International Journal of Production Research

ISSN: 0020-7543 (Print) 1366-588X (Online) Journal homepage: https://www.tandfonline.com/loi/tprs20

Cost analysis of alternative automated

technologies for composite parts production

B. A. R. Soares, E. Henriques, I. Ribeiro & M. Freitas

To cite this article: B. A. R. Soares, E. Henriques, I. Ribeiro & M. Freitas (2019) Cost analysis
of alternative automated technologies for composite parts production, International Journal of
Production Research, 57:6, 1797-1810, DOI: 10.1080/00207543.2018.1508903

To link to this article: https://doi.org/10.1080/00207543.2018.1508903

Published online: 18 Aug 2018.

Submit your article to this journal

Article views: 97

View Crossmark data

Full Terms & Conditions of access and use can be found at

https://www.tandfonline.com/action/journalInformation?journalCode=tprs20
International Journal of Production Research, 2019
Vol. 57, No. 6, 1797–1810, https://doi.org/10.1080/00207543.2018.1508903

Cost analysis of alternative automated technologies for composite parts production

a,b*
B. A. R. Soares , E. Henriquesc,d, I. Ribeiroc,d and M. Freitasc,d,e
a
Unidade de Investigação e Desenvolvimento em Engenharia Mecânica e Industrial (UNIDEMI), Caparica, Portugal; bFaculdade de
Ciências e Tecnologia Universidade Nova de Lisboa Campus da Caparica, Caparica, Portugal; cInstituto de Engenharia Mecânica
(IdMEC), Lisbon, Portugal; dInstituto Superior Técnico, Lisbon, Portugal; eATLÂNTICA – Escola Universitária de Ciências
Empresariais, Saúde, Tecnologias e Engenharia, Fábrica da Pólvora de Barcarena, Barcarena, Portugal
(Received 15 March 2018; accepted 25 July 2018)

Composite material usage in aircraft has been rising since the 1990s, with significant increases in manufacturing productivity
and repeatability due to automation in the production of aeronautic parts made of composite materials, becoming a strong
driver for widespread adoption of composites in this industry. Automated Tape Layup (ATL) and Automated Fibre
Placement (AFP) are two of the most important automated manufacturing technologies within aeronautics composites,
although their cost implications and economic comparison have not been widely studied. This paper presents an
economic evaluation of ATL and AFP technologies. Using process-based cost models, the manufacturing process of a
horizontal stabiliser is modelled, determining for each technology the associated consumption and use of resources and
their implications towards the part final cost. Results show that ATL is less expensive than AFP, due to lower material
costs, although with less material efficiency and slower cycle time.
Keywords: automation; operational research; process modelling; process-based cost modelling (PBCM); composites

Introduction
Composites have become a viable alternative to other materials in aeronautics since the 1990s (Marsh 1996). In some cases,
such as the Boeing 787 and the Airbus A350, they have become the most-used materials, with more than 50% by weight of
the aircraft. This has only become possible with automated processes, especially Automated Tape Layup (ATL) and Auto-
mated Fibre Placement (AFP). Although these technologies have been available industrially since the 1960s (Grant 2006),
the academic interest in them has been low until the 1990s. A bibliometric review (Lukaszewicz, Ward, and Potter 2012)
showed an increasing interest in these technologies: in 2010, publications in AFP were in the order of 150, and ATL
around 90 publications (compared to 50 and 30 in 1990). These numbers, however, are still below other composite pro-
duction processes such as filament winding, in which the same bibliometric review reported about 1200 publications in 2010.
Technological research in these production methods has increased, focused on various topics. Bader (2002) used material
selection to develop a model, estimating the cost and weight of producing a part using different fibres, resins, prepegs, and
manufacturing processes, and determined that, from a cost/performance metric, sometimes, more expensive materials are the
most effective due to weight reduction. Kaufmann, Zenkert, and Mattei (2008) studied part quality optimisation by creating
an optimisation framework that includes part quality as a variable in the manufacturing process to determine the impact on
production and maintenance costs. Apostolopoulos and Kassapoglou (2002) studied part size optimisation by determining an
optimal size considering learning curves and assembly processes that would minimise recurring costs, and Barbosa and Car-
valho (2013) created an analysis model to be used in the design stage of a part to foster the use of automation in composite
materials manufacturing, among others.
Cost modelling in these articles is somewhat generic, meaning that, in terms of scientific research, cost competitive-
ness analysis of technologies is still rare, despite its importance in the manufacturing world (Roy, Souchoroukov, and
Griggs 2008). In fact, cost modelling of composites-based manufacturing processes is still a small subset of research,
given the number of manufacturing technologies and the specificity of those processes. Despite the shortage of research,
authors such as Curran, Raghunathan, and Price (2004) provided a study on engineering cost modelling applied to aero-
space, stressing the genetic and causal nature of costs since they are inherited from and caused by specific design attri-
butes and production environment. The authors also conclude that costs can be scientifically modelled, integrated into

