See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/320824036

Survey on artificial intelligence for additive manufacturing

Conference Paper · September 2017

DOI: 10.23919/IConAC.2017.8082053

CITATIONS READS

54 4,052

4 authors, including:

Jimeng Yang Leo Chen

University of Glasgow Newcastle University
3 PUBLICATIONS 72 CITATIONS 98 PUBLICATIONS 1,429 CITATIONS

SEE PROFILE SEE PROFILE

All content following this page was uploaded by Jimeng Yang on 09 January 2018.

The user has requested enhancement of the downloaded file.

Proceedings of the 23rd International Conference on
Automation & Computing, University of Huddersfield,
Huddersfield, UK, 7-8 September 2017

Survey on Artificial Intelligence for Additive

Manufacturing
Jimeng Yang1, Yi Chen2, Weidong Huang3, Yun Li1
1
School of Engineering, University of Glasgow, Glasgow G12 8LT, United Kingdom
2
School of Computer Science and Network Security, Dongguan University of Technology, Dongguan 523808, China
3
State Key Laboratory of Solidification Processing, Northwestern Polytechnical University, Xi’an 710072, China
[email protected], [email protected], [email protected], [email protected]

Abstract—Additive manufacturing of three-dimensional construct each layer which is composed of massive

objects are now more and more realised through 3D segments along the predefined path. The process typically
printing, known as an evolutional paradigm in the uses the binder, laser or electron beam to solidify
manufacturing industry. Artificial intelligence is currently materials such as polymer, metal, and ceramic [5].
finding wide applications to 3D printing for an intelligent, Meanwhile, the extruder wheel can control the flow rate
efficient, high quality, mass customised and service-oriented of material [6, 7].
production process. This paper presents a comprehensive
survey of artificial intelligence in 3D printing. Before a At present, 3DP has been widely used in numerous
printing task begins, the printability of given 3D objects can domains including medical, architecture, food, mechanics,
be determined through a printability checker using machine aeronautics, chemical industry, education and social
learning. The prefabrication of slicing is accelerated culture, etc. Nevertheless, the limitations hindering the
through parallel slicing algorithms and the path planning is popularisation of 3DP still exist in the nowadays
optimised intelligently. In the aspect of service and security, manufacturing industry. The excessive time consumption,
intelligent demand matching and resource allocation shortage of real-time control, potential security hazard and
algorithms enable a Cloud service platform and evaluation transform from mass production to mass customisation are
model to provide clients with an on-demand service and principal problems that require being solved by
access to a collection of shared resources. We also present
computational intelligence, so-called artificial intelligence
three machine learning algorithms to detect product defects
(AI).
in the presence of cyber-attacks. Based on the reviews on
various applications, printability with multi-indicators, To date, AI has been developed in various forms and
reduction of complexity threshold, acceleration of applied successfully in a wide range of fields, including
prefabrication, real-time control, enhancement of security aviation, computer science, finance, education, healthcare,
and defect detection for customised designs are seen of good medicine, transportation, industrial manufacturing and so
opportunities for further research, especially in the era of on. It provides various algorithm, theories, methods and
Industry 4.0. offers great potential to transform the current
manufacturing technique under the situation of ever-
Keywords-3D Printing; additive manufacturing; artificial
intelligence; machine learning; algorithm; printability increased data repository. For example, machine learning
checking; prefabrication; path planning; cloud service (ML) is one representative of AI, which enables machines
platform; attack detection; security; real-time control. to learn and improve autonomously. Applying ML in
manufacturing can derive the useful information out of
I. INTRODUCTION existing data sets, so that provides a basis for
Today 3D printing (3DP) is known as one of the advanced approximations or predictions to operate machines with
manufacturing technologies and also a revolutionary future behaviours such as decision-making and automatic
progress for the next-generation manufacturing industry. It system improvement. It is also beneficial to detect certain
