satisfies multiple performance requirements. For example, a particular design geometry that has an optimal collision-tree robotic motion plan might have poor compressive and lateral load distribution or vice versa. Digital twins or close-to-accurate digital models of the setup are necessary for efficient and meaningful 4D simulations.
Figure 13.12 Comparison highlighting differences in design process for traditional and robotic construction.
Further, dealing with such a wide array of interconnected systems introduces multiple variations in how these systems can be configured with respect to the overall geometry of the building and vice versa. In such scenarios, it is essential to explore multiple design alternatives (or configurations) to make well informed design decisions. Technically, this requires generating/modeling multiple design configurations and analyzing them for multiple performance factors, such as structural, IEQ, energy, cost, and constructability. However, such concurrent-design-and-analysis approach requires a paradigm shift in the underlying design process as well as technological developments in terms of integrating the software and tools used by multidisciplinary stakeholders (Polit-Casillas & Howe, 2013). Multiple stakeholders dealing with various aspects such as toolpath design, material rheology, architectural systems, structural system, МЕР system, environmental control and life support system (ECLSS), and robotic systems use a variety of software for modeling, analyses, and simulations. Few of these even involve setting up physical testing apparatus (such as compression tests, rheology tests, etc.) and data collection. Based on measurable observations and inferences from such physical tests, material design, toolpath design, or geometrical specifications of the design might need to be modified appropriately. It is vital to incorporate such modifications to an integrated building information model (BIM) to identity the impact of those changes in terms of other performance factors. In such cases, it is essential to develop parametric modeling capabilities, which allow rapid changes to the geometry of the structure based on the observations from these physical tests. State-of-the-art BIM environments support parametric modeling and rapid generation of multiple design options. Additionally, there has been a recent surge in the use of multi-objective optimization and search algorithms tor structural, energy, and daylighting optimization in building design (Attia et al., 2013). However, these multi-objective optimization developments have been largely fragmented from advancements in BIM due to lack of integration between modeling and simulation environments (for multiple types of analyses) (Leicht et al., 2007; Haymaker et al., 2018). There is a need for an integrated BIM framework that can address and streamline the interoperability issues between multiple modeling, analysis, optimization, and construction simulation in the AEC field (Flager & Haymaker, 2009; Muthumanickam et al., 2020c) (Figure 13.13).
Figure 13.13 Multidisciplinary nature of design for additive construction (DfAC) involving a range of computational analyses and physical testing.
To address such technological gaps and streamline the additive construction design process, an end-to-end BIM framework was developed and used to design a Mars habitat from the conceptual design stages to additively constructing it using industrial robots in the final level of Phase 3 of NASA’s 3D-Printed Mars Habitat Centennial Challenge. The next section presents a detailed overview of the various components of the integrated BIM framework for modeling, analysis, optimization, and simulation to support additive construction.
Design for additive construction of Mars habitat—NASA Centennial Challenge
An interdisciplinary team from Penn State participated in NASA’s 3D-Printed Mars Habitat Centennial Challenge, in which multiple teams competed to push the state-of-the-art of additive construction technology to design and build sustainable habitats for humans to live in on Mars. The goal was to design a 3D-printable habitat that provided a pressure-retaining living area of at least 93 m2 with a minimum ceiling height of 2.25 m, with the intent of supporting four astronauts for one year with sleeping, eating/meal preparation, sanitation, recreation, laboratory/work area, communication, as well as МЕР, environmental control and life support systems (ECLSS), safety systems, and entry and exit hatch systems. The competition was hosted in multiple phases that included design (Phase 1), structural member (Phase 2), and on-site construction of a sub-scale habitat (Phase 3) with multiple levels under each phase. Phase 3 included two virtual construction levels and three actual construction levels. The virtual construction levels emphasized leveraging BIM to design the habitat (for simulated Martian conditions) as well as the underlying construction process and logistics involved in its additive construction. The actual construction levels emphasized leveraging autonomous robotic construction methods to additively construct parts of the habitat (in Earth-based conditions) with progressive complexity ranging from foundations at the initial level, construction of a cistern with two required pipes that penetrated its thick wall and performance of a hydrostatic test during the second level, and a 1:3 sub-scale habitat design (a simplified version of the full-scale habitat design submitted for the virtual construction level) during the third level of the competition.
End-to-end BIM framework
Building on the learning from the DtAM frameworks, challenges in streamlining and enhancing the DfAC framework were addressed by focusing on developing an end-to-end BIM framework that supported integrated modeling, analysis, optimization, and simulation (Figure 13.14) (Muthumanickam et al., 2020a, 2020b). A detailed overview of how the BIM framework was leveraged in DfAC of the Mars habitat for the NASA Challenge is outlined in the following section.
