Computers in Industry 139 (2022) 103659

Contents lists available at ScienceDirect

Computers in Industry
journal homepage: www.elsevier.com/locate/compind

A graph-based approach for module library development in

industrialized construction ]]
]]]]]]
]]

⁎
Jianpeng Cao a, , David F. Bucher a, Daniel M. Hall a, Mike Eggers b
a
Institute of Construction and Infrastructure Management, ETH Zurich, Stefano-Franscini-Platz 5, Zurich 8093, Switzerland
b
Project Frog, 99 Green St, San Francisco, CA 94111, United States

a r t i cl e i nfo a bstr ac t

Article history: A library of prefabricated parts and assemblies, i.e. module library, can help a firm in the construction
Received 26 October 2021 industry transition to a more industrialized and product-oriented approach. However, existing approaches
Received in revised form 3 March 2022 to manage such libraries are oriented around single-use projects. There is need for a more flexible data
Accepted 15 March 2022
structure to support storage, analysis and reuse of design information. This paper proposes a graph-based
Available online 29 March 2022
approach to develop a module library. The approach includes a graph representation of modules, a graph
database development, and a graph-based similarity analysis. The proposed approach is validated using a
Keywords:
Module library prefabricated timber panel system via a web-based application. Implementation demonstrates a more ef­
Graph database ficient process for bill of material generation and identifying the impacts of design changes.
Similarity analysis © 2022 The Author(s). Published by Elsevier B.V.
BIM CC_BY_NC_ND_4.0
Industrialized construction

1. Introduction However, it is not always easy for the construction industry to

shift to this new paradigm. The industrialized construction projects
Industrialized construction is experiencing a new wave of at­ sometimes suffer from the negative aesthetic perception of “mod­
tention and investment. In a 2019 McKinsey report, the construction ular”, “box-like”, or “cookie-cutter” architecture. As a result, de­
industry could deliver a $20 billion annual savings and 50% time signers might not be willing to accept this process due to limited
saving if industrialized construction is adopted (Bertram et al., 2019). design flexibility. Other challenges such as increased capital cost at
A crucial factor for achieving this goal is DfMA (Design for Manu­ the early adoption stage and lack of manufacturing knowledge for
facturing and Assembly). DfMA is originated from advanced manu­ designers also restrict the DfMA application in actual projects (Lu
facturing, and is used as the basis in concurrent engineering for cost- et al., 2020; Gao et al., 2020). To mitigate these challenges, it is
effective design. The main idea of DfMA is to consider the manu­ critical to provide designers with manufacturing input during their
facturing and assembly issues in the design phase. DfMA saves de­ design.
sign time and cost through the use of standard common parts, the Today, the most common approach to product libraries is to
reuse of previously designed details and the application of modular provide design flexibility in various configurations by selecting and
design (Anderson, 2003). Industrialized construction adapts this recombining well-designed and engineered building elements from
strategy of DfMA by developing standard types of prefabricated product libraries (Cui et al., 2020). Existing construction product li­
building elements (e.g., panel, component and/or volumetric mod­ braries e.g. NBS library, Open Source Wood, etc, enable designers to
ules) and then providing these modules to architects who can then find similar designs and templates on which to copy or modify.
fulfill design intentions. The strategy is particularly beneficial for a However, those libraries contain few prefabricated components, or
product structure that has a high degree of repeatability and stan­ modules, suitable for industrialized construction and do not provide
dardization (Bertram et al., 2019), such as hotels, affordable housing, easy access for designers to retrieve desired content (Li et al., 2020).
schools, hospitals and so forth. Some new industrialized construction firms, such as DMD Modular,
Project Frog, CIMC MBS, etc., start to develop their own module li­
braries as their core products, or product catalog. The module library
of industrialized construction companies almost falls into three ca­
⁎
Corresponding author. tegories: volumetric modular, flat-pack and kit-of-parts, according to
E-mail addresses: [email protected] (J. Cao),
[email protected] (D.F. Bucher), [email protected] (D.M. Hall),
their business strategies (Pullen et al., 2019). These firms claim to
[email protected] (M. Eggers). offer more competitive pricing, expedited project schedules, and

https://doi.org/10.1016/j.compind.2022.103659
0166-3615/© 2022 The Author(s). Published by Elsevier B.V.
CC_BY_NC_ND_4.0
J. Cao, D.F. Bucher, D.M. Hall et al. Computers in Industry 139 (2022) 103659

