Content uploaded by Philip Obst
Author content
All content in this area was uploaded by Philip Obst on Jun 24, 2021
Content may be subject to copyright.
Rapid.Tech 3D Conference 2021
Scientific Article with Double-Blind Review
Scientific Article with Double-Blind Review
Komplexität und wirtschaftlicher Nutzen künstlicher Intelligenz zur
automatisierten und industrialisierten Erkennung additiv gefertigter
Bauteile
Complexity and economical value of Artificial Intelligence for
automated and industrialized recognition of additive manufactured
components
Philip Obst1, Walied Nasser1, Stefan Rink2, Gerhard Kleinpeter1, Blanka Szost1,
Dominik Rietzel1, Gerd Witt3
1BMW AG; 2AM-Flow; 3Universität Duisburg-Essen
Kurzfassung
Die Additive Fertigung (AM) befindet sich an einem Wendepunkt zur
Industrialisierung und Automatisierung. Aufgrund der immer kürzer werdenden
Produktentstehungszeit in der Automobilindustrie steigt der Bedarf an einer
flexiblen Produktion und damit einhergehend auch der Bedarf an der Herstellung
größerer Stückzahlen von Prototypenbauteilen. Daher wächst das Bestreben, die
AM Technologien und deren Prozesskette, die den Weg zur Serie entscheidend
abkürzen, weiter zu optimieren und deren Effizienz zu steigern. Die derzeitige
Identifikation von AM-Bauteilen am Ende der Gesamtprozesskette stellt einen
nicht skalierbaren und kostenintensiven manuellen Vorgang dar. Die Vielfalt an
Geometrien im Prototypenbereich führt zu komplexen Herausforderungen, bei
denen bestehende Automatisierungslösungen nicht implementiert werden
können. Eine KI-basierte Bilderkennung kann hierbei eine Verbesserung der
Situation sein. Eine Analyse hinsichtlich Komplexität, Funktionsweise und
Einsatz soll Aufschluss über die Wirtschaftlichkeit der automatischen
Identifikation von Prototypenbauteilen geben.
Short Abstract
Additive manufacturing (AM) is at a turning point towards industrialization and
automation. Due to the ever shorter product development time in the automotive
industry, the need for flexible production methods is increasing, and with it the
need to manufacture larger quantities of prototype components. Therefore, there
is a growing effort to further optimize and increase the efficiency of AM
technologies and their process chain, which decisively shortens the way to series
production. The current identification of AM components at the end of the overall
process chain represents a non-scalable and cost-intensive manual, labor
intensive process. The variety of geometries in prototyping leads to complex
challenges where existing automation solutions cannot be implemented. AI-
based image recognition can be an improvement in this regard. An analysis with
regard to complexity, functionality and deployment will provide information about
the economic efficiency of the automatic identification of prototype components.
141
!"#$%&'()*+,-+./01(2(0)(+3435
Scientic Article with Double-Blind Review
SECTION I – Scientic Articles | 627$)8(+9$7*+-/:;8(<=8$0%+!(>$(9
-?@=AB+
=ACD-+
!BECBF
E-Book mit Wasserzeichenschutz (2752688167f14c6ba79cc7430ba71a8e, BIC media 06/2021)
Rapid.Tech 3D Conference 2021
Scientific Article with Double-Blind Review
Scientific Article with Double-Blind Review
1 Introduction
The application range of Additive Manufacturing (AM) continues to grow.
Technological progress enables higher production speeds, an increased choice
of materials and adjustable robust mechanical properties that are close to
conventional products [1–4].This leads to an increase in its adoption and
application in various industries, such as the automotive industry. The possibility
of adding functions, such as mechanical moving parts or the integration of two
different material components, results in new use cases in the field of testing and
validating vehicle components. At the same time, applications in the end
consumer sector are growing [5–7]. Due to the technological unique advantages,
like geometric freedom, it is possible to produce new structures, shapes and
highly personalized and individualized components in series [8]. With time-to-
market in the automotive industry steadily decreasing, demand for prototyping
components is higher than before. Besides this, the trend even goes so far that
the vision of large-scale production, delivering just-in-time to assembly lines, is
emerging.