As mentioned, the final phase of the NASA Challenge had two levels: a virtual construction level and an actual construction level. The virtual construction level required the design of a shelter that was to be optimized for Martian conditions, but for which a sub-scale version was to be 3D printed on Earth. Based on precedents and preliminary design charettes, the design concept took into account the maximum overhang angle that can be printed without formwork, which is ~60 degrees from horizontal; the minimum wall thickness required to provide protection from harmful cosmic radiation (on Mars), which varies between 2 and 3 feet (60 to 90 cm); and the maximum reach of a robotic arm. The resulting geometry
Figure 13.14 End-to-end BIM framework developed for DfAC of a Mars habitat for the NASA Challenge.
Figure 13.15 Design concept for a conical geometry along with variable configurations (tangential, overlapping, at a distance, and curvilinear arrangement).
included conical-shaped modules of varying sizes that could be connected and combined in different ways to obtain different shelter configurations and adjusted to different programmatic requirements and specific site conditions (Figure 13.15). The modular nature of the habitat guarantees expandability as it permits incremental construction of additional modules to host a growing human community on Mars. The same conical module concept was utilized for both the virtual and actual construction levels.
Each module was designed to cater to specific needs such as living area, working space, kitchen/dining, bed/bath, leisure area, and hydroponic food production module. The design for the virtual construction level had to include entry and exit hatch systems; МЕР systems and ECLSS to enable indoor living and work activities of astronauts; safety systems to mitigate emergency situations; food production systems; and other interior elements (along with foundational structure and envelope) as discussed further below. On the other hand, the design tor the actual construction level had to include only the entry and exit hatch systems, but not the МЕР and safety systems. The services and systems were designed to be centralized for ease of access, assembly, and serviceability.
A procedural modeling algorithm was developed in Dynamo for Revit using node-based modeling in which several nodes execute geometrical manipulation tasks computationally to result in the overall habitat design. The algorithm consisted of various clusters of nodes responsible tor different actions such as controlling input variables (for feeding input values); floor plan generator (for modeling floor plans with required rooms and interior spaces); interior systems modeler (tor modeling МЕР, ECLSS, and other service systems and components); conical module modeler (for modeling the conical enclosure); connector modeler (for solving intersections between adjacent modules); entry/exit and safety system modeler (for generating entry/exit hatches in selected modules); additive construction constraint solver (geometry slicing and toolpath related tasks); and application programming interfaces (APIs) (to connect the BIM model to various structural, environmental, toolpath analysis, and robotic simulation tools) (Figure 13.16). The parametric procedural modeling algorithm enabled rapid generation of multiple design alternatives with variable modular configurations. The algorithm was programmed to generate the enclosure (foundation, structural components, and walls) at level of detail (LOD) 300 & 400, and the МЕР and ECLSS system components at LOD 100 & 200 (for virtual construction level 1 & 2, respectively) (Figure 13.17).
Analyze + optimize
The generated building design options were then tested for multiple performance factors such as structural, IEQ, construction time and cost, etc. using discipline-specific computational analysis software and physical tests. For sound structural performance of a building, it is crucial to ensure that the building can withstand compressive loads caused by gravity
Figure 13.16 Parametric node based algorithm developed in Dynamo for Revit for procedural modeling of the Mars habitat.
Figure 13.17 (Left) Design options with variable module configurations and size and shapes and (Right) LOD 200 МЕР, ECLSS, and interior systems generated by the parametric algorithm.
and vertical loads in the building and lateral loads due to multiple factors such as wind loads, seismic activity, pressure difference between interior and exterior, and so on. Martian conditions such as a reduced gravity (3.71 m/s2, approximately one third that of Earth) and reduced atmospheric pressure (approx. 0.0006 MPa, Earth’s atmospheric pressure is ~ 0.101 MPa, i.e., less than 1% that on Earth) create increased lateral loads on the walls of the building due to increased pressure difference between the interior and exterior of the building. This produces a condition in which buildings with walls at certain angles might fail due to the increased outward lateral load if not designed properly. Contrary to this, the primary reason behind the selection of a conical geometry (angular walls that meet at an apex point) was to enable a unified concrete structure that can be 3D printed continuously from floor to roof without the need for any assembly post print, the use of any prefabricated parts, or any temporary structural support during construction. It should be noted that most of the state-of-the-art concrete 3D printing of buildings involve printing of components as kit of parts and assembly using a gantry or crane mechanism. Our team focused on developing techniques that can enable 3D printing of tapered concrete structures without any formwork
(or supporting structure) to avoid the complexity, cost, and logistics involved in setting up such formwork. The taper angle of the conical extrusion resulted in cantilevering of successive beads on top of each other. Due to the lack ot external formwork, it is essential to design the taper in such a way that the center ot gravity of each successive layer partially overlaps with the layer beneath it as a means to ensure continuous load transfer.