increased product quality. Even so, the overall development and units (e.g. factory-prefinished “boxes” that are then transported to
application of such module libraries are still limited in the broader the construction site). For example, Murtaza et al. (1993) described a
construction sector. module as “a volume fitted with all structural elements, finishes, and
Even when module libraries are developed, they are done so by process components that, regardless of system, function, or instal­
relying on specialist knowledge and accumulated experience ling craft, are designed to occupy that space”. De La Torre et al.
(Salama et al., 2017). Although many researchers propose various (1994) defined a module as “a product resulting from a series of
techniques and guidelines to support the module identification offsite assembly operations; it is usually the largest transportable
(Salama et al., 2017; Isaac et al., 2016; Samarasinghe et al., 2019), the unit or component of a facility.”
configuration of modules is often still determined on a project-by- To solve this ambiguity, we look for similar concepts adopted in
project basis (Gosling et al., 2016). However, this can lead to many the manufacturing industry. Salvador categorized five perspectives
unique typologies of modules that can be difficult to manage. A of product modularity, including component commonality, compo­
helpful analogy can be found from LEGO Group in the early 2000s, nent combinability, function binding, interface standardization, and
when the number of unique piece types reached 12,000 and nearly loose coupling (Salvador, 2007). Besides the general definition, two
bankrupted the company (Feloni, 2014). Therefore, it is crucial to characteristics of modular products are highlighted: 1) the similarity
consider the similarity among projects in order to develop a module between the physical and functional architecture of the design, and
library. The similarity indicates that a standard and adaptable design 2) the minimization of the degree of the coupling between the
can be applied in multiple projects. The reuse of the modules in physical and functional components (Ericsson and Erixon, 1999).
future projects will lead to the continuous improvement of project Accordingly, two modularization methods are similarity- and
quality and return of investment made initially inside a single pro­ coupling-based approaches (Borjesson and Hölttä-Otto, 2014). The
ject (Tetik et al., 2019). similarity-based approach to modularization defines modules based
To facilitate the reuse of the modules, the design is digitalized as on similar functions or properties among products to achieve pro­
building information models (BIM). BIM can provide 3D visualiza­ duct variations, or similar manufacturing methods or suppliers, for
tion, monitoring, analysis and prediction during the project life production efficiency. The coupling-based approach to modulariza­
cycle. Each module can be developed as a BIM file via BIM authoring tion detects modules by maximizing the coupling within the mod­
tools, such as Autodesk Revit. A BIM module library consists of di­ ules while minimizing the coupling in between the modules. The
verse models for architectural, civil, structural and mechanical ap­ objective is to reduce the amount of interface management, allowing
plications. However, a file-based management approach provides changes to one module without impacting others, and reducing re­
limited facility for users to extract the data from the respective BIM ciprocal interdependence between modules in order to minimize
authoring tool beyond simple queries on objects and their properties design feedback (Levitt, 2015; Thompson et al., 2017).
(Solihin et al., 2017). Besides, the query languages introduced so far In this paper, we follow the similarity-based approach and define
require a high level of knowledge about the IFC (Industry Foundation the module in terms of component commonality. To this regard, the
Classes) object model and about data mapping mechanisms modular product design problem refers to the designation of any
(Tauscher et al., 2016). Last but not the least, module reuse not only main component, such as volumetric elements and wall panels, that
requires the search and comparison in terms of parameters but also can benefit from standardized and reusable design in multiple pro­
the structural composition, which focuses on relationships between jects.
different parts of a product.
In this research, we propose to solve these challenges using a 2.2. Module library development
graph-based approach for module library development. There are
two key components to our solution. First, we propose the use of a Due to the loosely-coupled nature of the project-based organi­
graph database – a database that uses graph structures for data zation in conventional construction, design reuse remains an under-
storage and data manipulation – to store the BIM models of each utilized source to improve project performance (Dubois and Gadde,
module. Second, we propose to identify similar modules in a graph- 2002). Industrialized construction has the opportunity to reuse de­
based module library using a graph similarity analysis. To our sign through the use of a module library of standardized products.
knowledge, this is the first study on the development of a building The library can support the application of product platforms which
product library via graph database. contain common processes and technical solutions shared by mul­
The paper is organized as follows. In the next section, we conduct tiple projects (Bonev et al., 2015; Jensen et al., 2014; Peltokorpi et al.,
a literature review on the study of modules, focusing on the reusa­ 2018; Jansson et al., 2014), so as to save the design cycle and lower
bility of module libraries in project development. Next, we gen­ production costs. However, current module libraries in the con­
eralize the application areas of graph modeling in the construction struction industry do not apply any form of big data analytics for the
industry. Specifically, we analyze the use of the graph-based ap­ development and assessment of design options in industrialized
proach for module identification. Then, we introduce the graph da­ construction (Gbadamosi et al., 2020). These design options include
tabase for modules’ storage and management. From that, we propose the configuration of prefab modules, geometry and material of de­
a novel graph-based approach for module library development in sign parts and components, as well as alternative suppliers and
the BIM environment. After that, we gave an illustrative example of manufacturers. Previous scholars study modules mostly by case
how the approach is applied to a dataset of timber-framed panels, studies (Peltokorpi et al., 2018; Viana et al., 2017) and identify the
including a description of system architecture and implementation configuration of modules based on their experience and guidelines
results. Finally, we discuss the limitations and future research, and (Salama et al., 2017). To improve efficiency, there is an opportunity to
conclude with our intended contributions to the literature. apply a graph-based product modeling and graph database tech­
nology for module library management.
2. Literature review
2.3. Graph-based product modeling
2.1. Modules in construction projects
In computer science, a graph is an abstract data structure, con­
Construction literature does not have a consistent definition of a sisting of a finite set of nodes and a set of ordered or unordered pairs
module, modular system or modularization (Gosling et al., 2016). In of edges. The structure may also be assigned with certain values to
many instances, modular is used to refer to volumetric assembly each node or edge, such as a categorical label or a numeric value.

2
J. Cao, D.F. Bucher, D.M. Hall et al. Computers in Industry 139 (2022) 103659

More advanced graph structures, such as hypergraphs, contain edges

that can connect any number of nodes. Although a graph can be used
to represent complex engineering systems (Boccaletti et al., 2006;
Zawiślak and Rysiński, 2017), there is little research studying graph
modeling in industrialized construction.
Existing research using graph-based modeling in building design
is mostly limited to floorplans and spatial layouts (Wang et al., 2018;
Wong and Chan, 2009; Strug et al., 2014; Nauata et al., 2020; Strug
and Ślusarczyk, 2009). Few studies tested the graph representation
at the granularity of the element level and how it could support
module library development (Isaac and Navon, 2013). Khalili and
Chua developed a graph-based modeling approach to group single
precast elements into higher-level prefabrication assemblies (Khalili
and Chua, 2013). They searched for all subgraphs exhaustively and
filtered out the feasible configurations by constructability rules.
Isaac et al. applied a clustering algorithm to detect optimal config­
urations of modules in a housing unit (Isaac et al., 2016). A hier­
archical clustering algorithm was also taken by Samarasinghe et al.
to detect modules in mechanical, electrical and plumbing systems
(Samarasinghe et al., 2019). However, the above approaches are
based on a single project and detected modules which might not be
representative for standardization needed in industrialized con­ Fig. 1. A approach supporting the development of the module library.
struction. An increasing number of unique modules might deterio­
rate the production efficiency. In addition, the previous studies relationships can be detected automatically. The graph database for
perform limited similarity analysis for model reuse. Gbadamosi et al. a module library lay the foundation for design-related tasks.
(2020) and Li et al. (2020) propose a semantic-based similarity
analysis to retrieve BIM models. The structural similarity of module
3. Graph-based module library approach
compositions is not taken into account. Last but not the least, most
practices store those modules in a file-based system, without easy
In this section, the proposed graph-based approach for the
access for designers and manufacturers to edit, update, and reuse. To
module library development is described. The approach is composed
access the data locked inside the BIM tools, previous studies apply a
of four key steps (Fig. 1):
DB link from software developers to export BIM data into a relational
database and present domain-related operations as standardized
1. Determining the system boundary by selecting building elements
SQL queries (Solihin et al., 2017). Other studies using RDBMS (rela­
and their attributes as inputs.
tional database management system) develop a database structure
2. Representing the selected module information as a graph and
to integrate data among different software (Solihin et al., 2017; Liu
conducting the same process for all other modules concurrently.
and Issa, 2012; Park and Cai, 2017; Wu et al., 2019). However, a re­
3. Importing graph data into the graph database.
lational database is inefficient to store relationships between dif­
4. Performing graph-based applications, including common graph
ferent elements.
operations, and similar module retrieval.