refers to the layer-basis additive manufacturing of three patterns or explore regularities in a dynamic
dimensional object using digital data [1, 2]. The manufacturing environment [8].
prefabrication stage aims to enhance the reliability and In this paper, a comprehensive survey of various AI
feasibility of 3DP. A 3D printable model designed through methods applied throughout 3DP are presented with
Computer-Automated Design (CAD) or 3D scanning is analyses in Section 2. It includes AI in printability
presented as triangle meshes in Stereolithography (STL) checking, slicing acceleration, nozzle path planning, cloud
format and then converted into sliced layers with G-code service platform, service evaluation and security of attack
to instruct a 3D printer [3]. The prefabrication of image- detection. Moreover, future directions are discussed in
based slicing, path planning, support generation, Section 4 based on the reviews of above methods and
orientation, repairing and packaging can significantly applications. Finally, conclusions are drawn in Section 5.
speed up the construction time and reduce the cost and
material waste [1]. II. ARTIFICIAL INTELLIGENCE APPLICATIONS
IN 3D PRINTING
Throughout the printing process, a typical 3D printer
implements fabrication of a physical object according to AI is described as indicated intelligence performed by
G-codes [4]. The “ink-jet” printing nozzle and platform machines, typically aims to enable computers to construct
move respectively in horizontal and vertical directions to intelligent systems for learning and solving problems like
human brain [9]. Same as most sciences, AI is divided reliability, especially when meeting plenty of various
into several subdisciplines, sharing an essential approach features or data.
to solving problem but applied in different areas, ranging
from Game Playing to Expert Systems, from Machine Hence Lu proposed to apply the ML method into
Learning to Neural Nets and Genetic Algorithms [10]. In automatic rule adjustment, especially the parameters used
this section, several kinds of AI methods and potential to verify the printability. Different from the above method,
directions for further researches are presented throughout the estimation model can be trained for printability
different stages of 3DP. prediction by using the Support Vector Machine (SVM)
instead of predefining any rules [12]. In this way, the
A. Printability Checker optimal decision function could be obtained for further
Printability is the capability to closely reproduce a 3D classifications by using the same parameters. This method
model via 3D printer [11]. Theoretically, the 3DP was also proved by experiments, which effectively
technique is expected to print any three-dimensional reduced the feature extraction time of 3D models without
object. However, compared with traditional manufacturing any negative impact on product precision.
methods, the promotion and employment of 3DP are still 3) Further Developments
limited due to the geometrical attribute, time consumption
and specific material requirement. In practical manufacturing industry, the printability or
complexity calculation is not only based on a single
1) Original Printability Checker indicator but an integration of multiple indicators such as
To reduce the complexity of product fabrication and time, cost, raw material, model size and geometry, etc.
ensure the 3D model can be made in an optimal way, Lu The problems about how to test and determine multi-
proposed the Printability Checker (PC) scheme to judge indicators and what effect proportion each indicator has
whether an object is suitable to be 3D printed or produced should be converted into an optimisation problem solved
through other ways [12]. by genetic algorithms (GA) and genetics-based ML
method to achieve the optimal effect proportion and the
3D GUI minimum complexity value. The reason is GA is designed
especially for large spaces or data that could be expressed
in binary string format [14]. Compared to other methods,
PC this probabilistic search method only requires few
assumptions to build objective functions [15].
Feature Extractor
（FE） Furthermore, the PC scheme is developed on the basis
Verifier Engine Rule
(VE) Editor of the current level of 3DP. To popularise 3DP, further
Printer Manager research should focus on the optimisation of printing
Printer (PM)
technique to lower the complexity threshold under multi-
indicator environment. By combining the improved
printing method with the assist of printability checker,
External more and more products are possible to be classified from
Solver unprintable to printable.

Figure 1. Architecture of Printability Checker [12]