*Corresponding author. Email: ba.soares@fct.unl.pt

© 2018 Informa UK Limited, trading as Taylor & Francis Group

1798 B. A. Soares et al.
the engineering design process and becoming a key design variable. Also, Schubel (2012) compared various composites
processing techniques using Technical Cost Modelling (TCM), determining that automated processes can reduce man-
ufacturing costs by up to 8% in a 40 m wind turbine blade. Highlighting the influence of production volumes, Hagnell
and Åkermo (2015) developed a cost model with part complexity factors which compared Hand Layup, ATL with Hot
Drape Forming (ATL + HDF), and AFP, and concluded that, for their test case, Hand Layup was the most cost-effective
up to 150 parts with ATL + HDF taking over from there. However, these cost models are simplified and tend to use only
case-specific empirical relations that may, especially regarding material costs, fail to identify the range of components
where ATL and AFP are more cost-effective. The material cost is frequently the most important item in composite air-
craft components, and the amount of waste depends on the type of used technology and on the geometry of the part.
Furthermore, the application of the cost models in an industrial context involving real components is still less in
literature.
Given the growing trend of using composite materials in the aerospace sector and the prevalence of automated processes
in the production of composite parts, this paper presents the development and application of a process-based cost model
(PBCM) to analyse the cost-based competitiveness of alternative technologies by comparing production costs of a large air-
craft part. However, the model was developed to allow different parts with different areas, shapes, and thicknesses, as long as
both ATL and AFP are capable of producing them. This is shown by the process-based relations presented throughout the
paper. Although the whole production process is considered, this paper aims at supporting technology selection, an area quite
suited to cost modelling (Baumers et al. 2017), regarding two competitive processes for layup of composite fibres: the step
where most costs are defined. The cost model was developed with real industry data, based on an existing manufacturing
process, which reduces to minimum inferences and theoretical or lab-based data, in an area where even theoretical cost
models are rare. Furthermore, there is a contribution for calculating main cost driver of the process, the material cost,
relying on actual data instead of theoretical data, including losses due to quality, stemming from production issues, and tech-
nical losses, inherent to the process.
This will allow understanding the main cost drivers of each technology and its influence on final part cost. Additionally,
industrial contexts, such as machine usage, expected production volumes, and part dimensions are subjected to a sensibility
analysis to assess the best domains of application of each technology.

Research background
ATL has its beginning at the end of the 1960s (Grant 2006), when it was realised that unidirectional prepreg tape could be laid
up by automated methods in order to produce large composite structures with improvements in productivity and consistency,
in comparison to a manual layup. ATL is nowadays a highly productive process for prepreg layup, used preferentially to
produce large flat laminates with high areal weight. Its main drawback is its reduced capability to produce complex geometric
parts. Therefore, only simple parts, like tail planes, wing skins or centre wing boxes, are produced with this process (Lukas-
zewicz, Ward, and Potter 2012; Umer et al. 2016).
ATL is suitable to produce large flat parts and is usually capable of laying down 75, 150 or 300 mm wide unidirectional
prepreg tape. The width of the tape depends mainly on the curvature of the part, and to facilitate the laying process, tack of the
tape is strongly controlled. Lukaszewicz, Ward, and Potter (2012) state a maximum linear layup speed between 0.83 and
1 ms−1, with an acceleration of 0.5 ms−2. The compaction pressure depends on the width of the tape, generally around
0.1 MPa for thermosets.
Development of AFP started in the late 1970s/early 1980s, envisioned by Hercules Aerospace company (Campbell 2004;
Singh and Dwivedi 2009; Umer et al. 2016). By the end of the 1980s, AFP machines were commercially available and started
to establish themselves in the production of composites (Marsh 2011). They combined advantages from filament winding,
the roller design and deposition material guiding (Lukaszewicz, Ward, and Potter 2012), with the capabilities of ATL. Nowa-
days, AFP is a well-matured process, widely used in the production of composite structures, despite the initial investment,
mostly in machines and material (Mistry and Levy-Neto 1992; Croft et al. 2011; Debout, Chanal, and Duc 2011; Umer et al.
2016). Advantages include the possibility of reducing the number of parts, streamlining the assembly and maintenance
requirements, while improving the functional performance (Marsh 2011).
When compared with ATL, AFP machines layup prepeg material with smaller widths, from 3.2 to 12.7 mm. The head of
the AFP system can deliver 12, 24 or 32 tows side by side to form a band of material. However, each tow is individually
driven allowing it to be cut, placed, and restarted independently, which results in a more convenient process for producing
contoured shapes. Like ATL, the laminate orientation and material path are programmed to achieve the desired stiffness and
strength characteristics (Marsh 2011).
As mentioned in Campbell (2004) and in Hagnell and Åkermo (2015), the layup rate has a significant decrease in effi-
ciency as part size decreases, since the tape is rarely being laid at the optimum speed, with the machine having to decelerate,
 International Journal of Production Research 1799
cut the prepeg, and accelerate many times, making it unviable in terms of time and cost. The higher layup rate can only be
achieved, according to Hagnell and Åkermo (2015), for parts larger than 288 m2.
Automated prepeg deposition technologies also differ in the amount of the material technical scrap generated. Even if, as
pointed out by Grimshaw, Grant, and Luna Diaz (2001), the wasted prepeg is lower compared to a manual layup, the different
tape width in ATL and AFP introduces different scrap factors. Moreover, these scrap factors will be higher for small parts,
and smaller when scaling up parts until reaching a constant value. With the high cost of the raw material, this means that the
material technical scrap is a relevant factor when analysing the economic viability of the technologies.
Although ATL and AFP technologies have a similar nature, they have different approaches in a material layup. These
differences make the selection of the most appropriate one for the cost-effective production of a part a non-trivial decision.
Given these differences, this paper will focus on the manufacturing costs of an aircraft horizontal stabiliser, produced by
ATL and AFP using a carbon fibre/epoxy resin prepeg. The goal is to understand the technologies domains, determine the
main cost drivers, the cost variation, given differing production volumes and part dimensions, and to determine the effect of
machine uptime and technical scrap in the total cost of the stabiliser.