2.4. Graph database 3.1. System boundary definition

Compared with other types of database, such as relational da­ The categories of a module library in industrialized construction
tabase, graph database explicitly builds the dependencies between include volumetric modular, flat-pack and kit-of-parts, depending
nodes of data. This characteristic enables the representation of on companys’ business strategy (Pullen et al., 2019). To match dif­
products exactly as they are in the 3D modeling environment or real ferent strategies, the system boundary is restricted in terms of three
life. For example, when designers search for similar products, the aspects: building types, elements types, and element attributes for
graph database can not only take into account the existence of the library development. Take timber panelized buildings as an ex­
components, but also the structures of the product. Other features of ample. The input files are a collection of timber panels in different
using graph database in terms of data structures, query languages shapes. Each timber panel contains various types of elements, in­
and integrity constraints have been summarized by Angles (2012). cluding framing studs, wallboards, insulation layers, and metal fas­
Previous studies have introduced the methodologies to extract teners. To simplify the representation of the product, users are able
information from a BIM file (Hor et al., 2018; Ismail et al., 2017; to select the elements “framing studs” as the important components.
Ismail and Strug, 2018; van Treeck and Rank, 2007; Isaac et al., 2016). Finally, users can select dimensions, material properties, or cost of
However, BIM files are built upon a complicated hierarchical order the studs as the element attributes. The selected information is
and reference relationships. They also can contain redundant in­ stored in a graph database for similar model retrieval and project-
formation. Processing multiple large-scale BIM files is still a time- related queries, such as cost estimation.
consuming and error-prone task. Previous studies on BIM-to-graph
transformation do not take into account geometric information 3.2. Graph representation
(Ismail et al., 2017; Ismail and Strug, 2018), which is important for
module design and manufacturing. Finally, the adjacency relation­ The library is built upon the graph representation of selected
ships between elements are not explicitly defined in the BIM files. In elements. The graph model has the capacity to represent building
this study, we build up an easy and user-friendly interface sup­ elements at different levels of detail and at different stages of the
porting transformation from BIM into a graph database. The ele­ process (Ślusarczyk et al., 2017). In this study, we apply a labeled
ments and their attributes can be extracted, and the adjacency attributed graph to represent each module, as well as its

3
J. Cao, D.F. Bucher, D.M. Hall et al. Computers in Industry 139 (2022) 103659

components. For example, a timber panel contains studs, headers,

sills and fasteners. Let us first define a labeled attributed graph.
Definition 1. A labeled attributed graph over the node set (N) and
the edge set (E) is a system G = (N, E, LN, AN), where:

• N is the node set, representing a set of single building elements.

• E is the edge set, representing a set of relationships between
elements, such as adjacency relationships.
• LN is the label of the node, representing the distinct category of
building elements, such as exterior walls. Fig. 3. Node categorization by K-means.

• AN is the attributes of nodes, representing the properties of the

building elements, such as cost.
clustering to determine the distinct element types. K-means is an
In order to build such a graph structure, a web-based tool is unsupervised learning technique to partition n observations into K
programmed in the present research. The tool is written in Javascript clusters in which observations belonging to a cluster are similar to
using Autodesk Forge, a cloud-based development platform for BIM each other. Then, the elbow method is used to decide the optimal
data manipulation, and perform three main operations for the de­ number of groups (K). Finally, the elements within each group are
fined graph, including node identity extraction, node similarity considered as the same element and renamed as a sub-category. The
analysis and edge identity detection. results can be checked by users and updated as well. In this way, we
promote product standardization at the element level. Fig. 3 shows
3.2.1. Node identity extraction an example of clustering analysis for panel components, which are
This step is aimed to define node identity for graph representa­ categorized based on element length and type. For instance, there
tion. A node in a graph refers to a single building element, and node are three “2 × 8 SILL” instances, which are further classified into two
attributes refer to its properties. An element in BIM involves a large groups, namely 2.1 and 2.2 by length.
sum of information, including unique element IDs, geometries, lo­
cations, material properties, etc. Through a developed UI (Fig. 2), 3.2.3. Edge identity detection
users are able to visualize all element categories within the project. The goal of this step is to define the edges of a graph. An edge
Once a category is selected, the related properties can be further between a pair of nodes refers to a connection between two ele­
chosen for extraction. Besides, the interface also enables users to add ments. The previous study defines three types of connections be­
new attributes for other applications. For example, the assembly tween building components, including direct connections, indirect
sequence of a module that is not usually modeled in original design connections, and functional connections (Isaac et al., 2016). Con­
files can be stored as node attributes to facilitate module assembly. nection types can be specified as edge attributes for more precise
graph representation. In this study, we apply a 3D collision detection
3.2.2. Node similarity analysis algorithm to determine the spatial relationship (connected or dis­
Building elements in different projects might share similar connected) between components. The algorithm consists of the
properties. For example, a timber stud with a length 1 meter is si­ following procedures.
milar to a stud with a length 1.1 ms. The component similarity has a
large effect on construction cost and schedule performance (Staub- 1. Generating a bounding box for each element;
French and Nepal, 2007). To quantify the variations of element types, 2. Retrieving the minimal point (lower-left-rear corner of the box)
we first compute commonality among elements by establishing a and the maximal point (upper-right-front corner of the box)
common node identity. This step is to compare nodes by in­ 3. Iteratively implementing the logic equation between every two
corporating properties into a similarity analysis. We apply K-Means different bounding boxes as

Fig. 2. The user interface of node identity extraction.

