Submitted Successfully!
To reward your contribution, here is a gift for you: A free trial for our video production service.
Thank you for your contribution! You can also upload a video entry or images related to this topic.
Version Summary Created by Modification Content Size Created at Operation
1
Jelena Milošević
 -- 4717 2023-10-10 12:26:55 |
2 layout & references
Sirius Huang
 Meta information modification 4717 2023-10-12 02:44:49 |

Video Upload Options

Do you have a full video?
Send video materials Upload full video

Confirm

Are you sure to Delete?
Yes No
Cite
If you have any further questions, please contact Encyclopedia Editorial Office.
Živković, M.; Žujović, M.; Milošević, J. Architectural 3D-Printed Structures Created Using Artificial Intelligence. Encyclopedia. Available online: https://encyclopedia.pub/entry/50054 (accessed on 17 May 2024).
Živković M, Žujović M, Milošević J. Architectural 3D-Printed Structures Created Using Artificial Intelligence. Encyclopedia. Available at: https://encyclopedia.pub/entry/50054. Accessed May 17, 2024.
Živković, Milijana, Maša Žujović, Jelena Milošević. "Architectural 3D-Printed Structures Created Using Artificial Intelligence" Encyclopedia, https://encyclopedia.pub/entry/50054 (accessed May 17, 2024).
Živković, M., Žujović, M., & Milošević, J. (2023, October 10). Architectural 3D-Printed Structures Created Using Artificial Intelligence. In Encyclopedia. https://encyclopedia.pub/entry/50054
Živković, Milijana, et al. "Architectural 3D-Printed Structures Created Using Artificial Intelligence." Encyclopedia. Web. 10 October, 2023.
Architectural 3D-Printed Structures Created Using Artificial Intelligence
Edit
This entry is adapted from the peer-reviewed paper 10.3390/app131910671

Artificial Intelligence (AI) and 3D printing (3DP) play considerable roles in what is known as the Fourth Industrial Revolution, by developing data- and machine-intelligence-based integrated production technologies. In architecture, this shift was induced by increasingly complex design requirements, posing important challenges for real-world design implementation, large-scale structure fabrication, and production quality standardization.

architectural design digital design digital fabrication additive manufacturing 3D printing

1. Introduction

Similar to other engineering domains, automation is increasingly playing a significant role in architecture as advancements in Artificial Intelligence (AI) streamline various processes and enhance design capabilities [1][2][3][4][5]. AI-based automation is researched in diverse aspects of architecture design, including Building Information modeling (BIM), generative design, parametric design, performance analysis and simulation, project management, documentation generation and evaluation, as well as digital fabrication. As for the latter, automation is transforming the way various architectural forms and elements are designed and manufactured. Advanced additive manufacturing (AM) technologies, including 3D printing (3DP), allow designers to implement rapid prototyping design methodologies and fabricate complex and customized components directly, reducing material waste and construction time. Moreover, automation using AI and 3DP is widely explored across different fields using different materials and production scales [6][7], including metal 3DP technology for aerospace and mechanical industries [8][9], parts manufactured in composite materials using photopolymer 3DP [10], as well as biomedical materials advancement through AI-assisted 3D and 4D printing technology explorations [11]. In all the mentioned fields, robotic fabrication systems can automate tasks that are traditionally time- and effort-consuming, enabling efficient and precise construction processes [12][13][14][15][16][17][18]. It is important to note that, while automation can enhance production speed and creativity in architectural design, human expertise remains indispensable in the decision-making process.
Exploring the synergetic potential of AI and AM in architecture could advance both technologies and push the boundaries of what is possible in architectural design and construction. Studies on this topic are important because they aid technological improvements that support advances in design, optimization, customization and personalization, performance-driven explorations, and resource allocation towards increased efficiency and sustainability and facilitate industry applications. The integration of AI in the 3DP process promises to overcome specific challenges posed by AM technology. For example, reinforced concrete AM faces challenges regarding the accuracy of material placement during printing, phase transition control and measurements, cold joint formation during the layering process, reinforcement implementation, and surface finish [19]. Given that production usually begins after the design process is mostly or fully completed, 3DP might be viewed as inefficient because multiple iterations of the printing process are needed to produce the desired structure. Therefore, AI systems have been explored to overcome these challenges through real-time control of the printing parameters or autonomous adjustment of the printing process.