Figure 1: Separation of AM components within the BMW Group
However, in order to make larger AM volumes tangible, the process chains still
need to be optimized and further developed in many places. This is not only a
question of increasing output quantity and production speed, but also of economic
viability. The process chain of current available AM technologies still includes a
high amount of labor intensive work and process steps, which lead to a high
proportion of personnel costs and decreased product throughput. Also, these
operations lead to bottlenecks and downtimes in the overall process chain. For
this reason, a development towards automation and industrialization of the entire
AM industry can be observed today, which is shown by new solutions, applied
patents and announcements of collaborations and government-funded projects
[9]. The overall AM automation market is forecast to grow 23 % to a potential
revenue of $15 billion within this decade [10]. In order to effectively industrialize
an AM process chain, it is necessary to understand and analyze the overall
process chain of various technologies and to identify automation potential in form
of current production bottlenecks caused by manual operations.
142
!"#$%&'()*+,-+./01(2(0)(+3435
Scientic Article with Double-Blind Review
SECTION I – Scientic Articles | 627$)8(+9$7*+-/:;8(<=8$0%+!(>$(9
E-Book mit Wasserzeichenschutz (2752688167f14c6ba79cc7430ba71a8e, BIC media 06/2021)
Rapid.Tech 3D Conference 2021
Scientific Article with Double-Blind Review
Scientific Article with Double-Blind Review
2 State of the Art – Component identification in the AM
Production Workflow
Current available AM process chains reach a limit of productivity for large
production volumes [8]. AM processes are a complex sequence of individual
operation steps that mostly take place manually and physically. The production
limit is caused by machine capacity and runtime of the printing process itself,
which is connected to necessary manual operations such as machine cleaning,
preparation and unloading. In addition, process steps after printing include even
higher amounts of manual process steps which lead to bottlenecks, when scaling
up the production. One of these manual operations is the identification and
assignment to the acceptor or customer of the components by labeling the
components for further logistical transportation.
Figure 2 shows the overall process chain including the duration of component
identification in percentage. There are differences in the process phases
depending on the technology and used material. Here the HP Multi Jet Fusion
(MJF) technology with polymer powder as starting material is used as
representative.
Figure 2: Simplified overall AM process chain using HP MJF technology with percentage
duration of operations for manual component identification (Time measurement by AM-Flow)
Even though part identification is a small part in the overall process chain, it is
still a process step that does not scale and requires a high amount of manual
work in comparison to cooling for example, which does not require personnel
capacities. To achieve best economics (i.e. lowest cost per part), the components
are nested tightly with the support of specialized software. This leads to batches
with a large amount of different parts in one build job. However, this also means,
that the trackability of single components gets lost and the assignment to
customer orders needs to be done after the production. Usually it is a manual
143
!"#$%&'()*+,-+./01(2(0)(+3435
Scientic Article with Double-Blind Review
SECTION I – Scientic Articles | 627$)8(+9$7*+-/:;8(<=8$0%+!(>$(9
E-Book mit Wasserzeichenschutz (2752688167f14c6ba79cc7430ba71a8e, BIC media 06/2021)
Rapid.Tech 3D Conference 2021
Scientific Article with Double-Blind Review
Scientific Article with Double-Blind Review
process step that takes time and space, since the operator has to identify and
compare every single physical part with a 2D images from a list. Increasing
production throughput at this level increases the cost of additional employees,
slows down component delivery and requires more space. Therefore, an
automation of these manual operation steps should be implemented.
Figure 3: Manual component identification is the current state of the art at many companies
A distinction must be made between prototyping components and identical series
components. Series components are typically manufactured as identical parts,
so that only one geometry is produced per batch. Normally the typical operations
for final series components are quality control, determination of quantity and
collective bagging. Manual component identification consumes a low amount of
time when a batch only contains a little number of components like large objects.