B. Prefabrication
With the increasing design complexity, the optimisation of
In [12], a service-oriented printability problem computational prefabrication, also known as process
composed of feature extractor (FE), printer manager (PM) planning, has become a hot issue of 3DP [4]. At present,
and verifier engine (VE) is modelled. As shown in Figure lots of researchers have proposed their methods to
1, PC implements decision based on the findings of accelerate prefabrication. For instance, R. M. et al.
complexity values out of criteria. The calculation of proposed an asymptotical algorithm for adaptive slicing
complexity depends on the selection of different indicators problem [18]. Wang et al. presented a method to
such as tested runtime. With the assist of External Solver accelerate the slicing through parallel computing [13].
to decode the constraint rules coded in VE, three Vatani et al. optimised the slicing algorithms to reduce the
components can realise the information transfer and STL file sizes and computer memory by using nearest
cooperation each other. Specifically, FE is to extract the distance analysis [16]. Zhou et al. presented a hybrid
decidable features of a given 3D object. PM is to manage slicing process by integrating the laser-based vector
printers by utilising the relevant constraints and then send scanning and mask projection [17]. Fok et al. proposed a
the printer profiles to VE. Meanwhile, VE can match the path optimizer to search for the optimal printing trajectory
features and constraints from FE and PM so that tests the [6].
printability of the 3D object based on the final complexity
results [12]. 1) Slicing Acceleration
To convert the sliced 3D model into a set of slicing
2) Improvement by Machine Leaining planes in z-coordinates [4, 18], the layer information
Though the experiments prove the PC scheme has requires being extracted from triangular mesh by using
achieved the capability of checking printable model, the slicing algorithm.
practical implementation is still the lack of feasibility and
 In [13], Wang et al. presents a slicing algorithm used to find the shortest time of this visit. The transition
composed of three kernel modules, i.e. ray-triangle segment in intra-partition level is the connection of two
intersection (TRI), trunk sorting (TS) and layer extraction nearest end points located in two adjacent print areas [6].
(LE). Wherein TRI enables the slicing algorithm to
calculate the intersection points between vertical rays on
2D image pixel centres and triangle meshes in STL
format. A similar approach in [19, 20] utilises the plane-
triangle intersection to calculate the intersection point. TS
is to sort the intersection points in the order in the trunk.
According to the layer height and point position, LE is to
calculate the binary value of each pixel and then generate
the layer images for printing [13].
Though the slicing algorithm has established the Figure 2. Pixelwise Parallel Slicing [13]
foundation of subsequent layer-based addictive
manufacture, it has a few weaknesses in computational 3) Further Developments
complexity and difficulty of parallel implementation. In
Though the computational simulation has verified that the
today’s big data era, parallel computing has become a
scheme in [6] is capable of simplifying and accelerating
great potential for alleviating the computational demands
the printing process, there still exists a few defects which
of AI in the aspects of image processing, production
may affect the printing precision and optimal solution.
rules, mechanisation of logic, data filtering and data
mining, etc. [21]. Hence, Wang et al. proposed two Firstly, the Christofides algorithm utilising the
Graphic Processing Unit (GPU) schemes to accelerate distance as a criterion can only prove that the nozzle
this prefabrication process with the assist of pixel-wise traversing distance or time is optimal. But it fails to prove
parallel slicing (PPS) and fully parallel slicing (FPS) the total time consumption is the shortest under the
methods [13]. circumstance of neglecting the retraction time. In [6], it
points out that the retraction method has to be considered
The PPS aims to achieve the parallelism of slicing
due to the generation of excess filament leaking, which
algorithm. As shown in Figure 2, it enables GPU to
usually happens for a typical printer. It requires time to
support the threads and allocate them to each pixel ray,
retract the filament back when the nozzle traverses
which means all rays are operated by their specific
between segments. Thus, this retraction time should not
threads. Also, the threads can help to store the
be simply regarded as a constant due to its positive
intersection points and sorting results in a shared memory
correlation with the number of segments, which requires
in GPU, which significantly reduces the time
simulation and physical experiment to verify.
consumption. FPS further develops the parallelism of
RTI, TS and LE especially for solving the large size Secondly, to reduce the computing and printing time,
slicing problems. Compared to PPS, FPS enables the researchers proposed a simplified method to consolidate
multi-thread concurrency to operate three independent the small connected print segments into integrated
modules in a fully parallel method. [13]. segments based on the consolidation threshold value [6].
However, this approach is a lack of an optimal threshold
2) Path Optimisation
control. Because a large threshold is possible to effect the
A feasible printing trajectory can not only direct the boundary shape of the print area, causing wrong
nozzle to form the desired shape but also significantly transition segment between two adjacent inter-partitions
shorten the computational and printing time [1]. During and even the precision problem of the final 3D object.
the printing process, a printing nozzle basically spends Thus how and what degree the threshold control has that
time on traversing two types of segments, including print can achieve the optimal printing speed without negative
segment and transition segment [6]. Therefore it is impact on precision are key directions in future.
necessary to seek for an optimal path in the shortest
traversing time or distance. Finally, many researches have proved that parallel
slicing acceleration and path optimisation may offer an
Similar to Travelling Salesman Problem (TSP), Fok efficient performance. But there is few research on the
et al. proposed a relaxation scheme of 3DP path optimizer integration of printability checking, slicing and path
to compute the nozzle traversing time. The path planning. According to the layer-based printing process,
optimisation problem is formulated as TSP by comparing the parallel computing to both convert the 3D model into