Given such complexities, it is essential to identify a conical structure with a taper angle that is capable of withstanding both compressive and lateral loads as well as ensuring shape accuracy when 3D printed without formwork. Detailed computational FEM was used to aid the selection of the structurally optimal design option (Figure 13.18). However, additional physical testing might be required as design options identified as optimal by the FEM might fail under real-time conditions. For example, computational FEM analysis ot design options with taper angles of 60, 65, and 70 degrees seemed to yield optimal compressive strength.
Upon 3D printing small-scale samples to test the physical performance of these taper angles, it was identified that the 70-degree option performed well, whereas the other two angles (60 and 65 degrees) tailed, as shown in Figure 13.19. However, a 70-degree taper angle of the conical extrusion resulted in an increased height of the overall building design, which would
Figure 13.18 Sample of structural finite element model (FEM) of selective design options (Credits: Keunhyoung Park).
Figure 13.19 Failure of 65-degree taper angle due to material deformation caused by cantilevering of beads.
make it infeasible for the robot to print the top portion of the building without additional end effectors. Moreover, designing and fitting of an end effector to increase the range ot the robot resulted in vibrations near the material delivery nozzle, thereby leading to over-extrusion of materials at some parts of the geometry (Figure 13.20). The increase in vibrations can be attributed to the shift in center of gravity of the end effector among other reasons and a lower stiffness when compared to the robot arm itself. Such issues can be addressed in several ways such as modifying the nozzle size or shape, robot speed, end-eftector stiffness, or the overall geometry to avoid tight corners (to avoid kinks in toolpath). However, modifying these variables also impacts other aspects of the building design such as the overall structural performance, construction cost, material cost, and construction time. Hence, it is essential to capture such nuanced tradeoffs in additive construction in order to ensure successful printing.
Further, the taper angle also has an impact on the interior spatial planning, as a steep taper angle might result in reduced interior usable volume, essentially rendering the space near the wall—floor connection unusable. Hence it is also essential to incorporate the other elements of the buildings such as the interior furniture, МЕР systems, and ECLSS, (Figure 13.21) and
Figure 13.20 Over-extrusion of material around corner profile due to vibrations caused by end effector.
Figure 13.21 Schematic render of BIM showing multiple systems in the design selected for the virtual construction level.
also evaluate the overall spatial layout for accessibility and other spatial qualities. For the actual construction level, a minimum usable floor area (aj of 110 nr2 per the NASA habitability standards was prescribed as a requirement in the design brief (NASA and Bradley University, 2018). Incorporating other functional requirements—such as living and work areas, bedrooms, bathrooms, kitchen and dining area, leisure area and associated systems, services, and interior fixtures—introduces the need for interior walls and partitions. Such walls, partitions, and the intersections between the interior walls and the overall structure need to be designed within the constraints of the overall geometry. Subsequently, it is also equally important to design the МЕР and ECLSS systems in such a manner that the routing of the service systems (pipes, cables, and ducts) do not have collision paths with other service elements and are easily serviceable. The design for the virtual construction level was envisioned to have a centralized MEP/ECLSS distribution system along with other water storage and treatment tanks. The positioning of these components involves identifying optimal routing paths for the floor trenches, which carry the ducts, pipes, and electrical cables. The route also had to be designed in such a manner that the number of kinks and the minimum distance between consecutive kinks in the pipe were to be optimized to avoid excessive internal pressure in the ducts and pipes. The wet wall that distributes the water and electricity outlets to the kitchen/ dining and bath areas had to be positioned to align with the trench route. This, in turn, impacted the spatial planning of the room layout and other interior spaces.
In summary, there existed various tradeoffs while designing these interconnected systems and it was crucial to optimize the building design concurrently for multiple aspects (such as structural, IEQ, and constructability, i.e., toolpath and logistics) to ensure holistic performance of the additively constructed building (Figure 13.22). To enable such concurrent optimization for multiple design objectives (or goals), it is necessary to generate a large set of design options and concurrently analyze them tor their performance factors. Since the building design process involves multidisciplinary stakeholders dealing with multiple domain-specific modeling, analysis, and simulation tools, it was necessary to integrate these federated BIMs into an integrated (or master) BIM. Furthermore, to enable concurrent optimization, the integrated BIM was connected to an optimization routine via Dynamo for Revit (Figure 13.23). Dynamo was used as a platform enabling common data exchange for
Figure 13.22 Conceptual representation of the data flow between the modeling and analysis tools (structural, environmental, and constructability tools).