4
J. Cao, D.F. Bucher, D.M. Hall et al. Computers in Industry 139 (2022) 103659

Minui Maxvi , Maxui Minvi deletion and substitution of nodes and edges. Mathematically, the
GED is formalized as follows:
Where i refers to x, y, z coordinates of the minimal point and the
maximal point. For example, Minux stands for the x coordinate of Definition 2. The graph edit distance between graph G1 and graph
the minimal point of the bounding box u. If the equations are G2 is defined as:
satisfied with regard to x, y, z, the two bounding boxes (elements) k
are connected. Otherwise, they are not connected. GED(G1, G2) = min(e1, … , ek) P(G1,G2) i = 1 c(ei) (1)
4. Returning an adjacency matrix of a graph. where P(G1, G2) refers to the set of operations transforming G1 to
G2. Practically, it can embody the assembly operations needed to
3.3. Graph database development manufacture a product. The fewer operations needed, the more
similar the two products are. To indicate the similarity clearly, the
Finally, the adjacency matrix is transformed into a graph struc­ GED is further converted into the similarity score between (0,1]. The
ture, and the extracted element categories and properties are at­ conversion is performed as:
tached to node labels and node attributes respectively. After the GED (G1, G2)
designs of BIM files are transformed into graphs, this next step is to |G1| + |G2|
normalized GED(G1, G2) = e 2 (2)
store them in a graph database. In this study, Neo4j, a widely-used
graph database, is set up for data storage. As Neo4j is a schema-less A widely used method to compute GED is A* algorithm (Riesen
database, data stored on a disk is all linked lists of fixed-size records. et al., 2007). However, the A* algorithm suffers from high memory
More details about the database development are illustrated in consumption, as it explores all possible mappings between two
Section 4.1 by an implementation. graphs. The problem is even evident for large graphs. Considering
the building assemblies usually contain many components, the GED
3.4. Graph database applications computation might take a long time. Many improved solutions have
been proposed in the literature. In this study, we apply an upgraded
Once the database is developed, common operations, such as A* algorithm which uses a depth-first search strategy with less
insertion, deletion, update and selection, can be implemented in memory and computation time (Abu-Aisheh et al., 2015). The algo­
Cypher query language to visit data efficiently. Cypher is like SQL a rithm is packed in the Python NetworkX.
declarative, textual query language, but for graphs. Besides, custo­
mized functions can be developed for domain applications. To sup­ 3.4.3. Similar module retrieval
port module library reuse, we build a customized function to enable This function is to retrieve similar modules according to users’
similar module retrieval based on graph similarity analysis. input design. Modules with similar structures might serve for the
same design intention. Compared with the traditional keyword
search approach, the graph-based search can involve more structural
3.4.1. Common database operations information by taking advantage of the graph data structure. For
Storing BIM data in the graph database has advantages over other prefabricated assemblies, not only the geometry data but compo­
database systems. First, the graph database can help create a better nents data, such as the bill of materials (BOM), are crucial to de­
representation of the bill of materials from design and engineering termine the similarity in terms of manufacturing cost and time. The
data sources. Since a product is represented as a graph, the com­ retrieval is achieved by calculating the normalized GED score be­
ponents for manufacturing the product can be easily obtained by tween the users’ input design with the modules in the library. The
traversing all nodes of a graph, while RDBMS and other NoSQL da­ similar design is ranked by the GED score from high to low as output.
tabases typically see significant performance degradation when Additionally, by calculating the GED scores, the pairs of modules
traversing data beyond three levels of hierarchical depth (Rathle, with GED scores that exceed a threshold can be categorized as the
2020). Apart from the product data, the supply chain data, such as same group. For mass customization, it is important for companies
suppliers, customer data, such as claims, can also be stored as nodes to manage the trade-off between the variability and economy by
and connected to the component nodes in the graph. With such a controlling the number of modules in the same category.
demand-product-supply view of BOM, the communication among For the effective reuse of modules in the database, the library also
sales teams, engineering teams, manufacturers, and suppliers can be incorporates a multi-criteria similarity analysis according to users’
sped up. Second, the graph database can help manage design demands. In the case of complex modules, designers not only con­
changes. Changes in a construction project can occur during the sider the product structural similarity, but also building-related
design of building components. The changes can have direct impacts performances (e.g., cost, loading capacity, footprint, etc.) to satisfy
on the components which belong to the same system or indirect local building regulations and project-specific requirements. Those
impacts on components which belong to different systems. The in­ performances can be derived from the element properties in the
terdependency between components can be represented as an at­ empirical formula. For example, the load capacity of a timber panel
tribute of the edges. Once a component is revised, a real-time alert is one of the important design factors. The capacity is dependent on
will be triggered to notify the users of the affected components, panel length, height, stud spacing, sheathing nail spacing, sheathing
which are connected by the changed component. The functionality thickness, presence of drywall and opening, and different support
benefits the BIM users as BIM only offer visualization of the new conditions (Quayyum, 2020). With the properties stored in the da­
model design without highlighting the components that are affected tabase, we can perform a parameter-based model to calculate load
by the changed components (Moayeri et al., 2017). capacity for each panel. Then, the reuse query can be encoded as a
vector of metrics that signify the preferred performance of the
3.4.2. Graph similarity analysis product. Final outputs of module candidates can thus be retrieved by
In graph theory, two techniques can be used to conduct graph mapping user queries to product properties.
similarity analysis: 1. graph edit distance (GED) and 2. maximum
common subgraphs (MCS). In this research, we use a GED-based 4. Test case
similarity analysis. The graph edit distance is first proposed by
Sanfeliu and Fu (1983). It is defined as the transformation from the In order to verify the validity of our proposed approach, we have
input graph to the target graph by six operations, including insertion, developed a prototype system for a timber panel library (shown in

5
J. Cao, D.F. Bucher, D.M. Hall et al. Computers in Industry 139 (2022) 103659

Fig. 4. Example of an existing timber panel library used by industrialized construction firm.

Fig. 5. Example of a technical configurator for timber-panelized buildings.

Fig. 4). The library is originated from an industrialized construction structural analysis engine in the backend will check the design based
firm in the United States, mainly focusing on timber construction. on local building regulations. The prototype aims at supporting the
Their projects consist of prefabricated timber panels, that are com­ panel library storage, sharing, and reuse.
bined in various configurations to create mass-customized buildings.
The technical configurator (Fig. 5) is supported by a library of 3D 4.1. Complexity and relevance
models in Autodesk Revit, 2D engineering drawings in AutoCAD, and
technical documents that define all relevant design regulations. As a simple example, we select a timber-framed panel product
Using the configurator, the architects can first define the space for a test case. Timber prefabricated panels are an increasingly
boundary and room partitions on a grid system. Then, they select popular and utilized industrialized building system (Staib et al.,
panel types from the library and place them in panel slots. Finally, a 2008). The panel consists of multiple parts and sub-assemblies,

6
J. Cao, D.F. Bucher, D.M. Hall et al. Computers in Industry 139 (2022) 103659

Fig. 6. Graph representation of a panel where nodes represent structural studs and edges represent connection relationships.

making it a good case to demonstrate structural complexity. A high- been created. Fig. 6 shows an example of timber panel model and
level fabricated panel integrates structural frames, insulations, ser­ corresponding graph representation. The backend can now access
vices, vapor control layers, internal linings, doors and windows, the data and perform design-related tasks. The frontend can connect
joineries and claddings. In the test case, we limit the components of the geometry to parts with relations for visualization and show the
a panel to the structural frame. By doing so, we do not lose the results of the processed tasks. Fig. 7 shows the system architecture
generality because the structural diversity of panels mostly lies in and Fig. 8 shows the data processing workflow.
the frame layout and geometry. To be specific, the stud type, mate­
rial, dimensions, and spatial relationship are studied in illustrated 4.3. Application and results
applications. For more complex building products, such as pre­
fabricated pods, more components and properties need to be taken The application tier of the prototype processes the main func­
into consideration. The design process requires collaboration among tionalities. This research mainly paid attention to the design-related
different trades. This might require data integration among different tasks, including the bill-of-materials (BOM) generation, identifica­
trades’ databases. tion of impacts of design changes, and similar model retrieval. More
application services, such as cost analysis, could be added to the
4.2. System architecture and data model prototype to support the module library reuse.