2. AI Techniques in Architectural 3D-Printed Structures

ML is a computer science field that relates to the algorithms that learn how to solve complex real-world problems from given datasets. The most common problems met by ML include classification, clustering, and prediction [20]. The generally accepted types of ML, distinguished by significant functional differences, include (1) Supervised Learning (SL), which uses input data for identification and fulfillment of certain tasks, among which classification and regression are the most well-known; (2) Unsupervised Learning (UL), which is mostly concerned with identifying groups and organization patterns within unlabeled datasets, with the common task of clustering; (3) Semi-Supervised Learning (SSL), which learns from both labeled and unlabeled data, commonly used for classification and clustering purposes; and (4) Reinforcement Learning (RL), which functionality lies within a reward–punish system where the algorithm automatically evaluates the best behavior patterns and takes further actions to optimize the system [21]. Deep learning (DL) is a subset of ML with key features represented in numerous layers or stages of nonlinear information processing and supervised (SDL) or unsupervised (UDL) feature representation at progressively higher, more abstract layers [22]. Defining ANN in the scientific and practical domains remains challenging [23]. The architecture and functionality of artificial neurons forming ANNs are based on biological neurons. These networks function by processing information in their fundamental constituents—artificial neurons—in a non-linear, distributed, parallel, and local manner [24]. Lastly, computer vision represents a technical system whose primary goal is to mimic the functional modules of human vision, which is achieved through tasks including visualization, image formation, control of irradiance, focusing, irradiance resolution, tracking, and processing and analysis [25].
Out of the twenty-one reviewed studies, ML is found to be the main topic of six studies, with applications ranging from design, optimization, and diagnostic purposes for 3DP structures [26][27][28][29][30][31]. ANN as a category is found to be dominantly present in nine studies overall, with applications belonging to the diagnostic domain [32][33][34][35][36][37][38][39][40]. The combined appearance of ML and ANN is dominantly found in a total of four studies, where the use is focused around optimization and diagnostics [41][42][43][44]. Lastly, computer vision algorithms are present solely in the diagnostic domain and cover a total of four papers [27][35][45][46]. Figure 1 displays the identified AI techniques, their domains of application, and the related references.
Figure 1. Overview of AI techniques and main applications in 3DP in architecture [26][27][28][29][30][31][32][33][34][35][36][37][38][39][40][41][42][43][44][45][46].

3. Applications

Topics identified through the bibliometric analysis, including AI-driven (1) design, (2) optimization, and (3) diagnostics of architectural 3D-printed structures, are analyzed in depth in this subsection (Figure 2). Within each topic, the following aspects were reviewed:
Figure 2. Overview of the main identified uses for AI in large-scale 3DP.
  • general uses of different AI algorithms in a specific application domain;
  • challenges of large-scale structures 3DP overcomes by integration of AI;
  • modifications of AI algorithms made to suit a specific application domain.

Topic 1. AI-Driven Design of 3D-Printed Architectural Structures

In the reviewed articles, AI integration in the 3DP design process is mainly focused on the issues of material design and design optimization of the 3DP mixes. The previous is confirmed by the fact that 75% of the represented papers deal with these specific topics. Wang et al. [31] deliver a systematic review of the emerging digital technologies used for off-site construction leading towards Industry 4.0, where among the fifteen reviewed technologies, AI and 3DP are elaborated on separately, each representing an important factor for the future development of the construction sector. On the other hand, Nguyen-Van et al. [28] review the development of predictive modeling and design optimization and the current state of the art specifically for concrete 3DP, where four characteristic steps for the implementation of ML into the 3DP are given, including the problem definition, development of the model, collection of data, and process settings. Moreover, Geng et al. [26] deliver a review of the latest research on integrating ML into construction 3DP, with an extensive discussion of current problems and future trends in the field. In another study, Tan [27] proposes a framework for AI and 3DP combinations in five specific aspects, including 3DP materials, automation design, digital construction, 3DP robots, and a 3DP BIM platform. The overview of the main findings of AI models in 3DP for designing large-scale structures is given in Table 1.
Table 1. Overview of the specific techniques, applications, and research conclusions of AI-driven design of 3D-printed architectural structures.
Applications Techniques Main Conclusions Author(s)
References
Design for Additive Manufacturing (DfAM) of prefabricated architectural components. ML
  • Benefits: reduced effort and cost in the design process; components’ segmentation speed is increased.
  • Issues: time-consuming topology analysis preparation process; limitations in the components’ configuration design.
 Wang et al. [31]
Rapid design method development for product modeling and its structure selection. ML
  • Future research: exploring combinations of novel technologies such as BIM and AI.
 Tan [27]
Design problem definition aimed at AM. ML
  • Challenge: handling large amounts of data with high computational costs.
  • Digital twins of AM structures could support the ML techniques in attaining better outcomes.
  • ML included in tasks such as topology optimization, bio-inspired concept creation, and geometry creation using Computer-Aided Design.
 Nguyen-Van et al. [28]
AM material design with the real-time observation and automatic mixture alterations during the printing process. ML
  • ML application in the matching ratios’ selection can improve production efficiency and reduce construction costs.
  • Future studies: the integration of different research areas presenting specific challenges in the construction 3DP.
 Geng et al. [26]
The main challenges found in the studies include the time-consuming process and handling of large amounts of data in ML models [28][31]. Regarding the ML application in 3D concrete printing (3DCP), the challenge that may arise relates to data interpretability (specifically related to correlation-based ML models) and validation, since supervised regression models are one of the most commonly used in these types of cases. Having that in mind, for the valid application of ML, it is of utmost importance for the dataset to be correct, complete, and representative of a large data population, which is, apart from being time-consuming, also a costly process [28]. In the large-scale 3DP process, the challenges that could be met by AI include predicting and controlling material anisotropy, avoiding uneven pore distribution caused by the lack of fusion inside the material, and improving the warping behavior of the structure due to residual stresses caused by the rapid cooling characteristics of the 3DP process. In these cases, AI can be trained to learn certain principles and methods of structural design through ML models, allowing it to handle complex scenarios in the 3DP process in an efficient, intelligent, and environmentally conscious way [26].
The functional characteristics of AI models used for 3DP, specifically ML, remain true to their core logic, as explained in the introductory part of the Content Analysis section. The difference that arises is seen in the datasets that the ML models are trained on and the roles they are designed to embody. Specifically, ML is utilized for material design and property formulation to achieve optimal and desired targets [28]. Further, in the four registered stages of modeling for digital concrete fabrication, ML is introduced in the last stage, after the analytical modeling, experiments for the input data creation, and numerical simulations, contributing to the process with the ability to create a virtual 3DP simulation that could inform the design process [28]. Concerning different tasks set out for the AI algorithms, the models may vary, and researchers could select the optimal ML type regarding its future application in the 3DP process. This relatively subjective course of action could lead to an increased chance of poor ML training performance. The smoothness of the ML model training could be enhanced through the existence of a shared platform that contains the previously gathered experience from different ML models, since it can learn how to avoid making the same mistakes [26].