Figure 13.23 (Top) Interactive user interface for modifying high-level design goals and constraints; (Bottom) Conceptual representation of a connected BIM ecosystem for modeling, analysis, and optimization.
sharing information between the modeling, analyses, simulation, and optimization tools. It should be noted that, for the efficient use of such optimization algorithms, it is essential to define the quantitative design goals or objectives as a mathematical formulation.
First, the objective requirements and constraints were extracted from the design brief and encoded as rules in a procedural modeling algorithm (or generative algorithm) in Dynamo for Revit. Further, the minimum and maximum limits on usable living and functional area, access corridors, furniture spacing, and positioning per the habitability requirements from NASA-STD-3001, Volume 2 (NASA, 2019) were mathematically formulated into the procedural modeling algorithm as well. Design variables such as number of modules (n j, radius of the base profile of the module(s) (r), connector type (tangential, overlapping, at a distance), distance between modules (d), angle between modules (a), height of straight walls (h|lmv), overall height of the modules (hmJ, tapering angle of the conical extrusion (a ), wall thickness (fj, connector sectional profile manipulators, width of connector (w), height of straight walls in connector overall height of connector (hco), tapering angle of connector (a ),
width (wj of entry/exit modules, height of straight walls in entry/exit modules (hcsv), overall height of entry/exit modules geometrical manipulators for openings and penetrations in entry/exit modules, floor slab thickness (tf), number of floors (»f), number of crop shelves inside food production module (я), spacing between crop shelves (h), minimum unkinked МЕР duct/pipe length to maintain optimal pressure (/mcp), minimum room dimensions (rmjn), maximum room dimensions (rmx), and minimum usable area (aj. among other parameters were encoded as parametrically controlled input variables into the procedural modeling algorithm thereby enabling rapid change of these variables as required (Muthumanickam et al., 2020a).
The overarching design goals were to maximize usable floor area, maximize usable indoor volume, maximize compressive strength, minimize build cost based on construction material (concrete) quantity, and minimize overall 3D-printing time. The optimization routine aided in finding the combination ot design variables that would satisfy the design goals. Additional constraints such as the limits on the taper angle determined from the physical tests, optimal wall thickness tor Martian conditions, and optimal wall thickness based on material testing results were also used as guiding factors for the optimization routine. A web-based interactive user interface was developed that enabled multidisciplinary users to modify the design goals and constraints (Muthumanickam et al., 2020d). This, in turn, was connected to the master B1M model, analyses tools, and the optimizer via Dynamo for Revit. Hence, any change in the objectives or constraints triggered the optimizer to find a solution per the new inputs as well as update the master BIM model accordingly. The web-based interactive interface was hosted in Microsoft (MS) Azure and was connected to an optimizer utilizing nondominated sorting genetic algorithm II for searching the optimal design option. Whenever the design goals or constraints were modified on the interactive front end, the input values and the resultant optimal design option from the optimizer were appended to a database in MS Azure. This database of user inputs and optimal solutions were then used to train MS Azure’s inbuilt supervised ML models (a subset of Al techniques). These Al techniques inbuilt within MS Azure helped build a predictive analytics model, which can identify optimal solutions based on input values more quickly as the size of the training database increases. However, it should be noted that this approach was purely explored to decrease the querying time and more research is necessary to validate the accuracy of the ML models used.
Using the above BIM-based optimization framework, two design options were selected: a three-module version for the virtual construction level and a two-module version for the actual construction level (Figure 13.24). The actual construction level had to be designed with interlocking mechanism to prevent lateral slip due to cold joints, where fresh concrete is deposited over cured concrete after an intermittent stop in 3D printing.
Figure 13.24 Design option selected for (Left) virtual construction level (Render credits: Eric Mainzer) and (Right) actual construction level. (Far Right) Design of interlocking geometries to address cold joints.