The global system architecture, as well as the data model were 4.3.1. Bill-of-materials generation
developed with a focus on modularity and exemplary characteristics. The bill-of-materials (BOM) is the key to information systems
Specifically, we aim for a prototype that is reproducible by others within industrialized construction. A BOM can define a product as
and presents the main concepts of development in a simple way. they are designed, as they are ordered, as they are built and as they
Globally, the prototype was implemented with a three-tier archi­ are maintained. It acts as a bridge connecting islands of information,
tecture, using the relatively new mono repository (monorepo) ap­ such as MRP (Material Resource Planning), ERP (Enterprise Resource
proach. This provides for all applications and libraries to be Planning), and PDM (Product Data Management). Those systems
contained in the same repository and benefits the development with share the same information as BOM with different views.
easy dependency management. Changes can be easily verified across In existing BIM applications, a BOM is generated by selecting
all affected parts of the system. Our prototype includes a frontend element categories and filtering all individuals within those cate­
based on the Angular framework and the backend implemented in gories. In a large construction project, such an approach cannot
NestJS. We use a graph-based database solution, in our case Neo4J, automatically display the assembly-part relationship. For example, it
connecting to the backend. It consists not of tables, but of the data is difficult for production managers to know which specific parts
nodes with the properties and the edges accounting for relations. under the “stud” category are used in “Panel A205”, and further
Upon launching the platform, the backend ensures that the input conduct quality inspection. By comparison, an assembly is stored as
model is retrieved from Autodesk Forge. The geometry and proper­ a graph and all associated components are stored as nodes. Fig. 9
ties are processed and decoupled from each other. Geometry is ex­ shows the manufacturing BOM of “Panel A205”, which comprises a
tracted by Forge Viewer API and sent directly to the frontend, while collection of components, together with the required quantities. To
properties are extracted by Forge Model Derivative API and im­ support the production process, the BOM also contains product
plemented by casting into the Parts class. Next, the generated objects specifications, such as raw material, cutting length, etc.
are stored as nodes in the database. At this point, the edges between
the nodes are not obtained yet. A 3D collision detection algorithm is 4.3.2. Identification of impacts of design changes
performed to calculate the neighbors of the objects via the UUID Existing BIM applications are not capable of automatically
(Universally Unique Identifier). Derived from this calculation, objects manifesting the effect of a design change, unless the change violates
of the Relation class are created as two connected node pairs and a rigid constraint, such as a clash. As a result, designers need to keep
added to the database. Finally, a complete graph representation has the changes in mind and check the effects manually. Instead, with

7
J. Cao, D.F. Bucher, D.M. Hall et al. Computers in Industry 139 (2022) 103659

Fig. 7. System architecture.

the graph model, the element dependencies, such as physical ad­ and therefore is not efficient for complex assemblies analysis, such
jacency, are explicitly represented and stored as edges. The de­ as framed panels. As a result, the same query will return all panels
pendency modeling assists designers to identify the components in Fig. 9.
affected by a proposed change. In this case, the panels comprise To demonstrate the proposed approach is effective to retrieve
studs, headers, sills, plates, and fasteners. Changing any of the similar models, we attempt to set a target model and return a set of
components will directly affect its nearby components. The effect similar models. Fig. 11 provides a sample of the results. The panel
can be indicated by the neighborhood of the changed node. A models in the first column are used as targets, and a set of similar
transaction will be implemented in the graph database to perform models from the library are retrieved in the following columns. In
the changes made. The affected building elements can be visualized the first two rows, the targets are panels with window openings. The
through the interface. As an example in Fig. 10, the size of the header difference between the two targets is the window width. In the last
(light blue) is changed, the studs on the top of the header and on the two rows, the targets are panels with door openings. The difference
two ends of the header are affected and colored in red. If the change is the door width. The type of the opening (window or door) and the
is made by mistake, the designers can easily recover the results width of the opening affects the stud framing of the panels. Espe­
committed by the transaction and maintain consistency with the cially around the opening area, a double stud is needed. To help
data stored in the backend. users find the most similar models, the outputs are ranked based on
the GED score (below) from high to low. In this study, we filtered out
4.3.3. Similar model retrieval the outputs with GED scores lower than 0.6. From the results, it is
The similar model retrieval in the existing product library is built clear that the retrieved models belong to the same types of target
upon the attribute-based approach. For example, a query like “a panels and share a similar structure and geometry. One thing needs
timber panel with an opening” takes material and architectural to be noted. The threshold of GED score can be determined by users
features into account. By matching the attributes given by users’ and is dependent on the model types. If we lower the threshold of
requests and product information, similar products are obtained. GED score, the panels that belong to different types will be regarded
However, this approach does not consider the structural similarity as similar ones. However, this will not affect the search results.

8
J. Cao, D.F. Bucher, D.M. Hall et al. Computers in Industry 139 (2022) 103659

Fig. 8. Data processing workflow.

5. Limitation and future research 5.1. An industry product platform

The study offers a novel graph-based approach for module library The industrialized construction industry is seeking a platform
development. Previous studies neglect the importance of the reuse solution that can be deployed across multiple projects and sectors
of the prefabricated modules due to the loosely coupled nature of (Cao et al., 2021). This platform consists of a scalable module library.
construction projects (Dubois and Gadde, 2002). Industrialized Previous studies demonstrated that a BIM product repository can
construction shares a degree of repeatability and standardization simplify the design process and improve product efficiency
(Bertram et al., 2019). Therefore, by developing such module li­ (Gbadamosi et al., 2020; Nath et al., 2015). To support models’
braries, industrialized construction firms can speed up their project storing, sharing, and reuse, this research builds upon a collection of
development cycle and boost their productivity. To continue the BIM models and establish a graph-based approach. However, in our
research on product library development, we provide the following developed prototype, the library is limited to one specific company
discussion on the limitations and future research. doing timber panelized construction. To achieve a supply-chain

9
J. Cao, D.F. Bucher, D.M. Hall et al. Computers in Industry 139 (2022) 103659

Fig. 9. The manufacturing BOM of the panel “A205”.

Fig. 10. The red studs are affected when the blue header is changed [the model view (left) and database view (right)].

ecosystem to enable mass customization, a platform would need to 5.2. Product variety derivation and auditing
be developed that enabled products from multiple companies to be
added to this platform and shared with the industry. When com­ An increasing number of unique modules in the library might
pared to existing product libraries such as NBS Library, the use of a deteriorate product efficiency. In this research, we conduct a similar
graph-based approach can help recommend specific complementary model retrieval to compare the new design with the existing ones. If
products to help develop a module library. This idea is aligned with a high degree of similarity exists, the new design will not be added
the “Platform Design Programme” initiated by Construction to the library as a unique module to maintain economies of scale.
Innovation Hub. The industry participants can form an alliance and Furthermore, if a firm has completed multiple projects and wishes to
engage in the supply chain network. A decentralized framework can reduce the number of unique modules, they can use the model re­
be utilized to build the industrial level platform (Jiang et al., 2021), to trieval to audit their existing library. When several modules de­
facilitate cross-enterprise information sharing among multiple sta­ monstrate similarity, this can facilitate a structured and data-driven
keholders with transparency and security. In that case, when clients conversation about which modules to keep and which to remove
attempt to launch industrialized buildings on the platform, they can from the library.
easily retrieve the manufacturing quote, as well as an estimated For a company that wishes to begin the development of a module
production schedule, from different companies. library (or to begin a new product line), there is an opportunity to

10
J. Cao, D.F. Bucher, D.M. Hall et al. Computers in Industry 139 (2022) 103659

Fig. 11. Similar panels retrieval based on GED score calculation.