each print segment to a city and finding the fastest or layer image and automatically form an optimal path is
shortest tour to visit the whole country. In each layer, reliable to further accelerate the prefabrication.
TSP implements optimisation at inter-partition level
(print area) and intra-partition level (blank area) based on C. Service Platform and Evaluation
the boundaries of these areas. Between each inter- Service-oriented architecture (SOA), known as a core part
partition, TSP can compute the visit order and start the of cloud manufacturing, refers to a computing paradigm
tour to visit each area. Then the Christofides algorithm is that provides enabling technologies and services to fulfil
the client requirements in an efficient and fast manner [22-
24]. A feasible SOA is able to intelligently realise the high In addition, in spite of service and manufacturing
flexibility, integration and customisation of 3DP [25]. intelligence, the cloud platform is still a lack of real-time
control during a printing process. Once a printing task
To date, several researchers have studied the problem begins, it will be difficult to interrupt or terminate the task
of correlation among virtual services under multiple already in progress, especially when connected with the
demands and constraints [26]. For example, Li et al. online Cloud platform. To reduce the massive time and
proposed the impact of service-oriented cloud material wastes, what intelligent method should be used is
manufacturing and its applications [27, 28]. Ren Lei et al. a principal objective to research.
presented the resource virtualisation and allocation for
cloud manufacturing [29]. Wu proposed the 3DP D. Security
technique in cloud-based design and manufacturing Security problem in manufacturing industry has gained
system [7]. Y. Wu et al. also developed a conceptual more attention in recent years. The cyber-physical attack,
scheme of 3DP service-oriented platform and an a new vulnerability to a cyber-manufacturing system
evaluation model on the basis of cloud service [30]. including 3DP, may cause several defects of products
1) Cloud Service Platform including change of design dimensions, void infill, nozzle
travel speed, heating temperature and so on [32].
The cloud platform is an on-demand computing model
composed of autonomous, hardware and software 1) Attack Detection
resources. As an example, in [30], it provides clients with To real-time detect the malicious attacks in 3DP, Wu et al.
a convenient on-demand access to a shared collection of proposed an ML method on physical data, covering k-
resources and then integrates the resources and Nearest Neighbours (kNN) algorithm, random forest
capabilities into a virtualised resource pool [22, 26, 30]. algorithm and anomaly detection algorithm. Specifically,
The assist of service evaluation and demand matching each layer-based image with defects is converted into a
algorithm enables the platform to intelligently make a greyscale plot. The features of grayscale mean, standard
comprehensive evaluation of terminal printers, providing derivation and number of pixels larger than the threshold
an optimal resource allocation based on printing precision, are all extracted according to the greyscale value
quality, cost and time [30]. Similarly, W. Wang et al. distribution.
proposed the resource allocation algorithms to build a
flexible and agile collaborative scheduling and planning of Based on extracted features, the ML algorithms will
resources [26]. real-time detect the outliers in defect areas and trigger
alerts to the administrator. Wherein kNN is to determine
2) Service Evaluation the classification of defect areas when the probability
Based on above content, the method to evaluate and select density of some parameters are unsure [32, 33]. The
the services of terminal printers was further developed by anomaly detection is to detect unusual outliers that do not
using ML method. In [30], Y. Wu et al. presented a conform the predefined or accepted behaviours, such as
service evaluation model using principal indexes increase of mean value, standard derivation and number of
including time, cost, quality, trust, ability and pixels larger than a threshold [32]. Similarly, in a cyber-
environment. With the help of fuzzy number different physical system, this unsupervised learning method is also
quantization based on Hamming Distance algorithm, the applied to detect anomalies with low false rate by using a
optimisation algorithm is able to quantify the service Recurrent Neural Network and Cumulative Sum method
quality and improve the accuracy of service selection [30]. [34]. The random forest can not only classify the defects
Similarly, Dong, Y. F et al. also presented a quality of by estimating the posterior distribution of each image
service acquisition method and a trust evaluation model layer but also build process-based patterns and use
for cloud manufacturing service using the genetic proximities to detect outliers according to the images.
algorithm [31]. Through simulation and experiment, the anomaly
detection algorithm was found to achieve the highest
3) Further Developments accurate detection [32].
A comprehensive cloud platform is not only a collection
of abundant resources. Most researches still focus on the 2) Further Developments
technical innovation, resulting in the lack of attention to The experiment has indicated the feasibility of ML in 3DP
system security and safety for future deployment and security. However, the research only works on the
adoption. Though the sharing platform may highly component surface with regular shapes or patterns.
improve the resource usage and provides the opportunity Because the construction of 3D model depends on the
to middle or small scale manufacturers to accomplish their requirements from clients in an actual printing process.
production. It also leads to the malicious attacks from For a complex construction, it is unable to guarantee the
side-channel, causing the risks of information steals or detection system will never identify a correct component
losses in the nowadays competitive environment [26]. For as a defect by mistake. Hence the current research on
enhancing users’ privacy, F. Tao et al. proposed the defect detection is only feasible for standardised
establishment of a private cloud platform, offering components in mass production, requiring further
benefits and services from public platform environment development to fit into mass customisation only with the
but self-managed [24]. Although it may lower the risk to capability to classify the defects accurately.
some extent, the specific criteria and standard to
implement still require further research and simulation.
 TABLE 1. SUMMARY OF AI METHODS FOR 3D PRINTING