Digital twin-based 4D simulation
Once the design options were finalized, detailed digital twin models of the production setup were developed to perform detailed 4D simulations of the entire construction process. It should be noted with careful distinction that the constructability simulation (toolpath and robot motion planning) used as part of the analysis and optimization to arrive at a final geometry is different from the digital twin-based 4D simulation (outlined in this section). A digital twin-based 4D simulation includes of all the equipment used in the production process such as the material feeder system, material mixer system, storage silos, delivery truck, safety barricades, and enclosures, in addition to the robots. Ideally, such models can be used for analysis and optimization as well. But generating digital twin-based constructability simulations for multiple models as outlined in the previous section would be computationally expensive. Hence, a simplified toolpath simulation was used in the previous stage. The production system including the robots, equipment, and other logistics were modeled on top of the integrated BIM model in Revit. A custom node in Dynamo for Revit was used to stream this model into Grasshopper tor Rhino, within which a HAL Robotics plugin was used to develop the 4D simulation. The toolpath of the sliced geometry was used as the G-code to guide the robotic motion in the 4D simulation. The order of execution of the other construction tasks and logistics were programmed sequentially using a custom-built timeline editor within Grasshopper for Rhino. Separate digital twin—based 4D simulations were generated tor the virtual and actual construction levels of the NASA Challenge. The 4D simulation for the virtual construction level was developed to mimic a sequence of robotic construction tasks in the Martian environment (Figures 13.25 and 13.26), whereas the 4D simulation for the actual construction level was developed tor conditions of the competition venue in Peoria, Illinois (Figure 13.27). To resist lateral loads on Mars, the base of the structure was envisioned to start from an excavated area. Furthermore, in order to protect the freshly printed wet concrete structure from extreme atmospheric conditions, a retractable enclosure would be deployed after which the steps 3—16 shown in Figure 13.25 would be executed in sequence.
The second digital twin—based 4D simulation model was developed using the same software framework as mentioned above, but this time for the actual construction level setup, which included a delivery truck, trailer, silo for feed stock material storage, water tank, mixer, pumps, two ABB 1RB 6640 robots, computer stations, safety barricades, window fixtures, scissor jack for build space monitoring, and camera tor visual feedback (Muthuman- ickam et al., 2020a) (Figure 13.27).
Based on multiple iterations of the simulations and learning from the production system setup at our Additive Construction Laboratory (AddCon Lab) at Penn State, the digital twin model of the production system to be deployed at the competition venue in Peoria, Illinois was refined and details prescribed per the NASA on-site safety standards prescribed in the competition brief. Figure 13.28 is an image comparing the digital twin model with the actual setup at the finals of the actual construction level of the NASA Challenge.
Finally, a sub-scale version (one-third scale) of the proposed design for the actual construction level was 3D printed on-site during the finale of the NASA 3D-Printed Mars Habitat Centennial Challenge held in Peoria, Illinois. The final print occurred over three days, during which each team was given ten hours a day to 3D print their proposed designs. The allocated ten hours per day was exclusive of the time for site preparation, production system setup, and daily
Figure 13.25 Digital twin 4D simulation of additive construction of Martian habitat (virtual construction level) (Simulation credits: Eduardo Castro e Costa and Negar Ashrafi).
equipment preparation and cleaning after stopping at the end of each day (one half hour was allotted in the morning and an additional half hour at day’s end). Our proposed design was 3D printed in a cumulative time of 16.5 hours spanning three days with stops at the end of each day. Though most of the process was executed as per the digital twin-based 4D simulations, there were a few on-site anomalies in the robotic 3D printing process, which required manual intervention. Out of the three window fixtures that were programmed to be placed autonomously by a second ABB IRB 6640 robot, two windows were placed successfully, whereas the autonomous positioning of the third window failed. This was due to the over-extrusion of concrete at the edge surrounding the gap for the window, which made the gap smaller than the size of the window by a few millimeters. This prevented the window from sliding into the gap as planned and, hence, required manual scraping to remove the extra material deposited on the edge. Another challenge we faced during the competition was the failure of a few beads
Figure 13.26 Sequence of printing of enclosure and assembly of interior components and systems (Enclosure will be printed right after Step, but not shown for visual clarity of interior elements).
Figure 13.27 Digital twin 4D simulation of additive construction of one-third-scale habitat at competition venue (actual construction level) (Simulation credits: Naveen K. Muthumanickam and Eduardo Castro e Costa).
Figure 13.28 Comparison of BIM-based digital twin of the production setup and the actual setup at the competition venue during the finals.
on the interior of the walls at the location where the structure starts to taper. More specifically, this problem occurred during printing of the top of the structure, which required the robot arm to perform a 180-degree flip of its sixth axis (end effector) to reach such heights. Due to the flipping of the end effector, there was minor offset error in terms of the robot position calibration, which caused the toolpath to shift by a few millimeters, thereby resulting in the innermost beads falling without underneath support. However, despite such drawbacks, the team was able to complete 3D printing the entire structure as a unified tapering concrete structure from floor plate to the roof without any support structure or formwork (Figure 13.29).