Topic 2. AI-Driven Optimization of 3D-Printed Architectural Structures

Among the papers that explore ML applications for the optimization of 3DP, the study by Parisi et al. [30] delivers an approach towards intelligent AI-controlled tower crane 3DP with optimization of the extruder toolpath. On the other hand, Wang et al. [31] explore ML-based optimization targeted at specific construction tasks in off-site construction applications. Moreover, Nguyen-Van et al. [28] present an overview of the current state-of-the-art development of modeling and design optimization tools for 3DCP, highlighting the possibility of reducing printing time, improving structural performance, or allowing for the adaptation to printing structures with complicated geometries by using an intelligent toolpath-generating algorithm. Additionally, this study highlights the intricate relationships between ML and ANN models, as shown in the analyzed papers. As a result, both AI categories have been included in multiple articles, particularly in the realm of optimization problems. The study by Baduge et al. [42] highlights the use of ANN, ML, and DL algorithms and their applications in various domains of the construction process, pointing out the advantages of their integration into 3DP, which leads to optimized solutions through the increased level of automation, more advanced robots, and geometrical flexibility of the structures. Lao et al. [43] explored optimized nozzle shapes for delivering high-quality surface finishes in varying types of 3DP structures’ geometry. A combination of ML and ANN is found in material design, specifically in forming the optimal mixture for the 3DP process depending on the desired outcome, as presented by Fan et al. [41]. Table 2 summarizes the key features of the optimization-related studies.
Table 2. Overview of specific techniques, applications, and research conclusions of AI-driven optimization of 3D-printed architectural structures.
Applications Techniques Main Conclusions Author(s)
References
Formation of the cloud-based 3DP system that optimizes and enhances the printing process and identifies collision-free tool path; optimum geometry partitioning and material distribution optimization. ML, DL, ANN
  • Advantages: the generation of more optimized AM solutions.
  • Future trends: 4D printing.
  • AI-enabled collaborative robots will gain market in the 3DP industry, allowing the 3DP process to be configured during the printing with the material alterations.
 Baduge et al. [42]
Formation of a tower crane (TC) 3DP AI agent which dynamically activates the TC freedom degrees to minimize the swing effect, simultaneously maximizing the printing speed. DRL
TD3 architecture
  • Deep Reinforcement Learning (DRL) provides optimal 3DP control solutions, without strictly needing any formal and mathematical systems model.
  • DRL is effective in the case of complex and non-linear systems, including real-time image processing-led control.
 Parisi et al. [30]
Printing toolpath optimization for avoiding under and overfilling issue; printing mixture optimization. ML
  • Printing time is reduced, enhanced structural performance, and optimized complex shapes-adapted printing.
  • ML fixes the issue of the time-consuming process of the Finite Element Method (FEM) approach in the 3DCP, it aids the printing mix design, as well as assists in the modeling and optimization for the 3DCP process.
 Nguyen-Van et al. [28]
Improvement of the efficiency and accuracy of other technologies, one of them being AM. ML
  • Issues: time-consuming dataset preparation and the lack of real practice validation.
 Wang et al. [31]
Optimizing the mixture design of Ultra-High-Performance Concrete (UHPC). ML, back-propagation (BP) ANN, genetic algorithm BPNN (GA-BPNN)
  • Intelligent optimization has not been researched in domains such as UHPC micro-characteristics.
  • Future trends: more research should be done on the design theory of 3DP-UHPC to build a relation among material composition, performance, and applications.
 Fan et al. [41]
Finding a proper nozzle shape for production of designated extrudate geometries. ML, ANN
  • The proposed methodology improves the surface quality of the structures with different curvatures.
  • The proposed approach has the potential to improve the surface quality for other types of 3DP structures.
 Lao et al. [43]
Defined as one of the tasks of intelligent systems, optimization is a process executed using various AI algorithms. Specific challenges presented by the optimization tasks in 3DP include their application in the practical domain, where AI is still mostly limited to checking printability and modularization for prefabrication techniques [42]. Challenges posed include those related to the possibilities of the ML algorithms, such as one described by Lao et al. [43], which includes the non-invertible relationship between the targeted extrudate cross-sectional shape in the experiments and the nozzle shape in the ML model. The authors conclude that this challenge is still not overpowered by the benefits that the ML integration into the 3DP process creates, such as the enhanced overall efficiency for achieving extrudate control in practice [43]. The main challenge faced by the DRL method in the study by Parisi et al. [30] includes the creation of the control system, which produces an effective extruding toolpath for the 3D printer. Another common challenge in practice involves the higher geometric complexity that is derived from the optimized topologies. This issue has the potential to be resolved by generating an innovative toolpath, which would ensure the optimized concrete structures’ continuous printing [28]. Challenges faced in the UHPC domain include the inability to utilize common models of AI for the purpose of mixture design, since they produce low-accuracy results [41].
The algorithms that are used for the optimization tasks vary in their basic functionality and computational cost. For example, a novel AI framework was given by Parisi et al. [30], which is the core of the proposed extrusion-based 3DP system. The basic logic of the approach is an intelligent DRL agent that dynamically activates the tower crane’s degrees of freedom to minimize the extruder swing effect while maintaining maximum printing speed. The main inputs for the AI-controlled system contain the dynamic environment characteristics, the possible actions for the agent to take, the reward function, and the agent modeling with its learning algorithm. The specific type of algorithm is the twin-delayed deep deterministic policy gradient (TD3), which is suitable for models characterized by continuous action spaces [30]. The algorithms used for the optimized AM mixture design include the ones described by Fan et al. [41], where the back-propagation ANN was used to model the mixtures (ANN input), as well as the compressive strength and workability (ANN output) parameters. The algorithm was trained and tested on 53 different concrete mixtures. Another type of algorithm that was used for the AM-optimized mixture design is the Genetic Algorithm back-propagation neural network (GA-BPNN). The procedure involving this AI technique included creating a GA-BPNN prediction model based on obtained training datasets, determining the initial mixture of the UHPC in accordance with the ingredients’ boundaries, inputting the initial mixture for property prediction by the developed network, identifying the predicted results, and redesigning the initial mixture until the desired requirements are met [41]. The ML model workflow introduced by Lao et al. [43] included four steps: (1) pre-testing, with the setting up of the experiment with different nozzle shapes and conducting the experiment; (2) ANN training, which involved the optimization of the ANN topology, training the model with pre-testing results, and experimentally validating the ANN model; (3) building up the database, with the generation of enough volume of random nozzles and predicting the corresponding extrudate shapes with the ANN; and (4) target printing, which involved analyzing the target extrudate shape, finding the nozzle shape in the database, and conducting the printing with the nozzle [43].