However, as the number of components per batch grows, identification becomes
more time-consuming. The increasing build volumes and speed of new AM
machines reinforce this effect. The economy of scale allows a reduction of costs
per part due to minimized machine rates due to better utilization of equipment
and larger purchase quantities of material. Manual identification and sorting
however, results in higher costs per part for increased volumes. A recognition
from a larger total quantity leads to higher throughput times and growing error
margins. Therefore, there are strong indications that an automated
implementation of AM component identification has a high economical potential.
Figure 4: Shematic forecast of component cost shares and trends with larger quantities
144
!"#$%&'()*+,-+./01(2(0)(+3435
Scientic Article with Double-Blind Review
SECTION I – Scientic Articles | 627$)8(+9$7*+-/:;8(<=8$0%+!(>$(9
E-Book mit Wasserzeichenschutz (2752688167f14c6ba79cc7430ba71a8e, BIC media 06/2021)
Rapid.Tech 3D Conference 2021
Scientific Article with Double-Blind Review
Scientific Article with Double-Blind Review
However, in contrast to conventional production processes, automation of AM
component identification is associated with additional challenges. Other than the
basic idea of automation as replacement for recurring process sequences in order
to increase productivity and to reduce labor costs, a solution for this task needs
to be highly flexible. Flexibility in AM enables the economic production of different
batch sizes, different workpieces in any sequence and is a transition to the basic
idea of Industry 4.0 [11].
Today, in component recognition different methods are used to detect if a
component does not fit to a batch or to check simple quality features.
Differentiation of products is only done with unique recognition features, such as
bar codes. Unfortunately, although the technology allows it, it is not possible to
change functional prototyping designs by adding onset lettering or attached
identification labels. An application of such features affects component surfaces
and features. To automatically recognize components based on geometry,
following methods are available:
Table 1: Comparative expert evaluation of methods for component detection and
differentiation
Method Differentiation
ability
Processing
speed
Reduced costs
per part
Weighing 0 ++ ++
Point Cloud Scanning ++ 0 -
Image Recognition + + 0
Computer Tomography +++ - --
Most of the shown methods to classify and sort products are typically used for the
detection of series components. AM components in the automotive prototyping
sector are uniquely designed geometries, which are developed by different
engineers and departments. Weighing as a method is no accurate solution, since
weights of components are often close together, material inclusions may occur
and varying material conditions can result in differences in density. The use of
computer tomography or point clouds lead to high durations and is associated
with additional costs. Innovation in the field of Artificial Intelligence (AI) has made
the here necessary combination of flexibility and automation feasible. The
enormous development of especially deep learning algorithms within the last
decade enable systems to mimic human cognitive abilities that require strategic
thinking [12]. It enables machines to take over flexible tasks depending on data
input that previously could only be performed by humans. The input data can be
for example parameters, sensor measurement data or image material [13]. In
addition, AI has the ability to constantly improve and differentiate from image to
image. Therefore, camera based imaging recognition with image recognition in
combination with AI provides up to date the most promising approach.
145
!"#$%&'()*+,-+./01(2(0)(+3435
Scientic Article with Double-Blind Review
SECTION I – Scientic Articles | 627$)8(+9$7*+-/:;8(<=8$0%+!(>$(9
E-Book mit Wasserzeichenschutz (2752688167f14c6ba79cc7430ba71a8e, BIC media 06/2021)
Rapid.Tech 3D Conference 2021
Scientific Article with Double-Blind Review
Scientific Article with Double-Blind Review
3 Complexity of using Artificial Intelligence for Part Recognition
AI has already arrived in the AM sector. It is used for screening suitable
components, generating complex designs and for monitoring quality control
[14,15]. Deep learning and machine learning as application of AI helps to analyze
and evaluate the printing quality in-line and to implement optimization. AI helps
to manage large data volumes, where relevant information is not recognizable to
humans. This includes, for example, the detection of defects and pores [16].