automate the development of new modules with increasing product different types of modular products, such as timber panels versus
variety. Previous studies have demonstrated the graph-based mod­ volumetric modules. For volumetric modules, there exist different
eling can be used for automatic product variety derivation, but most decision-making factors affecting the module selection (Hwang
are limited in the floor plan generation (Wang et al., 2018; Wong and et al., 2018). Therefore, to facilitate module library reuse, the next
Chan, 2009; Strug et al., 2014; Nauata et al., 2020). To extend the step is to build a recommendation system incorporating key deci­
application to the modular building design, graph grammars can be sion-making factors for industrialized building design. It is beneficial
implemented on the graph represented modules as future research. for clients to be notified of the optimal product when there exist
The graph grammars denote a series of graph transformations from hundreds of product choices.
base products to a set of distinct end products. The typical trans­
formations include attaching, removing, swapping and scaling (Du 6. Conclusion
et al., 2002) existing modules. These grammars will be triggered by
the customer-selected product features or project requirements. As a Industrialized construction offers a faster completion time, better
result, a quick response to the clients’ needs can be generated by quality and is more environmental-friendly than traditional con­
configuring modules from the library. struction (Ferdous et al., 2019), especially for projects with a degree
of standardization. An increasing number of stakeholders adopt this
5.3. Multi-criteria reuse analysis approach by delivering their projects from design to manufacturing
and assembly. Modular products owned by industrialized con­
For the module library reuse in new projects, multiple decision- struction firms is a set of prefabricated components, including vo­
making factors need to be considered, such as cost, lead time of lumetric modules, flat-packs and kit-of-parts (Pullen et al., 2019).
production, etc. for different project settings. Other performance Although the categorization of modules (Gosling et al., 2016), as well
simulations can be added with a similar approach, such as LCA as a variety of modules’ designs (Liew et al., 2019; Lawson and
analysis (Hollberg et al., 2020). This lightweight solution replaces the Ogden, 2008), have been illustrated in previous studies, the appli­
process of importing CAD models into authorized software for cation of a module library is not being deployed at scale. Not all
running simulation. As a result, it mitigates error-prone data ex­ clients are aware of using them crossing multiple projects. One of
change and unnecessary remodeling. However, we do not explore an the significant reasons is a lack of a sharable industry product
optimal module selection under multiple criteria, especially among platform where stakeholders can easily obtain modules and adapt