Learnings, limitations, and future avenues of research
The unreinforced concrete structure that was 3D printed at the finals of the NASA Challenge was subject to a range of tests including a compressive strength test, during which a 90+ metric ton (MT) excavator pressed down on the top of the structure at the weakest location identified for the structure (Figure 13.30). The unreinforced concrete structure resisted the load until the tracks of the excavator were off the ground, at which time the structure partially failed. Even then, the structure only partially tailed with one module staying intact showcasing significant structural strength.
The use of an integrated BIM platform from the design conception stage (modeling) to design development (analysis, optimization, and simulation) to the actual 3D printing of the structure using industrial robots, was as an efficient platform for multidisciplinary collaboration, design, and construction. Specifically, it helped the team in adopting a data-driven design approach, which helps minimize conflicts and costly revisions. However, based on the tests conducted in our AddCon Lab at Penn State as well as learning from the competition, there is room for improvement to make the existing BIM-based design and optimization framework more robust. Key to ensuring successful execution of the additive concrete construction of the overall building is to conduct trial prints of small-scale samples and conducting physical tests (structural, thermal, and leakage tests) before the actual construction of the full-scale structure. Often, there might be anomalies uncovered during the printing tests even when the robot strictly follows the 4D simulations, as it is difficult to construct a digital
Figure 13.29 Sequence of snapshots of additive construction of the sub-scale habitat at the finals.
Figure 13.30 Compression test showing raising of excavator trails (dotted white lines).
twin model that exactly mimics the instantaneous conditions (e.g., temperature changes, humidity, and air flow) on the printing site. Apart from the printing tests, rheological tests are crucial for predicting anomalies such as over-extrusion ot materials. Though rheological experiments are predominantly physical experiments, detailed computational fluid dynamics (CFD) models can also be developed to virtually simulate the material flow for variable geometries. However, integrating such physics-based analysis tools with the BIM-based optimization framework would be difficult and, hence, is a potential area for future research.
Design of feedback mechanism
in case of geometrical errors due to material deformation or toolpath calibration error, some human intervention is necessary to handle such anomalies and avoid any potential failures or lapses in the additive construction process. To eliminate such human interventions, advanced computer vision sensors leveraging photogrammetric or LiDAR sensors can be used to collect real-time geometrical data about the printed geometry and deformations, if any. This can then be used for real-time corrective adjustment of motion-specific variables such as the toolpath and robot speed, or to control the material mixing ratio to avoid any potential failures. Such an effort requires creating a database for the storage of real-time computer vision data and sophisticated algorithms that can optimize the design in real time and update the 13IM model and G-code according to these observations. Further, ML techniques (especially supervised reinforcement learning) can also be used to compare the predicted performance of the structure (from various performance analyses tools) and the actual performance ot the structure (from the real time data collected via computer vision sensors) (Figure 13.31). Such an approach will help in increasing the accuracy of predicting the actual performance of the structure through simulations beforehand.
Figure 13.31 Conceptual framework for a feedback mechanism for real-time data collection from physical tests using computer vision sensors.
In summary, a novel connected software ecosystem utilizing state-of-the-art developments in BIM, parametric modeling, database management, and APIs was developed to support the design for additive concrete construction framework. Such a unified BIM platform allowed integration ot multidisciplinary modeling, analysis, optimization, and simulation tools and enabled the consideration of manufacturability and constructability constraints in addition to other performance factors such as structural, indoor environmental performance, and cost. In addition to the BIM framework that allowed the concurrent design of the overall building, and along with the design of toolpath and robotic production system and logistics, the BIM-based digital twin models were highly imperative in simulating the entire sequence of robotic construction tasks. Utilizing such a framework, it was possible to construct the world’s first fully enclosed concrete structure that was 3D printed from floor to roof, tapered toward its highest point, without any support structures/formwork at architectural scale. With promising trends yielded by the currently developed BIM framework, more robust and efficient designs can be developed by enhancing the BIM framework through its integration with real-time feedback systems and sophisticated CFD tools capable of simulating material flow.
This research was supported by prize money from NASA’s 3D-Printed Mars Habitat Centennial Challenge and by grants from the College of Arts and Architecture and the College of Engineering, and developed at the Stuckeman Center for Design Computing (SCDC) and at the Additive Construction Laboratory, The Pennsylvania State University. The authors acknowledge the valuable contribution to the research and express their gratitude to the following members of the Penn State team: Nick Meisel, Aleksandra Radlinska, Randal | Bock, Maryam Hojati, Keunhyang Park, Negar Ashrafi, Eduardo Castro e Costa, Samuel Dzwill, Andrew Przyjemski, Flavio Craveiro, Joe Straka, Aiden Smith, and Zhanzhao Li.