Topic 3. AI-Driven Diagnostics of 3D-Printed Architectural Structures

The reviewed articles include several approaches towards the diagnostic tasks for 3DP, based on (1) ML-driven prediction, simulation, and inspection; (2) ANN-based inspection and prediction; (3) combined use of ML and ANN for prediction; and (4) computer vision technology used for inspection and quality-monitoring purposes.
A variety of ML technologies are utilized for predictive and inspection purposes, ranging from SL, UL, SSL, and RL, with applications explored in the domains of material design optimization, control printing accuracy, printing defect detection and classification, state differentiation of the printing process, anisotropic behavior analysis, printing product classification in relation to the deformations, printing cost estimation, compensation of printing material deformation, printing process planning correction, large-scale printing product customization, and others, as presented by Geng et al. [26]. Additionally, Chen et al. [29] introduce a deep-learning module of the Dragonfly 3.6 software to extract the axes of the steel fibers in the X-ray micro-computed tomography (X-CT) images of 3DP concrete samples and evaluate their 3D orientational distribution statistics.
ANN models used for the inspective tasks in 3DP rely on several algorithm types, which differ in functionality and purpose. Yao et al. [34] explore the effect of steam curing conditions on the performance properties of 3DP materials at various ages of curing, using a specific set of algorithms to predict the performance of the material. Other types of applications are found, mostly related to the prediction of the tensile and compressive strengths of the researched materials [36][37][39]. Additionally, Rossi et al. [32] deal with modeling the curing conditions of cellulose-based 3DP components using a defined set of ANN models. The relationship between ANN and ML is intertwined, leading many researchers to employ both terms. Among the papers presented in this subgroup, two papers represent reviews of the current research status [41][44], whereas one paper presents an original methodology for finding the proper nozzle shape in the 3DP process [43].
Computer vision algorithms are seen as a promising method for assessing real-time 3DP process tracking, where the data collection usually consists of a camera being installed on the extruder to capture videos and images during the printing process. Among the four included papers, two represent frameworks for integrating computer vision technologies in the 3DP process [27][35], whereas the other two papers develop novel methods for this purpose. Kazemian et al. [46] develop a vision-based real-time extrusion quality-monitoring system for robotic construction. In another paper, four techniques for inline real-time extrusion quality monitoring during construction are given [45].
An overview of the key research aspects is given in Table 3.
Table 3. Overview of the specific techniques, applications, and research conclusions of AI-driven diagnosis of 3D-printed architectural structures.
Applications Techniques Main Conclusions Author(s)
References
Automatic robotic detection of the printing defects; printing parameters reconfiguration in real-time. ML
Computer
vision
  • AI and BIM provide useful methods for the issues which arise in the large-scale 3DP.
 Tan [27]
Extraction and 3D analysis of the centerlines of steel fibers in the X-ray micro-computed tomography image sequence. DL
U-Net module
  • U-Net is a newly approved neural network in the ML field where the computer is allowed to segment according to the semantics of the images.
 Chen et al. [29]
The automatic image segmentation of the 3DP fiber-reinforced materials. DCNN
U-Net module
  • Data augmentation procedure eliminates the time-consuming task of the manual annotation.
  • The automated image segmentation method is suitable for the efficient identification of complex micro-structures.
 Nefs et al. [33]
Real-time layer extrusion monitoring during 3DCP. Computer
vision
DCNN
  • The quality control methods should include easy-to-perform tests at a high frequency.
 Mechtcherine et al. [35]
Detection of bending deformations in the 3DP layers, during or after the printing process. DCNN
  • The developed model only defects bending of the layers. Other concerns are not addressed.
  • The performance of the system under an actual construction site environment should be verified.
  • The potential of computer vision and DL for automated inspection, quality control, and progress monitoring during 3DP is seen.
 Davtalab et al. [38]
Detection of the extruded layer and measurement of the layer width in real-time. Automatic adjustment of the material deposition rate. Computer
vision
  • The proposed method does not need sample preparation.
  • The method is limited to laboratory environment, without known outcomes if it were tested on a real construction site.
  • High precision and responsiveness of the developed extrusion monitoring system under the experimental conditions.
 Kazemian et al. [46]
Real-time quality-monitoring system which detects variations in the material properties before the deposition. Computer
vision
  • The vision-based techniques have the highest precision and responsiveness to material variations, compared to other methods.
 Kazemian & Khoshnevis
[45]
Panel performance prediction and the printing toolpath predictive generation. cGAN
  • The encoding and result parsing within Grasshopper makes the NN work as an immediate design tool for the user.
  • NN trained on digitally compiled datasets have potential to integrate analytic and predictive information into the real-time design and fabrication processes.
 Nicholas et al. [40]
Performance prediction for 3DP concrete. BAS method is used for the ANN hyperparameter adjustment, along with the cross validation which solves the ANN overfitting problem. ANN, ML
Beetle
antennae search (BAS)
Cross
validation
  • The construction of a suitable ML model with high precision and dependability is laborious and time-consuming.
  • The hardening performance of the material under the same printing parameters is maximized.
 Yao et al. [34]
Prediction of the fresh properties of cementitious materials, such as yield stress and mini-slump. ANN
  • ANN was successfully trained to be able to predict the yield stress as a function of the percentage in weight of cement of each admixture.