However, an automated component recognition of different geometries is
associated with new technical challenging complexity factors, which are beyond
data analysis, shown in Figure 4.
Figure 5: Complexity factors leading to a challenge for AI based AM component
identification in the automotive industry
Until recently, there was no automated solution on the market capable of solving
these complexity factors. Internal market analyses have shown that AM-Flow's
AM-VISION is designed for the AM market and does not require high training
efforts for objects that might only appear once. The system uses an AI-based
algorithm that enables a recognition of AM components based on their unique
geometric fingerprint at a high processing speed. Rendered stl-files and scans of
different component batches are used for a cyclewise retraining for continuous
improvement.
4 AM-VISION – An Automated machine learning part recognition
system
The AM-VISION is an industrial machine to identify AM components based on
their unique geometry, manufactured by AM-Flow. It uses cameras and machine
learning algorithms to recognize part geometries. Component detection itself
146
!"#$%&'()*+,-+./01(2(0)(+3435
Scientic Article with Double-Blind Review
SECTION I – Scientic Articles | 627$)8(+9$7*+-/:;8(<=8$0%+!(>$(9
E-Book mit Wasserzeichenschutz (2752688167f14c6ba79cc7430ba71a8e, BIC media 06/2021)
Rapid.Tech 3D Conference 2021
Scientific Article with Double-Blind Review
Scientific Article with Double-Blind Review
takes 0.2 seconds. The algorithm processed in two computers uses pattern
recognition in order to match an optically scanned part with the digital CAD model.
The CAD model needs to be uploaded and analyzed in advance by the system,
which can be done automatically via a manufacturing execution system (MES).
The processing of the CAD files includes rendering from different angles and a
gravity analysis to determine how the object is likely to lie on the conveyor belt.
Therefore, physical recognition does not need a full 360° view of the AM
component. The system can be calibrated to specific materials and different
colors in order to increase recognition rates. Since the recognition takes place
digitally, next machining step can be determined by communicating with a MES
via an Application Programming Interface (API).
Figure 6: left: AM-VISION pilot setup (two-way conveyor belt) for pre-test study
right: Operator interface “AM-LOGIC” with touch-screen and 3D rotating objects
Figure 7: Functional principle and procedure of AM-VISION
Depending on the degree of automation of the production, components can be
placed on a one-piece-flow conveyor belt or manually into the system. After
recognition and internal comparison, the appropriate label for the following logistic
process can also be printed out or attached directly.
The development of AI solutions is dependent on availability of annotated
examples to learn from. Deep learning models obtain their accuracy by learning
147
!"#$%&'()*+,-+./01(2(0)(+3435
Scientic Article with Double-Blind Review
SECTION I – Scientic Articles | 627$)8(+9$7*+-/:;8(<=8$0%+!(>$(9
E-Book mit Wasserzeichenschutz (2752688167f14c6ba79cc7430ba71a8e, BIC media 06/2021)
Rapid.Tech 3D Conference 2021
Scientific Article with Double-Blind Review
Scientific Article with Double-Blind Review
from a high number of annotated data samples. AI image classification model of
the AM-ViISION system was trained with a large number of digital models. The
models were expanded with simulated variations generated with a CAD engine.
Therefore, it is not necessary to train the system with high quantity of physical
parts beforehand. If the recognition score (0-100%) is below a certain score, the
top three matches detected by the algorithm are shown and an operator has to
select and approve the correct one. This guided learning is also data to train the
system. This validation is a synthetic training and ensures further accuracy
improvements. The machine learning model becomes more accurate the larger
the set of objects processed with AM-VISION.