11
J. Cao, D.F. Bucher, D.M. Hall et al. Computers in Industry 139 (2022) 103659

them to new project settings. As a result, modules are mostly de­ Construction Innovation Hub. Platform Design Programme Defining the Need.
veloped project by project, losing the benefits of economies of scale. 〈https://constructioninnovationhub.org.uk/wp-content/uploads/2020/12/FINAL_
VERSION_DEFINING-THE-NEED-REPORT-DEC-2020.pdf〉. (Accessed 8 March
To enable the adoption of industrialized construction by a greater 2020).
segment of the industry, this research proposes a graph-based ap­ Cui, Y., Li, S., Liu, C., Sun, N., 2020. Creation and diversified applications of plane
proach based on the BIM environment to support the storage, module libraries for prefabricated houses based on BIM. Sustainability 12, 453.
https://doi.org/10.3390/su12020453
sharing and reuse of the module library. This paper attempts to Dubois, A., Gadde, L.E., 2002. The construction industry as a loosely coupled system:
make several contributions to the literature. Most importantly, we implications for productivity and innovation. Constr. Manag. Econ. 20, 621–631.
apply a graph database for module library development. A collection https://doi.org/10.1080/01446190210163543
Du, X., Jiao, J., Tseng, M.M., 2002. Graph grammar based product family modeling.
of modules can be stored as graphs in the database. Graph modeling Concurr. Eng. Res. Appl. 113–128. https://doi.org/10.1177/106329303039729
has been used in many applications related to project design and Ericsson, A., Erixon, G., 1999. Controlling Design Variants: Modular Product Platforms.
control, such as floor planning, space navigation and modular design. Society of Manufacturing Engineers, Dearborn, MI (ISBN: 978-0872635142).
Feloni, R., 2014. How Lego Came Back From the Brink of Bankruptcy. 〈https://www.
Considering those applications are usually built independently and
businessinsider.com/how-lego-made-a-huge-turnaround-2014–2?r=US&IR=T〉.
supported by different systems, the development of a graph data­ (Accessed 8 March 2020).
base for module library will facilitate those applications in terms of Ferdous, W., Bai, Y., Ngo, T.D., Manalo, A., Mendis, P., 2019. New advancements,
common information management. Besides, to support module challenges and opportunities of multi-storey modular buildings – a state-of-the-
art review. Eng. Struct. 183, 883–893. https://doi.org/10.1016/j.engstruct.2019.01.
reuse in future projects, we create three application scenarios where 061
a module library can be used for design-related tasks, including BOM Gao, S., Jin, R., Lu, W., 2020. Design for manufacture and assembly in construction: a
generation, identification of impacts of design change, and similar review. Build. Res. Inf. 48, 538–550. https://doi.org/10.1080/09613218.2019.
1660608
module retrieval. Finally, a prototype system built upon web tech­ Gbadamosi, A.Q., Oyedele, L., Mahamadu, A.M., Kusimo, H., Bilal, M., Davila Delgado,
nology is developed and implemented in a timber panel system of an J.M., Muhammed-Yakubu, N., 2020. Big data for design options repository: to­
industrialized construction firm. This makes the first step towards wards a DFMA approach for offsite construction. Autom. Constr. 120, 103388.
https://doi.org/10.1016/j.autcon.2020.103388
the development of a shared, industry-wide product platform for Gosling, J., Pero, M., Schoenwitz, M., Towill, D., Cigolini, R., 2016. Defining and cate­
mass customization. gorizing modules in building projects: an international perspective. J. Constr. Eng.
Manag. 142, 1–11. https://doi.org/10.1061/(ASCE)CO.1943-7862.0001181
Hollberg, A., Genova, G., Habert, G., 2020. Evaluation of BIM-based LCA results for
CRediT authorship contribution statement building design. Autom. Constr. 109, 102972. https://doi.org/10.1016/j.autcon.
2019.102972
Jianpeng Cao: Conceptualization, Methodology, Data curation, Hor, A.E.H., Sohn, G., Claudio, P., Jadidi, M., Afnan, A., 2018. A semantic graph database
for BIM-GIS integrated information model for an intelligent urban mobility web
Writing – original draft, Software. David F. Bucher: Software, application. ISPRS Ann. Photogramm., Remote Sens. Spat. Inf. Sci. 4, 89–96.
Visualization, Writing – review & editing. Daniel Hall: Supervision, https://doi.org/10.5194/isprs-annals-IV-4-89-2018
Writing – review & editing, Funding acquisition. Mike Eggers: Hwang, B.G., Shan, M., Looi, K.Y., 2018. Knowledge-based decision support system for
prefabricated prefinished volumetric construction. Autom. Constr. 94, 168–178.
Resources.
https://doi.org/10.1016/j.autcon.2018.06.016
Isaac, S., Bock, T., Stoliar, Y., 2016. A methodology for the optimal modularization of
Declaration of Competing Interest building design. Autom. Constr. 65, 116–124. https://doi.org/10.1016/j.autcon.
2015.12.017
Isaac, S., Navon, R., 2013. A graph-based model for the identification of the impact of
The authors declare that they have no known competing fi­ design changes. Autom. Constr. 31, 31–40. https://doi.org/10.1016/j.autcon.2012.
nancial interests or personal relationships that could have appeared 11.043
Ismail, A., Nahar, A., Scherer, R., 2017. Application of graph databases and graph theory
to influence the work reported in this paper. concepts for advanced analysing of BIM models based on IFC standard. In: Digital
Proceedings of the 24th EG-ICE International Workshop on Intelligent Computing
Acknowledgements in Engineering 2017, pp. 146–57. 〈https://www.researchgate.net/publication/
318600860_Application_of_graph_databases_and_graph_theory_concepts_for_
advanced_analysing_of_BIM_models_based_on_IFC_standard〉. (Accessed 8 March
This work was supported by Swiss National Science Foundation 2020).
[Grant number: 197294]. Ismail, A., Strug, B., 2018. Building knowledge extraction from BIM/IFC data for ana­
lysis in graph databases. In: Proceedings of the International Conference on
Artificial Intelligence and Soft Computing, p. 652–64. 〈https://doi.org/10.1007/
References 978-3-319-91262-2〉.
Jansson, G., Johnsson, H., Engström, D., 2014. Platform use in systems building. Constr.
Abu-Aisheh, Z., Raveaux R., Ramel J.Y., Martineau, P., 2015. An exact graph edit dis­ Manag. Econ. 32, 70–82. https://doi.org/10.1080/01446193.2013.793376
tance algorithm for solving pattern recognition problems. In: ICPRAM 2015 – 4th Jensen, P., Olofsson, T., Smiding, E., Gerth, R., 2014. Developing products in product
International Conference on Pattern Recognition Applications and Methods, platforms in the AEC industry. In: Computing in Civil and Building Engineering
Proceedings, pp. 271–8. 〈https://doi.org/10.5220/0005209202710278〉. 2014. American Society of Civil Engineers, Reston, VA, pp. 1062–1069. https://doi.
Anderson, David M., 2003. Design for Manufacturability & Concurrent Engineering; org/10.1061/9780784413616.132
How to Design for Low Cost, Design in High Quality, Design for Lean Manufacture, Jiang, Y., Liu, X., Kang, K., Wang, Z., Zhong, R.Y., Huang, G.Q., 2021. Blockchain-enabled
and Design Quickly for Fast Production. CIM press. (ISBN: 978-1878072238). cyber-physical smart modular integrated construction. Comput. Ind. 133, 103553.
Angles, A., 2012. A comparison of current graph database models. In: Proceedings – https://doi.org/10.1016/j.compind.2021.103553
2012 IEEE 28th International Conference on Data Engineering Workshops, ICDEW Khalili, A., Chua, D.K.H., 2013. IFC-based framework to move beyond individual
2012, IEEE, pp. 171–7. 〈https://doi.org/10.1109/ICDEW.2012.31〉. building elements toward configuring a higher level of prefabrication. J. Comput.
Bertram, N., Fuchs, S., Mischke, Palter R., Strube, G., Woetzel, J., 2019. Modular con­ Civ. Eng. 27, 243–253. https://doi.org/10.1061/(ASCE)CP.1943-5487.0000203
struction: from projects to products. In: McKinsey & Company: Capital Projects & Lawson, R.M., Ogden, R.G., 2008. “Hybrid” light steel panel and modular systems.
Infrastructure, pp. 1–34. 〈https://www.ivvd.nl/wp-content/uploads/2019/12/ Thin-Walled Struct. 46, 720–730. https://doi.org/10.1016/j.tws.2008.01.042
Modular-construction-from-projects-to-products-full-report-NEW.pdf〉. Levitt, R.E., 2015. An extended coordination theory for governance of tasks, projects,
(Accessed 8 March 2020). firms, and business ecosystems., Working Paper, Global Projects Center, Stanford
Boccaletti, S., Latora, V., Moreno, Y., Chavez, M., Hwang, D.U., 2006. Complex net­ University. 〈http://gpc.stanford.edu/publications/extended-coordination-theory-
works: structure and dynamics. Phys. Rep. 424, 175–308. https://doi.org/10.1016/j. governance-tasks-projects-firms-and-business-ecosystems〉. (Accessed 7 June
physrep.2005.10.009 2019).
Bonev, M., Wörösch, M., Hvam, L., 2015. Utilizing platforms in industrialized con­ Liew, J.Y.R., Chua, Y.S., Dai, Z., 2019. Steel concrete composite systems for modular
struction: a case study of a precast manufacturer. Constr. Innov. 15, 84–106. construction of high-rise buildings. Structures 21, 135–149. https://doi.org/10.
https://doi.org/10.1108/CI-04-2014-0023 1016/j.istruc.2019.02.010
Borjesson, F., Hölttä-Otto, K., 2014. A module generation algorithm for product ar­ Liu, R., Issa, R.R.A., 2012. Automatically updating maintenance information from a BIM
chitecture based on component interactions and strategic drivers. Res. Eng. Des. database. In: Computing in Civil Engineering 2012. American Society of Civil
25, 31–51. https://doi.org/10.1007/s00163-013-0164-2 Engineers, pp. 373–380. https://doi.org/10.1061/9780784412343.0047
Cao, J., Bucher, D.F., Hall, D.M., Lessing, J., 2021. Cross-phase product configurator for Li, N., Li, Q., Liu, Y.S., Lu, W., Wang, W., 2020. BIMSeek++: retrieving BIM components
modular buildings using kit-of-parts. Autom. Constr. 123, 103437. https://doi.org/ using similarity measurement of attributes. Comput. Ind. 116, 103186. https://doi.
10.1016/j.autcon.2020.103437 org/10.1016/j.compind.2020.103186