Materials and methods

Process-based cost model
Process-based cost modelling (PBCM), introduced by Field, Kirchain, and Roth (2007), is an analytical cost estimation
method, similar to activity-based costing (Rezaie, Ostadi, and Torabi 2008), which proposes that cost is a function of tech-
nical factors, such as materials consumed and cycle time, and includes process and operational inefficiencies that are deter-
minant in establishing the amount of resources required to achieve a production volume, in a time horizon, with a technology
(Kirchain 2001). By knowing the technical factors influencing the process, engineers can better understand how to minimise
production costs (Fuchs et al. 2008; Lin, Lee, and Bohez 2012). PBCM can also be used as a technological forecast tool for
new manufacturing technologies (Krachangphiphop, Wannasin, and Meemongkol 2017), since the model can be built based
on estimates, best guesses, and literature data (Nadeau et al. 2010). In its essence, PCBM defines the production process of a
product as a series of interdependent steps relying on a major technology. In its simple form, the output of the manufacturing
step i is the input of step i + 1.
The PBCM is decomposed into three interconnected models describing the Process, Operational and Financial aspects of
production. The Process model takes the product description, including geometry and material characteristics, and the
process technical parameters to determine the processing requirements, such as material quantities, cycle time, and equip-
ment requirements. The Process model determines what is technically required to produce the part, and its outputs are
then used in the Operations model to reflect the industrial environment, with specific operating conditions, in which the pro-
duction takes place. Its aim is to determine the resources use and consumption including operational inefficiencies. The infor-
mation gathered in the Operations model is then fed to the Financial model, which turns these requirements into costs to
determine production costs. By linking the sequence of technological steps in the manufacturing process and relying on
theoretical, practical, and empirical relationships among process parameters, a PBCM can be built to analyse production,
make predictions, change parameters in which uncertainty or future evolution can be expected and perform sensitivity analy-
sis (Storck 2010).

Test case
In this paper, the PBCM developed is applied to an aircraft part, currently produced by ATL and AFP. The test case is a wing
skin of a horizontal stabiliser, with an area of 8.25 m2, an expected annual production volume of 50 parts and a geometry
suitable to be produced by ATL and AFP. The ply stacking sequence is [0°/90°/+45°/−45°] balanced and symmetrical,
repeated 10 times for a total of 40 layers.
The data for the Operations and Financial models were obtained in the plant producing the part, where an annual pro-
duction volume of 50 parts is expected soon. The Process model also included information from literature (Campbell 2004;
Field, Kirchain, and Roth 2007).
Whether using ATL or AFP for the layup, the production of the horizontal stabiliser involves several process steps from
material reception to painting before going to the assembly plant. Figure 1 presents a sketch of this manufacturing flow.
It starts with the material reception, where the data for the composite prepeg are checked (date of manufacture and
shelf life) and placed in cold storage. When required, the material is taken out of cold storage to heat up to ambient temp-
erature (1). The part manufacturing begins with the cleaning and preparation of the mould, followed by the hand layup of a
copper mesh (to allow for skin electrical conductivity), and glass fibre layers as an interface material, since carbon fibre
1800 B. A. Soares et al.

Figure 1. Global ATL- and AFP-based process flow.

promotes galvanic corrosion in other materials. This initial deposition is then vacuum bagged to remove voids and con-
solidate the layers (2).
Next, it is moved to the fibre placement machine, where layers of composite prepeg are automatically deposited, with pre-
set ply orientations (3). Afterwards, the part goes into another hand layup of glass fibre, and vacuum sealed to prepare for the
autoclave process (4), in which it is subjected to heat and pressure cycles for up to 12 hours to cure the prepeg (5). The part is
then demoulded and trimmed to its net shape (6). The last process considered is non-destructive ultrasound testing, where the
part is analysed to detect possible non-conformities, such as voids, wrinkles, and debonding, among others (7).
The manufacturing process flow is the same either using ATL or AFP except for the Fibre Placement step (number 3 in
Figure 1). Based on the assumption that differences on using these technologies are confined to the prepeg layup and to the
raw material costs, which is in accordance with literature and industrial experience, this process step will be the focus of this
study.