Ashrafi, N., Duarte, J. P., Nazarian, S., & Meisel, N. A. (2019). Evaluating the relationship between deposition and layer quality in large-scale additive manufacturing of concrete. Virtual Physical Prototyping 14, 135-140. https://doi.org/10.1080/17452759.2018.1532800 Attia, S., Hamdy, M., O’Brien, W., & Carlucci, S. (2013). Assessing gaps and needs for integrating building performance optimization tools in net zero energy buildings design. Energy and Buildings, 60, 110-124. https://doi.Org/10.1016/j.enbuild.2013.01.016 Craveiro, F., Bartolo, H., Duarte, J., & Bartolo, P. J. (2017). Designing cork-based functionally-graded concrete walls. In F. Moreira da Silva, H. M. Bartolo, P. Bartolo, R. Almendra, F. Roseta, H. A. Almeida, A. C. Lemos (eds.), Challenges for Technology Innovation: An Agenda for the Future (pp. 431-434). London: CRC Press, https://doi.org/10.1201/9781315198101-86 Craveiro, F., Nazarian, S., Bartolo, H., Bartolo, P. J., & Duarte, J. P. (2020). An automated system for 3D printing functionally graded concrete-based materials. Additive Manufacturing, 33, 1—10. https:// doi.org/10.1016/j.addma.2020.101146
Duarte, |. P. (2001). Customizing Mass Housing: A Discursive Grammar for Siza’s Malagueira Houses. PhD. Thesis, Massachusetts Institute of Technology, Cambridge, MA. Retrieved from https://dspace. mit.edu/handle/1721.1/8189
Duarte, |. P. (2019). Customizing mass housing: Toward a formalized approach. In B. Kolarevic & J. P. Duarte (eds.), Mass Customization and Design Democratization (pp. 129—142). New York: Routledge.
Flager, F., & Haymaker, J. (2009). A Comparison of Multidisciplinary Design, Analysis and Optimization Processes in the Building Construction and Aerospace Industries. CIFE Technical Report. Stanford, CA: CIFE. Retrieved from https://purl.stanford.edu/mcl98xh9178
Gan, Y., Dai, X., & Li, D. (2013). Off-line programming techniques for multirobot cooperation system. International Journal of Advanced Robotic Systems, 10(7), 282.
Haymaker, J., Bernal, M., Marshall, M. T, Okhoya, V., Szilasi, A., Rezaee, R., Chen, C., Salveson, A., Brechtel, J., Deckinga, L., Hasan, H., Ewing, R, & Welle, B., (2018). Design space construction: A framework to support collaborative, parametric decision making. Journal of Information Technology in Construction, 23(8), 1Б7—178. Retrieved from http://www.itcon.Org/2018/8
Labonnote, N., Ronnquist, A., Manum, B., & Riither, P. (2016). Additive construction: State-of-the- art, challenges and opportunities. Automation in Construction, 72, 347—366. https://doi.Org/10.1016/j. autcon.2016.08.026.
Leicht, R., Fox, S., Makelainen, T., & Messner, J. (2007). Building Information Models, Display Media, and Team Performance: An Exploratory Study. Espoo: VTT Technical Research Centre of Finland. VTT Working Papers, No. 88. Retrieved from http://www.vtt.fi/inf/pdf/workingpapers/2007/W88.pdf
Muthumanickam, N. K., Park, K., Duarte, J. P., Nazarian, S., Memari, A. M., & Bilen, S. (2020a). BIM for parametric problem formulation, optioneering and 4D simulation of a 3D-printed martian habitat: A case study of the NASA 3D-printed habitat challenge. In Proceedings of the 5th Residential Building Design and Construction Conference. March 4—6, 2020, State College, PA. Retrieved from https://www.researchgate.net/publication/341451080_BIM_for_parametric_problem_formu- lation_optioneering_and_4D_simulation_of_3D-printed_Martian_habitat_A_case_study_of_ NASA’s_3D_Printed_Habitat_Challenge
Muthumanickam, N. K., Duarte, J. P., Nazarian, S., Bilen, S. G. & Memari, A. M. (2020b). BIM for design generation, analysis, optimization and construction simulation of a Martian habitat. In Proceedings of the ASCE (American Society of Civil Engineers) Earth & Space Conference 2021. April 1—22, Seattle, WA. (Forthcoming)
Muthumanickam, N. K., Brown, N., Duarte, J. P., & Simpson, T. W. (2020c). Multidisciplinary analysis and optimization in architecture, engineering, and construction: A detailed review and call for collaboration. (Submitted to Journal of Structural and Multidisciplinary Optimization).