5 Economic viability of the AM-VISION
Test studies of build jobs containing a high geometry mix already proof that the
pilot setup saves time for identification and labelling. Components can be
processed up to 50% faster compared to the manual operation. Final machine
setups can be integrated in the production line with an automated one-way
conveyor belt. AM-Flow estimates that this will result in 6 to 10 times faster
processing time per part. The recognition rate is between 80 and 95 % if the build
job contains a high diversity of geometries. The more similar the geometries are,
the more the recognition rate can decrease. During testing, a calibration to the
gray HP MJF component appearance leads to an increase in the detection rate.
In the ideal case, the technology used and therefore the appearance color of the
physical component is also included in the name or code of the CAD file visible
for the system. This also makes it possible to mix technology batches in a single
pass. The functionality however, strongly depends on the component batch and
its included component variants.
It is still crucial to analyze from which output quantity and for which components
the application is functional and economical. Studies of panels, customized for
customers that differ only by finely embossed and engraved pattern or letters on
one surface side, show that AI is not yet sufficient for differentiation. Although the
geometry differs slightly, this case must be classified as still unsuitable due to the
series property. Figure 8 shows the percentage distribution of AM component
variants within the BMW AM production in a tested time frame. In addition, it must
be considered that an increased number of component geometries leads to an
exponential increase in maximum manual comparisons. Thus, the qualitative
effort due to the growth of the production volume can be visualized.
148
!"#$%&'()*+,-+./01(2(0)(+3435
Scientic Article with Double-Blind Review
SECTION I – Scientic Articles | 627$)8(+9$7*+-/:;8(<=8$0%+!(>$(9
E-Book mit Wasserzeichenschutz (2752688167f14c6ba79cc7430ba71a8e, BIC media 06/2021)
Rapid.Tech 3D Conference 2021
Scientific Article with Double-Blind Review
Scientific Article with Double-Blind Review
Figure 8: left: Percentage of polymer powder-based AM components of a short period in
2020; right: Increasing number of maximum component geometries to be compared in one
job
Series components that have no variance or large prototyping components, such
as a front grille for functional validation are displayed as 1 component variant per
job in this figure. This means that during the tested time frame, the sum of all jobs
containing more than 10 different geometries have a total proportion of around
25 %. However, a completely industrialized process flow also enables efficient
automated identification and labeling of individual prototyping parts in batches.
As it is a momentary observation, this distribution is not a guideline, as the
demand is constantly changing. The question therefore arises as the number of
high mixed parts to create a profitable business case in comparison to manual
operations.
The determination of a business case depends on several factors and framework
conditions, in order to calculate a cost comparison, like parts per day, failure
costs, full time equivalent and labor costs. Following automotive model input
values from AM-Flow determine the business model:
Table 2: Input values for comparing manual and automated component identification
Parts per day 130 (Manufacturer information)
Yearly growth of throughput 5%
Batch process steps 1,2 (polishing, tumbing or coloring)
Revenue per part 35 €/part (Manufacturer information)
Error rate 0.1% (Manufacturer information)
Shift times 7 h/day
Factory operational 5 days/week
Manual sorting (Sm) 2,9 hours/day
Automated sorting (Sa) 0.4 hours/day
Operation duration 60 sec/identification (model value)
Operator costs (including overhead)
120.000 € (approximate industry value)
Yearly machine costs 50.000 € (approximate industry value)
Yearly salary increase 3% (Manufacturer information)
Proportion Manual Mistakes 0.1%
Proportion Automated Mistakes 0.05%
0%
2%
4%
6%
8%
10%
12%
14%
16%
0102030
Percentage of AM jobs
Different geometries per job
0
100
200
1471013161922
Component
comparisons
Component variants
Maximum comparisons per job
149
!"#$%&'()*+,-+./01(2(0)(+3435
Scientic Article with Double-Blind Review
SECTION I – Scientic Articles | 627$)8(+9$7*+-/:;8(<=8$0%+!(>$(9
E-Book mit Wasserzeichenschutz (2752688167f14c6ba79cc7430ba71a8e, BIC media 06/2021)
Rapid.Tech 3D Conference 2021
Scientific Article with Double-Blind Review
Scientific Article with Double-Blind Review
Based on the parameters, a business case with a produced number of 130 high
mixed components can be calculated:
Yearly manual costs = FTEm * Operator salary + Failure costs
Yearly auto. costs = FTEa * Operator salary + Machine costs + Failure costs
With:
FTEm = Sm/8 * Factory operational/5 * 8/Direct productive hours
FTEa = Sa/8 * Factory operational/5 * 8/Direct productive hours
Failure costs = (Parts per day * Revenue per part * Mistakes * Shift times)/7 * 365
Figure 9: Overview of cost increase-decrease comparison for manual and automated
identification
6 Results and Outlook
The AM market is working on the emerging challenges for further industrializing
the overall process chain. While the process up to the production process itself
is digital and automated, a high proportion of manual work is required in the post-
processing. New developments and improvements in increasing the output
quantity lead to higher efforts in component identification. The deployment of AI
is effective here, even if the algorithms face challenges that are solved gradually.