Material requirements
After the process flow has been delineated, a mass balance is performed across every step, where the physical quantities of
materials and work-in-process parts entering and leaving each process step are determined considering inefficiencies (tech-
nical scrap) and quality issues (quality scrap). The objective is to determine how much material is required and how many
parts must be produced in each step to meet the expected production volume. It is constructed backwards, starting with the
required production volume as the output in the last process step to the material input in the first step, considering the material
losses in the process (Figure 2).
The number of parts required in process step i is given by

NPi+1
NPi = (1)
1 − (Scrpi + rwi × rscrpi )

Figure 2. Framework of non-quality losses in process step i.

 International Journal of Production Research 1801
where NPi is the number of parts in process step i, Scrp is the percentage of defective parts that are directly considered scrap,
rw is the percentage of defective parts that potentially can be recovered, and rscrp is the percentage of parts that after being
reworked do not pass inspection and are scrapped.
In other words, given a part shape to be produced, the tape layup will always generate a surface that is larger than the
part itself. This excess of material, called technical scrap, is different in ATL and AFP technologies. In fact, it depends
not only on part geometry and tape layup orientation but also on the tape width. Although the range of typical scrap rates
can be found in literature, (Lukaszewicz, Ward, and Potter 2012, Hagnell and Åkermo 2015) , the scrap rates for the
studied part can be determined, by superposing its geometry with continuous tows of appropriate width (300 mm for
ATL and 12.7 mm for AFP), in the four orientations used (0°,90°, and ±45°).
By knowing the number of parts required in the fibre placement step and the technical scrap generated per part, the
amount of prepeg material consumed for the production volume can be calculated. Since the prepeg is sold in rolls of
600 linear meters, the number of rolls required for the production volume is given by

surfa × n◦ l × NPi × ( 1 + TechScrap )

Nrr = (2)
tw × tl

where Nrr is the number of rolls required, surfa is the part surface area, n°l is the number of prepeg layers, tw and tl are the
tape width and length, respectively, and TechScrap is the process losses inherent to the manufacturing technology, percentage
wise according to


◦ totai
åni=1plySt −1
surfa
TechScrap = (3)
n◦ plySt

in which n°plySt is the number of ply orientations in the stacking sequence, and totai is the total area of material deposited
in each i ply orientation.
Note that in Equation (2) the scrap derived from quality issues is already considered in the NPi variable, as the number of
parts that are necessary to process in step i to obtain the required number of good parts.
Since incineration is the most probable disposal scenario of composite materials, scrap generated in the fibre placement
process represents a parcel in the manufacturing cost, due to its acquisition, and also its disposal. The mass of scrap generated
can be calculated by

Scrap = r × Surfa × n◦ l × NPi × ( TechScrap + scrp + rw × rscrp) (4)

in which ρ is the prepeg material density per unit of area.

Time-based requirements
Costs are highly influenced by the time to manufacture the part since during that time resources are allocated and consumed.
The manufacturing cycle time, the time necessary to produce a part, is the sum of the set-up time of the machine with the time
the machine takes to produce the part. In this study, the machine time for ATL and AFP is adapted from the layup rates pub-
lished by Lukaszewicz, Ward, and Potter (2012). These layup rates are affected mostly by the dimension of the part surface
since, for smaller part sizes, the equipment spends less time at maximum speed, significantly reducing the layup rate. The
layup rates were developed for a flat rectangular part, considering effects of part size, areal weight, and ply orientation, but
they do not distinguish between effective and wasted deposition.
The machine time per part, meaning the time required to layup all the prepeg layers, depends on the part surface area,
weight, layup rate, and the number of layers specified for the part. Finally, the part cycle time on the machine is the sum
of the set-up and the machine time. Once the cycle time is known the time required for the fibre placement step, Treqi, is
given by

Treqi = CTi × NPi (5)

where NPi is the number of parts that are required to be produced in step i for the intended production volume, consider-
ing non-quality part losses, and CTi is the part cycle time in this process step.
1802 B. A. Soares et al.
In an industrial environment, the hourly cost of a machine is significantly influenced by the machine uptime, understood
as the machine productive time in a specific time frame (generally a year). The machine uptime is a variable calculated by

Treqi
Uptimei = (6)
Alloci

where Alloci is the percentage of the machine uptime occupied by the production of the part in required quantities, Treqi is the
time required to produce these parts, and Uptimei is the uptime of the machine in process step i.
It was assumed that production of the required quantities of the part induces a certain occupation of the machines’ uptime.
This assumption reflects somehow the context of aeronautic parts manufacturing. If a change occurs in the required quantity
of the analysed part, this means that more aircrafts are being assembled and thus, the production volumes of the remaining
parts, produced in the same machines, are also expected to increase. Thus, the occupation induced by the part is expected to
be approximately constant and so any change in its production volume affects its machine cost item, as far as it impacts the
machine uptime.