  • Future research: involving the time factor in the ANN to describe the evolution of the yield stress during printing.
 Charrier & Ouellet-Plamondon [37]
Quality control of the fused deposition modeling (FDM) printing technology, predicting the ultimate tensile strength, and optimizing the printing and material parameters. ANN
  • The model is a credible guideline for designers and researchers to manufacture FDM of optimal mechanical properties.
  • The developed ANN model accurately predicts the UTS of FRP.
  • Future activities: creation of an open-source tool for users of FDM 3D printers.
 Alhaddad et al. [36]
Prediction of the 3DP concrete compressive strength. ANN
  • Difficulties: the accuracy of the model depends on the number of patterns.
  • Limited amount of research on patterns concerning 3DP concrete’s compressive strength.
  • For accurate ANN model creation, one hidden layer in its structure is enough.
 Izadgoshasb et al. [39]
Properties prediction of the Ultra-High Performance Concrete (UHPC) such as mechanical strength, flowability, filling capacity and segregation. ML, ANN
  • The complexity of the UHPC system results in insufficient predicting accuracy when using common models of AI.
  • There are not enough studies on the topic of the micro-structures of the UHPC.
 Fan et al. [41]
Statistical modeling of the curing of cellulose-based 3DP components. ANN
  • The dataset size is a constant constraint for getting good predictions out of physically generated datasets.
 Rossi et al. [32]
CNN is utilized for optimal proportions of 3DP products prediction; SL is utilized for printing products’ geometric deviations prediction; UL is utilized for 3DP material porosity prediction; RL is utilized for print trajectory geometry prediction and planning. ML
  • ML techniques are used for printing concrete filament geometrical shapes prediction, and automatic detection of layer deformations.
  • ML is used for the automatic detection of surface defects; ANN is used for predicting crack patterns and stress-crack width curves of pore structures in 3DP.
  • Future research: interlayer bond performance, real-time status monitoring, anisotropic behavior control.
 Geng et al. [26]
DfAM, including geometry deviation prediction, material analytics, prediction of defect and others. ML, ANN
  • Although many ML methods for DfAM have been researched in various applications, only several research programs have been conducted in the construction industry.
 Tuvayanond & Prasittisopin [44]
Development of a predictive model for extrudate geometry in 3DCP. ML, ANN
  • A predictive model was developed to correlate the nozzle shape and the extrudate shape counterpart using ANN.
 Lao et al. [43]
The main issue in which DCNN was utilized as an effective solution, as presented by Nefs et al. [33], is the manual annotation of micro-structural objects in 3DP strain-hardening cementitious composite (SHCC) materials, which would be an extremely time- and effort-consuming task because the fibers may be oriented in arbitrary directions and never positioned in the same plane. This issue would arise if the typical ML algorithm was used instead of the DCNN. Instead, a new methodology is presented in which the image segmentation of fiber-reinforced materials is performed with automated annotations of physical sample data [33]. One of the common large-scale 3DP issues that has been addressed is the fact that in the robotic construction process, which could take as little as one day to complete, there is not enough time for manual inspections to resolve problems. The reason for the high-speed construction is seen in the structural benefits that arise from it, such as enhanced interlayer bonding. Therefore, to monitor the printing process, these systems should ideally be equipped with an automated process inspection system that documents all relevant printing parameters [38]. Another set of 3DP-related issues is the assessment of material characteristics that impact the 3DP process, such as pumpability, yield stress, viscosity, and cement hydration, for which various scholars have focused their research on ML and ANN [37]. Specifically regarding UHPC, one of the main issues involves the integration of fibers, which could impact structures in a positive or negative manner, depending on their amount and distribution. Therefore, to maximize the positive role of fibers, a multi-disciplinary approach is necessary, including material science, mechanics, and intelligent manufacturing [41]. Another common issue in large-scale 3DP production relates to surface quality problems, such as the jagged surface or staircase effect on the 3DP object. This issue has been acknowledged and introduced to specific ANN models to affect the nozzle shape during 3DP [43].
Among the studies in the diagnostic domain, the Deep Convolutional Neural Network (DCNN) has been explored in three out of six papers [33][35][38], representing the algorithms that excel in processing organized arrays of data, such as images, as their primary functional characteristic [47]. Another explored type of ANN is the Conditional Generative Adversarial Network (cGAN) [40], whose main characteristic is the fact that it constitutes a pair of networks, known as the forger/generator and expert/discriminator, which compete in the parallel training process [48]. Further, the ANN combined with the Backpropagation Learning model was used together with the Artificial Bee Colony optimization algorithm, which creates an effective feed-forward model for the inspection and prediction of the 3DP printing and material parameters [36]. Similarly, specific types of ANNs are employed to predict the characteristics of 3DP structures. In the case presented by Nefs et al. [33], the dataset for the DCNN training was created by scanning a basic specimen composed of a single, prestressed yarn of fiber surrounded by a cementitious matrix with air voids. Later, the scan was divided into smaller windows, whose dimensions corresponded to those at which fibers in real fiber-reinforced materials appear straight. To generalize the algorithm, a data augmentation procedure was employed, where the windows of the obtained scan were rotated along the three axes at arbitrary angles, which allowed the DSCNN to train for the segmentation of arbitrarily oriented fiber samples. The DCNN network was constructed using Python, with the utilization of the Tensorflow and Keras packages [33].