3D Printing Process Status Application Method Problems &Future Work

Original printability 1. Multi-indicator test and
Printability Checking PC scheme (FE, PM, VE)
checker optimal effect proportion
(Design & Offline using GA.
Preparation) Automatic checking ML (SVM) 2. Lower complexity threshold.

Slicing Algorithm 1. Evaluation of time

(TRI, TS, LE) consumption closed to a
Slicing acceleration practical situation.
GPU (PPS and FPS
Prefabrication parallelism) 2. Consolidation of print
Offline segments and threshold
(Planning) control.
TSP based optimisation 3. Parallel computing of
Path Optimiser
Christofides algorithm printability checking, slicing
and path planning.

Cloud Service Demand matching algorithm

Service Platform & Platform Resource allocation algorithm
1. Security enhancement.
evaluation
Online ML 2. Real-time control.
(Design & Printing
Evaluation Model (Multi-criteria fuzzy decision 3. Design for printing.
&Service& Control)
based on Hamming Distance
algorithm; GA)
Security kNN, anomaly detection, Defect detection for mass
Online Attack Detection
(Control) random forest customisation.
correlation between time consumption and the number of
III. FUTURE DIRECTIONS print segments.
Based on the above survey, the developments and C. Mass Customisation in the Era of Industry 4.0
analyses of different methods for 3DP are simply outlined
in Table 1. The directions to further improve the Image recognition is an essential application of Neural
performance of 3DP are given as follows, Nets in deep ML of AI, inspired by human eyes of
neurology to implement recognitions from boundary to
A. Improvement of Prefabrication Acceleration angle, from complex construction to final object. By
For further shortening the prefabrication time, the applying ML in 3DP area, the recognition of defects or
threshold problem of print segment consolidation requires specific construction could be more accurate for
seeking for an optimal solution in case of excessive customisation production in future.
simplifications resulting in the deformed boundary and D. Online Real-time Control
wrong traversing path in each layer image. The relevant
factors effecting the control of threshold value should be To enhance the manoeuvrability, the Intelligent control
considered such as time, segment length, model size, method is a sound solution to automatic and real-time
geometry and even printer properties. Furthermore, an control the printing progress, especially when connected
integrated parallel computing of printability checking, with the online Cloud platform. It should be developed
slicing and path planning may accelerate computing time throughout whole 3D printing from the offline
as well. On the basis of FPS, this parallel method can be prefabrication to the accomplishment of physical printing.
developed to enable multiple types of threads to operate E. Design for Printing
different tasks in global memory.
Instead of spending time on checking printability, design
B. Closer to Practical Situation for printing should be considered as an efficient and
In path planning process, the evaluation of time convenient approach, which means all models will be
consumption should closely conform to a practical designed to match 3DP in particular. Thus clients will not
situation, which may provide the clients with high-quality consider the printability but only follow the 3DP criteria
services and optimal solutions for actual manufacturing. to design with the help of intelligent algorithms and CAD.