Cost relations
Costs can be divided into variable costs, i.e. production volume dependent, and fixed costs, generally associated with the
process step and requiring an initial investment (Table 1).
Material costs are calculated based on their price per unit (weight, area or rolls) and on the required quantities determined
by the mass balance. The following equations are used:

Matcos t = C Kg × ra × surfa × n◦ l × NPi (7)

Matcos t = C sqm × surfa × n◦ l × NPi (8)

Matcos t = C roll × Nrr (9)

where Matcost is the total cost of the material for the production volume, Ckg is the material price per unit of mass, Csqm is
the material price per unit of area, Croll is the material price per roll, ρ is the material areal weight, surfa is the part surface
area, n°l is the number of layers, NPi is the number of parts produced in the process step i, and Nrr is the number of material
rolls required for the production.
Directly associated with the materials, the scrap costs reflecting the disposal expenses are calculated by the material
losses in the process multiplied by the disposal cost per unit.
Labour costs can be defined as

Labor= CTi × n◦ wi × dedi × wi × NPi (10)

where n°wi and dedi are the number of workers and the percentage of their dedication to the process step i, w€i is the average
wage per unit of time, CTi is the cycle time, and NPi is the number of parts produced in the step.
Energy cost is the energy directly consumed in part production. It depends on the machine time required, power, and
energy unit price.
Machine, tooling, and building costs are calculated based on an annual equivalent rent associated with the capital
investments. The annual rent, tooling or building infrastructures depend on the initial acquisition cost of the asset j (invest-
ment), Ij, useful life, nj, and opportunity cost annual rate. Assuming a null residual value at the end of useful life, it can be

Table 1. Variable and fixed costs in a given process step.

Variable costs Fixed costs
Material cost Main machine cost
Scrap cost Tooling cost
Labour cost Fixed overhead cost
Energy cost Building cost
 International Journal of Production Research 1803
calculated by

(1 + r)nj × r
EqCost ji = Ii (11)
(1 + r)nj − 1

To transform these equivalent annual costs into a cost associated with the part under analysis, the following equations were
used:

åj (EqCost ji + Mant ji ) × Alloci

Machine cos ti = (12)
NPi

åj EqCost ji
Tooling cos ti = (13)
NPi

åj EqCost ji × SfAreai × Alloci

Building cos ti = (14)
SfArea × NPi

where SfArea is the global shop-floor area, SfAreai is the area occupied by the infrastructure of the manufacturing process
i, and Mantji is the annual maintenance cost of the machines. One should note that since tooling is part-specific, its cost is
fully allocated to the part.
Finally, the overhead costs, indirect costs associated with support engineering labour, were determined as a percentage of
direct labour costs. The part total manufacturing cost can then be calculated as the sum of these cost items, either per part or
per global production volume.

Results and discussion

A full table with the inputs for the fibre placement step is shown in the Appendix.

Material requirements
The material requirements involve the determination of the number of parts that must be produced and the technical scrap in
each process step, taking into consideration the effect of non-quality along the production flow and the technologies
involved.
The material losses resulting from part non-conformity issues, gathered in the industrial field with industry experts, were
considered similar for ATL and AFP. Table 2 presents the non-quality part losses, together with the resulting number of parts
that must be produced to fulfil the intended production volume of 50 parts.
Based on the number of parts required along the process (55), the material requirements are calculated (Equation (2))
considering the effect of technical scrap, which is dependent on the technology since ATL uses one 300 mm width tape
and AFP uses multiple individual 12.7 mm width tows (Figure 3).
As shown in Table 3, there is a significant difference on the technical scrap generated when ATL is compared with AFP:
17.7% and 1.15% of the lost material, respectively, meaning that material requirements are lower in AFP than for ATL.

Table 2. Quality issues and number of parts required in each step for an annual production of 50 final parts.
Reworking Direct scraping Scraping after rework Passing parts Parts required
Reception, check and storage 0 0 0 100% 55a
Copper and glass fibre placement 2% 1% 0 97% 54.7
Fibre placement 2% 1% 1% 97% 54.2
Glass fibre placement 4% 1% 10% 95% 53.6
Autoclave 0 1% 0 99% 52.9
Demould., finish. and adjustment 0 3% 0 97% 52.3
Non-destructive testing 1% 1% 50% 98% 50.8
a
Rounded up.
1804 B. A. Soares et al.

Figure 3. Basic scheme for the different amount of technical scrap in ATL and AFP technologies.

Table 3. Material losses as technical scrap in the fibre placement step.

Material amount Difference
2
Part area (m ) 8.26
ATL AFP
Ply orientation Material at 0° (m2) 10.6 8.37 21.0%
Material at 90° (m2) 9.43 8.31 11.9%
Material at +45° (m2) 10.06 8.37 16.8%
Material at −45° (m2) 10.06 8.37 16.8%
Minimal cut length (mm)a 100
Total material in part (m2) 401.39 334.19 16.7%
Total material required (m2) 330.35 –
Technical scrap (m2) 71.04 3.84 94.6%
Technical scrap (%) 17.7% 1.15% –
a
Additional material to allow cutting the tape in each deposition

Time-based requirements
The cycle time for each process step was determined by collecting time data in situ, with the exception of the fibre placement
step, where fitting equations based on the (Lukaszewicz, Ward, and Potter 2012) data was used for ATL and AFP (with a part
set-up time of 2 h). As shown in Table 4, only the cycle time in the fibre placement is different, with the cycle time and,
consequently, the time required being significantly lower in AFP compared to ATL, because of its higher time-based
productivity.
Table 4 also presents the percentage of equipment uptime occupied by the annual production volume of the part. Since
the equipment involved in the different process steps is not dedicated only to parts produced by ATL and AFP technologies,
different percentages of uptime occupation were considered to reflect this.