References

  1. As, I.; Basu, P. The Routledge Companion to Artificial Intelligence in Architecture, 1st ed.; Basu, P., As, I., Eds.; Routledge: Oxford, UK, 2021; ISBN 978-0-367-74959-0.
  2. Bernstein, P. Machine Learning: Architecture in the Age of Artificial Intelligence; RIBA Publishing: London, UK, 2022; ISBN 978-1-00-060068-1.
  3. Del Campo, M. Neural Architecture: Design and Artificial Intelligence, 1st ed.; G—Reference, Information and Interdisciplinary Subjects Series; Applied Research & Design: San Francisco, CA, USA, 2022; ISBN 978-1-951541-68-2.
  4. Chaillou, S. Artificial Intelligence and Architecture: From Research to Practice; Birkhauser Verlag GmbH: Basel, Switzerland; Boston, MA, USA, 2022; ISBN 978-3-0356-2400-7.
  5. Leach, N. Architecture in the Age of Artificial Intelligence: An Introduction to AI for Architects, 1st ed.; Architecture in the Age of Art; Bloomsbury Academic: London, UK, 2022; ISBN 978-1-350-16552-6.
  6. Fu, K.; Yao, Y.; Dai, J.; Hu, L. Progress in 3D Printing of Carbon Materials for Energy-Related Applications. Adv. Mater. 2017, 29, 1603486.
  7. Nguyen, N.; Park, J.G.; Zhang, S.; Liang, R. Recent Advances on 3D Printing Technique for Thermal-Related Applications. Adv. Eng. Mater. 2018, 20, 1700876.
  8. Syuhada, A.; Shamsudin, M.S.; Omar, M.F.; Ghoshal, S.K.; Harun, S.W.; Aziz, M.S. Incorporating 3D Metal Printing with Artificial Intelligence in Meeting Aerospace Demands. JPCS 2021, 1892, 012015.
  9. Gupta, M. 3D Printing of Metals. Metals 2017, 7, 403.
  10. Cherevko, A.I.; Denisov, G.L.; Nikovskii, I.A.; Polezhaev, A.V.; Korlyukov, A.A.; Novikov, V.V. Composite Materials Manufactured by Photopolymer 3D Printing with Metal-Organic Frameworks. Russ. J. Coord. Chem. 2021, 47, 319–325.
  11. Pugliese, R.; Regondi, S. Artificial Intelligence-Empowered 3D and 4D Printing Technologies toward Smarter Biomedical Materials and Approaches. Polymers 2022, 14, 2794.
  12. Brell-Cokcan, S.; Braumann, J. Rob|Arch 2012: Robotic Fabrication in Architecture, Art and Design; Springer: Vienna, Austria, 2013; ISBN 978-3-7091-1465-0.
  13. Claypool, M.; Retsin, G.; Jimenez, M.; Garcia, M.J.; Soler, V. Robotic Building: Architecture in the Age of Automation; DETAIL Special; Detail Business Information GmbH: München, Germany, 2019; ISBN 978-3-95553-424-0.
  14. Habibi, S. Building Automation and Digital Technologies; Woodhead Publishing Series in Civil and Structural Engineering; Elsevier Science: Amsterdam, The Netherlands, 2022; ISBN 978-0-12-822129-7.
  15. Jebelli, H.; Habibnezhad, M.; Shayesteh, S.; Asadi, S.; Lee, S.H. Automation and Robotics in the Architecture, Engineering, and Construction Industry; Springer International Publishing: Cham, Switzerland, 2022; ISBN 978-3-030-77163-8.
  16. Raphael, B. Construction and Building Automation: From Concepts to Implementation, 1st ed.; Routledge: Oxfordshire, UK, 2022; ISBN 978-1-00-316562-0.
  17. Willmann, J.; Block, P.; Hutter, M.; Byrne, K.; Schork, T. Robotic Fabrication in Architecture, Art and Design 2018: Foreword by Sigrid Brell-Çokcan and Johannes Braumann, Association for Robots in Architecture; Springer International Publishing: Berlin/Heidelberg, Germany, 2018; ISBN 978-3-319-92294-2.
  18. Yuan, P.F.; Xie, Y.M.; Leach, N.; Yao, J.; Wang, X. Architectural Intelligence: Selected Papers from the 1st International Conference on Computational Design and Robotic Fabrication (CDRF 2019); Springer: Singapore, 2020; ISBN 9789811565687.