The investigated AM-VISION System from AM-Flow is able to perform reliable
object detection of high mixed AM components based on a partial representation
of the geometry. The implementation results in a reduction in throughput time that
leads to cost savings due to reduced identification time and reduced failure costs,
even an operator is still required in the shown setup.
It is crucial to choose the right field of application for the automated identification.
Identical parts, parts that only differ on specific component areas or too little
volume of manufactured components do not lead to an economic benefit today.
It is this particular field that should be optimized in the future, since similar
components are also difficult to distinguish for humans and lead to bottlenecks.
108% 117% 127% 137% 148% 161% 174% 188%
-2% -9% -15% -21%
-47% -51% -54% -58%
-60%
-10%
40%
90%
140%
190%
2021 2022 2023 2024 2025 2026 2027 2028
Cost ratio compared to 2020
Cost increase of manual component identification
Cost reduction with automatic identification
150
!"#$%&'()*+,-+./01(2(0)(+3435
Scientic Article with Double-Blind Review
SECTION I – Scientic Articles | 627$)8(+9$7*+-/:;8(<=8$0%+!(>$(9
E-Book mit Wasserzeichenschutz (2752688167f14c6ba79cc7430ba71a8e, BIC media 06/2021)
Rapid.Tech 3D Conference 2021
Scientific Article with Double-Blind Review
Scientific Article with Double-Blind Review
Current research efforts are focused on modular and flexible process steps. This
facilitates flexibility in terms of the process steps used and enables loose linking
of the manufacturing steps. Thus, not every component goes through all process
steps, such as coloring and finishing [11,17]. In this case, it will be necessary to
identify components frequently in the overall chain.
Occurring fluctuations of the recognition rate are solved through the continuous
improvement of the deep learning algorithm. Updates during testing have already
helped to distinguish mirror-inverted automotive components. In addition to the
AI image classification model, it also has an AI decision model, which can handle
multiple simultaneous inputs. Further input could help enable automated
comprehensive quality control by using point clouds for example to measure
dimensional accuracy. Further improvements like laser triangulation could also
solve the restriction on the prototyping sector. With such an improvement, AM
series components could be identified, by fine patterns, serial numbers or small
Data Matrix Codes code integrated to the geometry.
The algorithms could additionally be applied in grippers for component handling
which is important in order to fill remaining bottlenecks for the fully automated
process chain. Nevertheless, the current available automated identification of a
high mix at high volume with the AM-VISION is already another step towards a
large scaled AM production.
7 References
[1] Wohlers T, Campbell RI, Diegel O, Huff R, Kowen J. Wohlers report 2020:
3D printing and additive manufacturing state of the industry. Fort Collins,
Colo.: Wohlers Associates; 2020.
[2] Osswald PV, Obst P, Mazzei Capote GA, Friedrich M, Rietzel D, Witt G.
Failure criterion for PA 12 multi-jet fusion additive manufactured parts.
Additive Manufacturing 2020:101668.