Cost calculation
The costs for the fibre placement step were calculated according to Equations (7)–(14). The results (Table 5) show that the
main cost driver is material consumption, followed by machine costs. Material cost accounts for 73.7% and 77.5% of the part
total cost for ATL and AFP, respectively.

Table 4. Cycle time and time requirements for an annual production volume of 50 final parts.
Time required (h/ Uptime
Cycle time (h) year) occupation (%) Uptime (h/year)
ATL AFP ATL AFP ATL AFP ATL AFP
Reception, check and storage 4 218,8 25% 875
Copper and glass fibre placement 6 328 30% 1094
Fibre placement 5.58 4.42 302 239 55% 55% 550 435
Glass fibre placement 6 322 25% 1287
Autoclave 14 740 60% 1233
Demould., finish. and adjustment 5 262 45% 581
Non-destructive testing 6,8 345 40% 863
 International Journal of Production Research 1805
Table 5. Fibre placement costs per part for an annual production volume of 50 final parts.
ATL AFP
Variable costs Material cost (€) 10,419 73.7% 12,126 77.5%
Labour cost (€) 344 2.4% 306 2.0%
Energy cost (€) 242 1.7% 192 1.2%
Total variable cost (€) 11,005 77.9% 12,624 80.7%
Fixed costs Machine cost (€) 2069 14.6% 2069 13.2%
Tooling cost (€) 103 0.7% 103 0.7%
Overhead cost (€) 894 6.3% 796 5.1%
Building cost (€) 60 0.4% 60 0.4%
Total fixed cost (€) 3126 22.1% 3028 19.3%
Total production cost (€) 14,131 100.0% 15,652 100.0%

The total variable costs are lower in ATL than in AFP due to the material cost item. Although as shown previously the
amount of material loss in the layup process is significantly higher in ATL, which not only negatively affects the material
acquisition but also the disposable costs of the scrap material, the cost per unit of area of the material for ATL is substantially
lower than the cost per unit of area for AFP. According to the data collected, the cost of ATL material is 26% lower than for
AFP. Even if it does not balance the material cost difference, AFP has a slight advantage in labour and energy costs, due to its
lower cycle time.
Except for the overhead costs, which are a linear function of labour costs, the fixed costs are the same for both technol-
ogies. Since acquisition cost, useful life and opportunity cost rate are the same for both equipment, it might be expected that a
significant difference in cycle times would have induced a divergence in machine costs. However, it was assumed that in any
case, the part production occupies 55% of the equipment uptime, the lower required time of AFP results in a lower uptime
and so in a lower cost per occupation hour.
The total cost of the fibre placement per part in the test case is lower in ATL by 9.7% due to its lower material cost.
However, this difference gets diluted when the global process is considered, and the total production cost of the horizontal
stabiliser is determined. Following the same methodology, the whole process was modelled. The results show that globally
the final production cost of the ATL option is 5.1% lower than the AFP option (Tables 6 and 7). Material is the highest cost
item of the horizontal stabiliser, responsible for over 40% of the total costs, a result which is in line with Grant (2006). The
machine costs are similar to the material costs, with both representing more than 80% of the part cost.
Fibre Placement is the highest cost step for both technologies, being responsible for around 50% of the total cost, fol-
lowed by the Autoclave and the Demoulding Finishing and Adjustment (DFA) steps. However, this is because more than
90% of the raw material is consumed in the Fibre Placement step (10,419 € and 12,126 € for ATL and AFP material
cost, respectively). If the material cost was considered as a separate cost item, reflecting the principle that this material
cost is a part cost and not a technology cost, the Fibre Placement step would cost less than both the Autoclave and DFA steps.
The split of material and time-based costs allows a better perception of their importance in the Fibre Placement step,
focusing the attention on reducing the material losses (non-conformity and technical scrap account for 9% and 22% of
material costs for AFP and ATL cases, respectively) or on reducing the part cycle time to increase the cost-related
competitiveness.

Table 6. Total production costs items per part for an annual production volume of 50 final parts.
ATL AFP
Variable costs Material cost (€) 11,424 40.20% 13,133 43.90%
Labour costs (€) 1059 3.80% 1021 3.40%
Energy cost (€) 864 3.00% 813 2.70%
Total variable cost (€) 13,347 € 47% 14,967 50.00%
Fixed costs Machine cost (€) 11,892 41.90% 11,892 39.70%
Tooling cost (€) 103 0.40% 103 0.30%
Overhead cost (€) 2543 9.00% 2445 8.20%
Building cost (€) 521 1.80% 521 1.70%
Total fixed cost (€) 15,059 53.00% 14,961 50.00%
Total production cost 28,405 € 100% 29,928 € 100%
1806 B. A. Soares et al.
Table 7. Total production costs per part for an annual production volume of 50 final parts.
ATL AFP
Cost (€) % Cost (€) %
Reception, check and storage 1156 4.10% 1156 3.90%
Copper and glass fibre placement 1237 4.40% 1237 4.10%
Fibre placement 14,131 49.70% 15,652 52.30%
Glass fibre placement 584 2.10% 584 2.00%
Autoclave 5695 20% 5695 19.00%
Demoulding, finishing and adjustment 3961 13.90% 3961 13.20%
Non-destructive testing 1641 5.80% 1643 5.50%
Total 28,405 € 100.00% 29,928 € 100.00%

Of note is that although all the steps other than fibre placement are equal, there is a difference in cost of the Non-Destruc-
tive Testing, since, although the rework rate and material use were considered equal for both processes, the material cost in
AFP is higher than on ATL.