Due to the limitations of current printing technique, the Although it seems to be a long way to realise, especially
time spent on nozzle traversing, printing and filament under the comparable situation with traditional
retracting should all be considered. It also requires manufacturing industry, it requests both improvements of
simulations and physical experiments to determine the 3DP techniques and also the conversion to a digitised and
 personalised production mode which is integrated with [16] M. Vatani, A. Rahimi, and F. Brazandeh, “An enhanced slicing
products and services, so called Industry 4.0. algorithm using nearest distance analysis for layer
manufacturing,” Proceeding World Acad. Sci. Eng. Technol., vol.
3, no. 25, pp. 721–726, 2009.
IV. CONCLUSION
[17] C. Zhou, H. Ye, and F. Zhang. A Novel Low-Cost
A comprehensive survey of AI methods in 3DP has been Stereolithography Process Based on Vector Scanning and Mask
presented, ranging from various algorithms to the design Projection for High-Accuracy, High-Speed, High-Throughput, and
of the service-oriented platform. While current challenges Large-Area Fabrication. ASME. J. Comput. Inf. Sci. Eng.
have also been discussed, covering the real-time control, 2015;15(1):011003-011003-8.
parallel computing, attack detection for mass [18] C. Zhou, Y. Chen, Z. Yang, and B. Khoshnevis, “Digital material
customisation and conceptual scheme of design for fabrication using mask-image-projection-based
stereolithography,” Rapid Prototyp. J., vol. 19, no. 3, pp. 153–165,
printing. Although many researches and tests have proved 2013.
that 3DP is expected to become the mainstream mode of
[19] R. Minetto, N. Volpato, J. Stolfi, R. M. M. H. Gregori, and M. V.
production, further developments on AI in this domain are G. Silva, “An Optimal Algorithm for 3D Triangle Mesh Slicing
necessary to accelerate the realisation of this prospect in and Loop-Closure,” pp. 1–31.
the future Industry 4.0. [20] R. M. M. H. Gregori, N. Volpato, R. Minetto and M. V. G. D.
Silva, "Slicing Triangle Meshes: An Asymptotically Optimal
REFERENCES Algorithm," 2014 14th International Conference on Computational
[1] F. Yang, F. Lin, C. Song, C. Zhou, Z. Jin and W. Xu, "Pbench: a Science and Its Applications, Guimaraes, 2014, pp. 252-255.
benchmark suite for characterizing 3D printing prefabrication," [21] L.N. Kanal H. Kitano V. Kumar C.B. Suttner, “Parallel Processing
2016 IEEE International Symposium on Workload for Artificial Intelligence 1,” Volume 14 - 1st Edition. North
Characterization (IISWC), Providence, RI, 2016, pp. 1-10. Holland, p. 1-10, 1994.
[2] I. Gibson, D. W. Rosen, B. Stucker, et al. Additive manufacturing [22] Q. F. Hassan, “Demystifying Cloud Security,” CrossTalk, pp. 16–
technologies. Springer, 2010. 21, 2011.
[3] Jacobs, Paul Francis, “Rapid prototyping & manufacturing: [23] Q. F. Hassan, “Aspects of SOA: An Entry Point for Starters.,”
fundamentals of stereolithography,” Society of Manufacturing Ann. Comput. Sci. Ser., vol. 7, no. 2, p. 125, 2009.
Engineers,1992.
[24] F. Tao, L. Zhang, V. C. Venkatesh, Y. Luo, and Y. Cheng, “Cloud
[4] P. Kulkarni, A. Marsan, and D. Dutta. A review of process manufacturing: a computing and service-oriented manufacturing
planning tech- niques in layered manufacturing. Rapid Prototyping model,” Proc. Inst. Mech. Eng. Part B J. Eng. Manuf., vol. 225,
Journal, 6(1):18–35, 2000. no. April 2016, pp. 1969–1976, 2011.
[5] I. J. Petrick and T. W. Simpson, “Point of View: 3D Printing [25] G. Da Silveira, D. Borenstein, and H. S. Fogliatto, “Mass