  19. Wangler, T.; Lloret, E.; Reiter, L.; Hack, N.; Gramazio, F.; Kohler, M.; Bernhard, M.; Dillenburger, B.; Buchli, J.; Roussel, N.; et al. Digital Concrete: Opportunities and Challenges. RILEM Tech. Lett. 2016, 1, 67–75.
  20. Rebala, G.; Ravi, A.; Churiwala, S. Machine Learning Definition and Basics. In An Introduction to Machine Learning; Springer International Publishing: Cham, Switzerland, 2019; pp. 1–17. ISBN 978-3-030-15728-9.
  21. Sarker, I.H. Machine Learning: Algorithms, Real-World Applications and Research Directions. SN Comput. Sci. 2021, 2, 160.
  22. Deng, L. Deep Learning: Methods and Applications. Found. Trends Signal Process. 2014, 7, 197–387.
  23. Guresen, E.; Kayakutlu, G. Definition of Artificial Neural Networks with Comparison to Other Networks. Procedia Comput. Sci. 2011, 3, 426–433.
  24. Krenker, A.; Bester, J.; Kos, A. Introduction to the Artificial Neural Networks. In Artificial Neural Networks—Methodological Advances and Biomedical Applications; Suzuki, K., Ed.; InTech: London, UK, 2011; ISBN 978-953-307-243-2.
  25. Computer Vision and Applications: A Guide for Students and Practitioners; Jähne, B.; Haussecker, H. (Eds.) Academic Press: San Diego, CA, USA, 2000; ISBN 978-0-12-379777-3.
  26. Geng, S.; Luo, Q.; Liu, K.; Li, Y.; Hou, Y.; Long, W. Research Status and Prospect of Machine Learning in Construction 3D Printing. Case Stud. Constr. Mater. 2023, 18, e01952.
  27. Tan, K. The Framework of Combining Artificial Intelligence and Construction 3D Printing in Civil Engineering. In Proceedings of the MATEC Web of Conferences, Chengdu, China, 13–15 April 2018; Lim, C., Zhu, X., Eds.; EDP Sciences: Chengdu, China, 2018; Volume 206.
  28. Nguyen-Van, V.; Li, S.; Liu, J.; Nguyen, K.; Tran, P. Modelling of 3D Concrete Printing Process: A Perspective on Material and Structural Simulations. Addit. Manuf. 2023, 61, 103333.
  29. Chen, Y.; Zhang, Y.; Pang, B.; Wang, D.; Liu, Z.; Liu, G. Steel Fiber Orientational Distribution and Effects on 3D Printed Concrete with Coarse Aggregate. Mater. Struct. 2022, 55, 100.
  30. Parisi, F.; Sangiorgio, V.; Parisi, N.; Mangini, A.; Fanti, M.; Adam, J. A New Concept for Large Additive Manufacturing in Construction: Tower Crane-Based 3D Printing Controlled by Deep Reinforcement Learning. Constr. Innov. 2023.
  31. Wang, M.; Wang, C.C.; Sepasgozar, S.; Zlatanova, S. A Systematic Review of Digital Technology Adoption in Off-Site Construction: Current Status and Future Direction towards Industry 4.0. Buildings 2020, 10, 204.
  32. Rossi, G.; Chiujdea, R.; Hochegger, L.; Lharchi, A.; Harding, J.; Nicholas, P.; Tamke, M.; Thomsen, M. Statistically Modelling the Curing of Cellulose-Based 3d Printed Components: Methods for Material Dataset Composition, Augmentation and Encoding. In Proceedings of the RLUK—Research Libraries UK, Virtual, 22–24 March 2023; Gengnagel, C., Baverel, O., Betti, G., Popescu, M., Thomsen, M., Wurm, J., Eds.; RLUK: London, UK, 2023; pp. 487–500.
  33. Nefs, K.; Menkovski, V.; Bos, F.; Suiker, A.; Salet, T. Automated Image Segmentation of 3D Printed Fibrous Composite Micro-Structures Using a Neural Network. Constr. Build. Mater. 2023, 365, 130099.
  34. Yao, X.; Lyu, X.; Sun, J.; Wang, B.; Wang, Y.; Yang, M.; Wei, Y.; Elchalakani, M.; Li, D.; Wang, X. AI-Based Performance Prediction for 3D-Printed Concrete Considering Anisotropy and Steam Curing Condition. Constr. Build. Mater. 2023, 375, 130898.