Sensitivity to production volume and machine uptime

The last section dealt with a cost analysis for a production volume of 50 horizontal stabilisers per year occupying 55% of the
uptime of the fibre placement equipment. With this baseline case established, several studies can be performed such as:
. how do the fibre placement-related costs vary with the production volume;
. when is capacity reached in the baseline case;
. what happens to cost per part if the machine is dedicated, and
. what is the minimum cost per part if the 55% uptime assumption is removed and do they vary in a different way
between ATL and AFP.
Tests were performed to determine the variation of costs with production volume, using the baseline case with 55%
uptime and two limit situations where the 55% uptime assumption is removed. The first assumes that the uptime of the
fibre placement equipment is equal to its available time, i.e. equal to the time in which the equipment is in its specified con-
ditions and can operate. The second assumes that the machine is dedicated to the production of the part, and so the equipment
uptime is equal to the time required to produce it. It should be noted that these two situations induce the low and high part
cost limits, in which all the intermediate situations are contained. The results are shown in Figure 4.
The results show the unit cost reduction with the horizontal stabiliser production volume. Note that at a certain point, the
available capacity is no longer enough to guarantee the required production level and new equipment must be acquired. In the
baseline case, additional equipment is needed when the production volume is 490 and 618 in ATL and AFP, respectively.
Although at these points the cost per part is affected (a jump in the cost per part followed by a similar evolution pattern)
and the effect occurs sooner for ATL, it is not sufficient to make a change in the most cost-effective technology.
The limit situations show that when the equipment is always working, a minimum cost per part is achieved, which is
almost insensitive to production volume. In opposition, when the equipment is dedicated, a high sensitivity exists, especially
for low volumes. For a production volume of 50 units, the cost per part can range between 13,690 € and 17,598 € in AFP and
from 12,180 € to 16,210 € in ATL, depending on the equipment uptime.

Figure 4. Limits of fibre placement cost per part as a function of production volume and equipment uptime.
 International Journal of Production Research 1807
When the production volume increases significantly the effect of the equipment becomes smaller: the cost curves for each
technology meet at the point where the fibre placement machines are producing the horizontal stabiliser using all their avail-
able time. This point is theoretically the most cost-efficient point (reflecting most the variable costs per part), with the equip-
ment at maximum capacity. Above it, it would be necessary to invest in additional machines.

Sensitivity to part dimensions

For a production volume of 50 units, under the assumption of 55% of occupation of equipment uptime, the material cost is
clearly the main cost driver, since the material costs include the material losses in the process, due to non-conformity issues
and due to material technical scrap. While dependent on fibre placement technologies, the material technical scrap is also
significantly influenced by the geometry and size of the part. The question on how much different part dimensions influence
the technical scrap and the fibre placement costs on both technologies deserves to be explored.
Keeping its geometry, the baseline part was scaled, by changing the length of the wing, with the equivalent change in
width and area, from 3 m (the baseline) to 1, 4, and 5 m. Table 8 shows the technical scrap for the horizontal stabiliser,
scaled to the different part areas.
While the technical scrap in AFP remains small even when the part area changes significantly, severe material losses
occur for a small horizontal stabiliser (0.92 m2 of area) made by ATL. In fact, the percentage of technical losses in both
cases is a negative rational power of the part area (Figure 5). However, the behaviour of the curves diverges significantly
as the area is reduced. Naturally, the coefficients and power exponents depend on the geometry of the horizontal stabiliser
together with the characteristics of the equipment, such as minimum cut length shown in Table 8. This significant increase of
the material losses in ATL when parts get smaller means that ATL can become costlier than AFP due to the technical scrap
produced offsetting the lower material-specific cost.
The fibre placement cost per part increases with the part area since a higher part area implies more material and more
cycle time with a direct impact on the variable costs. For the part and operative conditions considered, AFP becomes less
expensive than ATL for parts with a layer area smaller than 1.76 m2 due to the differences in technical scrap (Figure 6).