Disrupts Manufacturing: How Economies of One Create New customization : Literature review and research directions,” Int. J.
Rules of Competition,” Res. Manag., vol. 56, no. 6, pp. 12–16, Prod. Econ., vol. 72, no. 49, pp. 1–13, 2001.
2013.
[26] W. Wang, Y. Wang, W. Williams, and A. Browne, “Secure Cloud
[6] K. Fok, C. Cheng, C. K. Tse, and N. Ganganath, “A Relaxation Manufacturing: Research Challenges and a Case Study,” pp. 1–12.
Scheme for TSP-based 3D Printing Path Optimizer,” 2016.
[27] Li B H, Zhang L, Wang S L, et al. Cloud manufacturing: a new
[7] D. Wu, J. L. Thames, D. W. Rosen, and D. Schaefer, “Enhancing service- oriented networked manufacturing model[J]. Computer
the Product Realization Process With Cloud-Based Design and integrated manufacturing systems, 2010, 16(1): pp. 1-7.
Manufacturing Systems,” J. Comput. Inf. Sci. Eng., vol. 13, no. 4,
p. 41004, 2013. [28] Li Bohu, Zhang Lin, Ren Lei, et al. Further discussion on cloud
manufacturing [J]. Computer integrated manufacturing systems,
[8] T. Wuest, D. Weimer, C. Irgens, and K.-D. Thoben, “Machine 2011, 17(3): pp. 449-457.
learning in manufacturing: advantages, challenges, and
applications,” Prod. Manuf. Res., vol. 4, no. 1, pp. 23–45, 2016. [29] Ren Lei, Zhang Lin, Zhang Yabin, et al. Resource virtualization in
cloud manufacturing [J]. Computer integrated manufacturing
[9] R. Brooks, “Intelligence without Reason,” Artif. Intell., vol. 47, systems, 2011, 17(3): pp. 511-518.
no. 1–3, pp. 139–159, 1991.
[30] Y. Wu, G. Peng, L. Chen, and H. Zhang, “Service Architecture
[10] G. F. Luger, Artificial Intelligence: Structures and Strategies for and evaluation model of distributed 3D printing based on cloud
Complex Problem Solving, vol. 5th, pp. 20-31, 2005. manufacturing,” pp. 2762–2767, 2016.
[11] A. Telea and A. Jalba, “Voxel-Based Assessment of Printability of [31] Dong, Y. F., and Gang Guo. "Evaluation and selection approach
3D Shapes,” in Mathematical Morphology and Its Applications to for cloud manufacturing service based on template and global trust
Image and Signal Processing: 10th International Symposium, degree." Computer Integrated Manufacturing Systems 20.1
ISMM 2011, Verbania-Intra, Italy, July 6-8, 2011. Proceedings, P. (2014): 207-214.
Soille, M. Pesaresi, and G. K. Ouzounis, Eds. Berlin, Heidelberg:
Springer Berlin Heidelberg, 2011, pp. 393–404. [32] M. Wu, Z. Song, and Y. B. Moon, “Detecting cyber-physical
attacks in CyberManufacturing systems with machine learning
[12] T. Lu, “Towards a fully automated 3D printability checker,” Proc. methods,” J. Intell. Manuf., pp. 1–13, 2017.
IEEE Int. Conf. Ind. Technol., vol. 2016–May, pp. 922–927, 2016.
[33] H. Xie, D. Liang, Z. Zhang, H. Jin, C. Lu and Y. Lin, "A Novel
[13] A. Wang, C. Zhou, Z. Jin, and W. Xu, “Towards Scalable and Pre-Classification Based kNN Algorithm," 2016 IEEE 16th
Efficient GPU-Enabled Slicing Acceleration in Continuous 3D International Conference on Data Mining Workshops (ICDMW),
Printing,” pp. 623–628. Barcelona, 2016, pp. 1269-1275.
[14] D. E. Goldberg and J. H. Holland, “Genetic Algorithms and [34] J. Goh, S. Adepu, M. Tan and Z. S. Lee, "Anomaly Detection in
Machine Learning,” Mach. Learn., vol. 3, no. 2, pp. 95–99, 1988. Cyber Physical Systems Using Recurrent Neural Networks," 2017
[15] T. Weise, “Global Optimization Algorithms–Theory and IEEE 18th International Symposium on High Assurance Systems
Application,” URL http//www. it-weise. de, Abrufdatum, vol. 1, p. Engineering (HASE), Singapore, 2017, pp. 140-145.
820, 2009.

View publication stats