12
J. Cao, D.F. Bucher, D.M. Hall et al. Computers in Industry 139 (2022) 103659

Lu, W., Tan, T., Xu, J., Wang, J., Chen, K., Gao, S., Xue, F., 2020. Design for manufacture Ślusarczyk, G., Łachwa, A., Palacz, W., Strug, B., Paszyńska, A., Grabska, E., 2017. An
and assembly (DfMA) in construction: the old and the new. Archit. Eng. Des. extended hierarchical graph-based building model for design and engineering
Manag. 0, 1–15. https://doi.org/10.1080/17452007.2020.1768505 problems. Autom. Constr. 74, 95–102. https://doi.org/10.1016/j.autcon.2016.11.008
Moayeri, V., Moselhi, O., Zhu, Z., 2017. Design change management using BIM-based Solihin, W., Eastman, C., Lee, Y.C., Yang, D.H., 2017. A simplified relational database
visualization model. Int. J. Archit., Eng. Constr. 6, 1–9. https://doi.org/10.7492/ schema for transformation of BIM data into a query-efficient and spatially enabled
ijaec.2017.001 database. Autom. Constr. 84, 367–383. https://doi.org/10.1016/j.autcon.2017.10.
Murtaza, M.B., Fisher, D.J., Skibniewski, M.J., 1993. Knowledge‐based approach to 002
modular construction decision support. J. Constr. Eng. Manag. 119, 115–130. Staib, G., Dörrhöfer, A., Rosenthal, M., 2008. Components and Systems Modular
https://doi.org/10.1061/(ASCE)0733-9364(1993)119:1(115) Construction – Design, Structure, New Technologies. Walter de Gruyterhttps://doi.
Nath, T., Attarzadeh, M., Tiong, R.L.K., Chidambaram, C., Yu, Z., 2015. Automation in org/10.11129/detail.9783034615662
construction productivity improvement of precast shop drawings generation Staub-French, S., Nepal, M.P., 2007. Reasoning about component similarity in building
through BIM-based process re-engineering. Autom. Constr. 54, 54–68. https://doi. product models from the construction perspective. Autom. Constr. 17, 11–21.
org/10.1016/j.autcon.2015.03.014 https://doi.org/10.1016/j.autcon.2007.02.013
Nauata, N., Chang, K.-H., Cheng, C.-Y., Mori, G., Furukawa, Y., 2020. House-GAN: re­ Strug, B., Grabska, E., Ślusarczyk, G., 2014. Supporting the design process with hy­
lational generative adversarial networks for graph-constrained house layout pergraph genetic operators. Adv. Eng. Inform. 28, 11–27. https://doi.org/10.1016/j.
generation. In: European Conference on Computer Vision, pp. 162–77. 〈https:// aei.2013.10.002
arxiv.org/abs/2003.06988〉. (Accessed 8 March 2020). Strug, B., Ślusarczyk, G., 2009. Reasoning about designs through frequent patterns
Park, J., Cai, H., 2017. WBS-based dynamic multi-dimensional BIM database for total mining. Adv. Eng. Inform. 23, 361–369. https://doi.org/10.1016/j.aei.2009.06.007
construction as-built documentation. Autom. Constr. 77, 15–23. https://doi.org/ Tauscher, E., Bargstädt, H.-J., Smarsly, K., 2016. Generic BIM queries based on the IFC
10.1016/j.autcon.2017.01.021 object model using graph theory. In: Proceedings of the 16th International
Peltokorpi, A., Olivieri, H., Granja, A.D., Seppänen, O., 2018. Categorizing modular­ Conference on Computing in Civil and Building Engineering. Article 04014046.
ization strategies to achieve various objectives of building investments. Constr. Tetik, M., Peltokorpi, A., Seppänen, O., Holmström, J., 2019. Direct digital construction:
Manag. Econ. 36, 32–48. https://doi.org/10.1080/01446193.2017.1353119 Technology-based operations management practice for continuous improvement
Pullen, T., Hall, D.M., Lessing, J., 2019. White Paper: A Preliminary Overview of of construction industry performance. Autom. Constr. 107, 102910. https://doi.
Emerging Trends for Industrialized Construction in the United States, pp. 1–24. org/10.1016/j.autcon.2019.102910
〈https://doi.org/10.3929/ethz-b-000331901〉. Thompson, J.D., Zald, M.N., Scott, W.R., 2017. Organizations in Action. Routledge (ISBN:
Quayyum, S., 2020. Parametric models for wind load resistances of wood-frame walls. 9781315125930).
J. Struct. Eng. 146, 04019188. https://doi.org/10.1061/(asce)st.1943-541x.0002471 De La Torre, M.L., Sause, R., Slaughter, S., Hendricks, R.H., 1994. Review and Analysis of
Rathle, R., 2020. Driving Innovation in Retail with Graph Technology How Top Modular Construction Practices. Lehigh University, pp. 1–152. 〈https://preserve.
Retailers Use Neo4j. 〈https://neo4j.com/whitepapers/retailers-graph-technology- lib.lehigh.edu/islandora/object/preserve%3Abp-4308350〉. (Accessed 8 March
neo4j/〉. (Accessed 6 May 2020). 2020).
Riesen, K., Fankhauser, S., Bunke, H., 2007. Speeding up graph edit distance compu­ van Treeck, C., Rank, E., 2007. Dimensional reduction of 3D building models using
tation with a bipartite heuristic. In: Mining and Learning with Graphs, pp. 21–4. graph theory and its application in building energy simulation. Eng. Comput. 23,
〈http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.114.6893&rep=rep1& 109–122. https://doi.org/10.1007/s00366-006-0053-7
type=pdf〉. (Accessed 8 March 2020). Viana, D., Tommelein, I., Formoso, C., 2017. Using modularity to reduce complexity of
Salama, T., Salah, A., Moselhi, O., Al-Hussein, M., 2017. Near optimum selection of industrialized building systems for mass customization. Energies 10, 1622.
module configuration for efficient modular construction. Autom. Constr. 83, https://doi.org/10.3390/en10101622
316–329. https://doi.org/10.1016/j.autcon.2017.03.008 Wang, X., Yang, Y., Zhang, K., 2018. Customization and generation of floor plans based
Salvador, F., 2007. Toward a product system modularity construct: literature review on graph transformations. Autom. Constr. 94, 405–416. https://doi.org/10.1016/j.
and reconceptualization. IEEE Trans. Eng. Manag. 54, 219–240. https://doi.org/10. autcon.2018.07.017
1109/TEM.2007.893996 Wong, S.S.Y., Chan, K.C.C., 2009. EvoArch: an evolutionary algorithm for architectural
Samarasinghe, T., Gunawardena, T., Mendis, P., Sofi, M., Aye, L., 2019. Dependency layout design. CAD Comput. Aided Des. 41, 649–667. https://doi.org/10.1016/j.cad.
structure matrix and hierarchical clustering based algorithm for optimum module 2009.04.005
identification in MEP systems. Autom. Constr. 104, 153–178. https://doi.org/10. Wu, S., Shen, Q., Deng, Y., Cheng, J., 2019. Natural-language-based intelligent retrieval
1016/j.autcon.2019.03.021 engine for BIM object database. Comput. Ind. 108, 73–88. https://doi.org/10.1016/j.
Sanfeliu, A., Fu, K.-S., 1983. A distance measure between attributed relational graphs compind.2019.02.016
for pattern recognition. Trans. Syst. Man Cybern. 13, 353–362. https://doi.org/10. Zawiślak, S., Rysiński, J., 2017. Graph-Based Modelling in Engineering. Springer, Cham.
1109/TSMC.1983.6313167 https://doi.org/10.1007/978-3-319-39020-8

13