Muthumanickam. N. K., Duarte, J. R, & Simpson, T. W. (2020d). Multidisciplinary concurrent optimization framework for multi-phase building design processes. (In preparation).
NASA. (2019). NASA-S'I D-3001: NASA spaceflight human-system standard, Volume 2: Human Factors, Habitability, and Environmental Health, Revision B. Washington, DC: NASA. Retrieved from https:// www.nasa.gov/sites/default/files/atoms/files/nasa-std-3001_vol_2_rev_b.pdf
NASA and Bradley University. (2018). On-site Habitat Competition Rules, NASA 3D-printed Mars habitat challenge - Phase 3. Retrieved from https://www.bradley.edu/sites/challenge/assets/doc- umentsZ3DPH_Phase_3_Rules-v3.pdf
Nazarian, S., Pantano, C., Colombo, P., & Marangoni, M. “Ceramic glass joints: Transitioning interface from glass to geopolymer cement.” [aka “Seamless architecture: Innovative material interfaces”, US provisional patent application 62/322,864 filed on April 15, 2015; converted to a Patent Cooperation Treaty (PCT) application PCT/US2017/027976 on April 17, 2017; Published on October 19, 2017 as WO2017181191. Retrieved from https://patents.google.com/patent/ WO2017181191 Al/en
Polit-Casillas, R., & Howe, S. A. (2013). Virtual construction of space habitats: Connecting building information models (BIM) and SysML. In AIAA Space 2013 Conference and Exposition, pp. 1—19. San Diego, CA. Pasadena, CA: Jet Propulsion Laboratory, National Aeronautics and Space Administration. https://doi.Org/10.2514/6.2013-5508
Rael, R., & San Fratello, V. (2011). Developing Concrete Polymer Building Components for 3D Printing. Retrieved from http://www.rael-sanfratello.com/media/emergmg_objects/papers/243.pdf
Simon, H. (1969). The Sciences of the Artificial (1st ed.). Cambridge, MA: MIT Press.
Stiny, G. and Gips J. (1972) Shape grammars and the generative specification of painting and sculpture. In С. V. Freiman (Ed.), Information Processing, 71 (pp. 1460—1465). Amsterdam: North-Holland. Retrieved from https://architecture.mit.edu/sites/architecture.mit.edu/files/attachments/publica- tions/SGIFIPSubmitted.pdf
Sutherland, I. (1963) Sketchpad: A Man-Machine Graphical Communication System. PhD Thesis, Massachusetts Institute ofTechnology, Cambridge, MA. Retrieved from http://hdl.handle.net/1721.1/14979
Thompson, M. K., Moroni, G., Vaneker, T., Fadel, G., Campbell, R. I., Gibson, I., Bernard, A., Schulz, J., Graf, R, Ahujai B. and Martina, F. (2016). Design for additive manufacturing: Trends, opportunities, considerations, and constraints. CIRP Annals, 65(2), 737—760. https://doi.Org/10.1016/j. cirp.2016.05.004
Turing, A. (1948). Machine intelligence. In B. J. Copeland (ed.), The Essential 1'nring: The Ideas that Gave Birth to the Computer Age (p. 412). Oxford: Oxford University Press, van der Zee, A., & Marijnissen, M. (2017). 3D concrete printing in architecture: A research on the potential benefits of 3D printing in architecture. In A. Fioravanti, S. Cursi, S. Elahmar, S. Gargaro, G. Loffreda, G. Novembri, & A. Trento (Eds.), SI10CK— Sharing of Computable Knoudedge - Proceedings of the 35th International Conference on Education and Research in Computer Aided Architectural Design in Europe (Vol. 2, pp. 299—308), Rome, Italy, 20—22 September 2017. Rome: eCAADe. Retrieved from http://papers.cumincad.org/cgi-bin/works/ShowPecaade2017_087 Watson, N. D., Meisel, N. A., Bilen, S. G., Duarte, J. P., & Nazarian, S. (2019). Large-scale additive manufacturing of concrete using a 6-axis robotic arm for autonomous habitat construction. Solid Freeform Fabrication 2019: Proceedings of the 30th Annual International Solid Freeform Fabrication Symposium, Austin, TX, 12—14 August 2019, pp. 1583—1595. Retrieved from http://utwl0945.ut- web.utexas.edu/sites/default/files/2019/134%20Large-Scale%20Additive%20Manufacturing%20 of%20Concrete%20Usi.pdf
Winston, P. (1992). Artificial intelligence (3rd ed.). Reading, MA: Addison-Wesley.