[3] Obst P, Riedelbauch J, Oehlmann P, Rietzel D, Launhardt M, Schmölzer S
et al. Investigation of the influence of exposure time on the dual-curing
reaction of RPU 70 during the DLS process and the resulting mechanical
part properties. Additive Manufacturing 2020;32:101002.
[4] Osswald TA. Understanding polymer processing: Processes and
governing equations. 2nd ed. Cincinnati, Munich: Hanser Publications;
Hanser Publishers; 2017.
[5] Boissonneault T. AM Focus Automotive: An exclusive eBook addressing
the intersection of additive manufacturing and automotive. 3dpbm -
Insights 2020.
[6] BMW Group. MINI Yours Customised: Vom Original zum persönlich
gestalteten Unikat; 2017.
[7] Lachmayer R, Lippert RB, Kaierle S (eds.). Additive Serienfertigung.
Berlin, Heidelberg: Springer Berlin Heidelberg; 2018.
[8] Obst P, Osswald P, Friedrich M, Rietzel D. Lean Additive Manufacturing:
Chancen für die wirtschaftliche variantenflexible Produktion. In: Kompass-
I: Gesellschaft für Ressourceneffizienz und Additive Technologien, 28–31,
2020.
151
!"#$%&'()*+,-+./01(2(0)(+3435
Scientic Article with Double-Blind Review
SECTION I – Scientic Articles | 627$)8(+9$7*+-/:;8(<=8$0%+!(>$(9
E-Book mit Wasserzeichenschutz (2752688167f14c6ba79cc7430ba71a8e, BIC media 06/2021)
Rapid.Tech 3D Conference 2021
Scientific Article with Double-Blind Review
Scientific Article with Double-Blind Review
[9] Photonik Forschung Deutschland. Integrierte Linienanwendung von
polymerbasierten AMTechnologien (POLYLINE); Available from:
https://www.photonikforschung.de/projekte/photonische-
prozessketten/projekt/polyline.html.
[10] Boissonneault T. AM Automation: The road to a fully automated and
digitalized additive manufacturing factory of the future 2020.
[11] Bauernhansl T, Hompel M ten, Vogel-Heuser B. Industrie 4.0 in
Produktion, Automatisierung und Logistik. Wiesbaden: Springer
Fachmedien Wiesbaden; 2014.
[12] VDMA Bayern. Leitfaden Künstliche Intelligenz: Potenziale und
Umsetzungen im Mittelstand; 2020.
[13] Murphy KP. Machine learning: A probabilistic perspective. Cambridge,
Mass.: MIT Press; 2012.
[14] Carolin Seidel. Industrialisierung von 3D-Druck schreitet bei der BMW
Group voran; Available from:
https://www.press.bmwgroup.com/deutschland/article/detail/T0322259DE/i
ndustrialisierung-von-3d-druck-schreitet-bei-der-bmw-group-voran.
[15] Wang C, Tan XP, Tor SB, Lim CS. Machine learning in additive
manufacturing: State-of-the-art and perspectives. Additive Manufacturing
2020;36:101538.
[16] Krabusch J, Meixlsperger M, Burkert T, Schleifenbaum JH. Prediction of
the Quality of L-PBF Parts Using Process Monitoring Image Data and
Deep Learning Models 2020.
[17] Photonik Forschung Deutschland. Industrialisierung und Digitalisierung
von Additive Manufacturing (AM) für automobile Serienprozesse (IDAM);
Available from: https://www.photonikforschung.de/media/photonische-
prozessketten/pdf/IDAM-Additive-Fertigung-Projektsteckbrief-bf-C1.pdf.
152
!"#$%&'()*+,-+./01(2(0)(+3435
Scientic Article with Double-Blind Review
SECTION I – Scientic Articles | 627$)8(+9$7*+-/:;8(<=8$0%+!(>$(9
E-Book mit Wasserzeichenschutz (2752688167f14c6ba79cc7430ba71a8e, BIC media 06/2021)