  35. Mechtcherine, V.; van Tittelboom, K.; Kazemian, A.; Kreiger, E.; Nematollahi, B.; Nerella, V.; Santhanam, M.; de Schutter, G.; Zijl, G.; Lowke, D.; et al. A Roadmap for Quality Control of Hardening and Hardened Printed Concrete. Cem. Concr. Res. 2022, 157, 106800.
  36. Alhaddad, W.; He, M.; Halabi, Y.; Almajhali, K. Optimizing the Material and Printing Parameters of the Additively Manufactured Fiber-Reinforced Polymer Composites Using an Artificial Neural Network Model and Artificial Bee Colony Algorithm. Structures 2022, 46, 1781–1795.
  37. Charrier, M.; Ouellet-Plamondon, C. Artificial Neural Network for the Prediction of the Fresh Properties of Cementitious Materials. Cem. Concr. Res. 2022, 156, 106761.
  38. Davtalab, O.; Kazemian, A.; Yuan, X.; Khoshnevis, B. Automated Inspection in Robotic Additive Manufacturing Using Deep Learning for Layer Deformation Detection. J. Intell. Manuf. 2022, 33, 771–784.
  39. Izadgoshasb, H.; Kandiri, A.; Shakor, P.; Laghi, V.; Gasparini, G. Predicting Compressive Strength of 3D Printed Mortar in Structural Members Using Machine Learning. Appl. Sci. 2021, 11, 10826.
  40. Nicholas, P.; Rossi, G.; Williams, E.; Bennett, M.; Schork, T. Integrating Real-Time Multi-Resolution Scanning and Machine Learning for Conformal Robotic 3D Printing in Architecture. Int. J. Archit. Comput. 2020, 18, 371–384.
  41. Fan, D.; Zhu, J.; Fan, M.; Lu, J.; Chu, S.; Dong, E.; Yu, R. Intelligent Design and Manufacturing of Ultra-High Performance Concrete (UHPC)—A Review. Constr. Build. Mater. 2023, 385, 131495.
  42. Baduge, S.; Thilakarathna, S.; Perera, J.; Arashpour, M.; Sharafi, P.; Teodosio, B.; Shringi, A.; Mendis, P. Artificial Intelligence and Smart Vision for Building and Construction 4.0: Machine and Deep Learning Methods and Applications. Autom. Constr. 2022, 141, 104440.
  43. Lao, W.; Li, M.; Wong, T.N.; Tan, M.J.; Tjahjowidodo, T. Improving Surface Finish Quality in Extrusion-Based 3D Concrete Printing Using Machine Learning-Based Extrudate Geometry Control. Virtual Phys. Prototyp. 2020, 15, 178–193.
  44. Tuvayanond, W.; Prasittisopin, L. Design for Manufacture and Assembly of Digital Fabrication and Additive Manufacturing in Construction: A Review. Buildings 2023, 13, 429.
  45. Kazemian, A.; Khoshnevis, B. Real-Time Extrusion Quality Monitoring Techniques for Construction 3D Printing. Constr. Build. Mater. 2021, 303, 124520.
  46. Kazemian, A.; Yuan, X.; Davtalab, O.; Khoshnevis, B. Computer Vision for Real-Time Extrusion Quality Monitoring and Control in Robotic Construction. Autom. Constr. 2019, 101, 92–98.
  47. Xu, Y.; Zhang, H. Convergence of Deep Convolutional Neural Networks. Neural Netw. 2022, 153, 553–563.
  48. Creswell, A.; White, T.; Dumoulin, V.; Arulkumaran, K.; Sengupta, B.; Bharath, A.A. Generative Adversarial Networks: An Overview. IEEE Signal Process. Mag. 2018, 35, 53–65.
More
© Text is available under the terms and conditions of the Creative Commons Attribution (CC BY) license; additional terms may apply. By using this site, you agree to the Terms and Conditions and Privacy Policy.
Upload a video for this entry
Information
Subjects: Construction & Building Technology
Contributors MDPI registered users' name will be linked to their SciProfiles pages. To register with us, please refer to https://encyclopedia.pub/register :
Milijana Živković
,
Maša Žujović
,
Jelena Milošević
View Times: 224
Revisions: 2 times (View History)
Update Date: 12 Oct 2023
Table of Contents
    1000/1000
    Hot Most Recent
    Feedback