Table 8. Material losses as technical scrap in the fibre placement step as a function of part dimension.
Material required Material required Material required
2
Part area (m ) 0.92 14.68 22.94
ATL AFP ATL AFP ATL AFP
Ply orientation Material at 0° (m2) 1.68 0.95 17.85 14.82 27.01 23.11
Material at 90° (m2) 1.30 0.94 16.37 14.75 25.07 23.04
Material at +45° (m2) 1.48 0.96 17.16 14.84 25.93 23.13
Material at −45° (m2) 1.48 0.96 17.16 14.84 25.93 23.13
Minimal cut length (mm) 100
Total material in part (m2) 59.25 38.03 685.45 592.45 1039.43 924.15
Total material required (m2) 36.66 36.66 587.29 587.29 917.64 917.64
Technical scrap (m2) 22.59 1.37 98.16 5.16 121.79 6.51
Technical scrap (%) 38.1% 3.6% 14.3% 0.9% 11.7% 0.7%

Figure 5. Percentage of material technical scrap as a function of part area.

1808 B. A. Soares et al.

Figure 6. Fibre placement cost per part as a function of part area (50 parts/year; 55% of uptime occupation).

Conclusions
With the prevalence of different automated technologies in the production of composite parts for the aeronautic sector, their
costs must be studied to identify and understand the main drivers and main economic domains of these technologies con-
sidering their characteristics. Automated Tape Layup (ATL) and Automatic Fibre Placement (AFP) were studied and
process-based cost models were developed considering technological aspects, such as the required material and losses,
the productivity of the material deposition, and operational issues like production volume and equipment uptime. A repre-
sentative aircraft part was used as a case study. For the geometry analysed, the results show that AFP is both faster and gen-
erates less technical scrap, but ATL has a lower cost, due to lower material-specific costs, which, given the volumes of
production studied, is in line with Hagnell and Åkermo (2015) findings. Exploration of some variables shows that a reduction
of 5% in material cost differential, or an increase in technical scrap, makes AFP less costly than ATL. The part size evaluation
is consistent with (Grimshaw, Grant, and Luna Diaz 2001) findings, with scrap rates decreasing and eventually stabilising
with area increase, although this effect is geometry dependent. AFP has other advantages outside the scope of this article,
the most important being the possibility of producing parts with more complex geometries, increasing the ability to
produce more parts. Although the paper sheds some light on the subject, one possible line of future research is the relation
between part geometry and scrap and its influence on the choice of the technological process.
The results achieved for the comparative analysis of both technologies were validated with the industry manufacturer that
provided some of the operations and cost data, which supports the relevance of this study based on real data for the academic
and industrial world.

Disclosure statement
No potential conflict of interest was reported by the authors.

Funding
The authors gratefully acknowledge the funding by Fundação para a Ciência e Tecnologia, FCT, Portugal , under grants [CMUP-ERI/TPE/
011/2013] and [MITP-TB/PFM/0005/2013].

ORCID
B. A. R. Soares http://orcid.org/0000-0003-2737-1154

References

Apostolopoulos, Panayotis, and Christos Kassapoglou. 2002. “Recurring Cost Minimization of Composite Laminated Structures –
Optimum Part Size as a Function of Learning Curve Effects and Assembly.” Journal of Composite Materials 36 (4): 501–518.
doi:10.1177/0021998302036004556.
Bader, Michael G. 2002. “Selection of Composite Materials and Manufacturing Routes for Cost-Effective Performance.” Composites Part
A: Applied Science and Manufacturing 33 (7): 913–934. doi:10.1016/S1359-835X(02)00044-1.
Barbosa, G. F., and J. Carvalho. 2013. “Analytical Model for Aircraft Design Based on Design for Excellence (DFX) Concepts and Use of
Composite Material Oriented to Automated Processes.” The International Journal of Advanced Manufacturing Technology 69 (9):
2333–2342. doi:10.1007/s00170-013-5211-7.
 International Journal of Production Research 1809
Baumers, Martin, Luca Beltrametti, Angelo Gasparre, and Richard Hague. 2017. “Informing Additive Manufacturing Technology
Adoption: Total Cost and the Impact of Capacity Utilisation.” International Journal of Production Research 55 (23): 6957–
6970. Taylor & Francis. doi:10.1080/00207543.2017.1334978.
Campbell, F. C. (Flake C.). 2004. Manufacturing Processes for Advanced Composites. Oxford: Elsevier. ISBN: 1856174158.
Croft, Kaven, Larry Lessard, Damiano Pasini, Mehdi Hojjati, Jihua Chen, and Ali Yousefpour. 2011. “Experimental Study of the Effect of
Automated Fiber Placement Induced Defects on Performance of Composite Laminates.” Composites Part A: Applied Science and
Manufacturing 42 (5): 484–491. Elsevier Ltd. doi:10.1016/j.compositesa.2011.01.007.
Curran, R., S. Raghunathan, and M. Price. 2004. “Review of Aerospace Engineering Cost Modelling: The Genetic Causal Approach.”
Progress in Aerospace Sciences 40 (8): 487–534. doi:10.1016/j.paerosci.2004.10.001.
Debout, Pierre, Hélène Chanal, and Emmanuel Duc. 2011. “Tool Path Smoothing of a Redundant Machine: Application to Automated
Fiber Placement.” CAD Computer Aided Design 43 (2): 122–132. Elsevier Ltd. doi:10.1016/j.cad.2010.09.011.
Field, Frank, Randolph Kirchain, and Richard Roth. 2007. “Process Cost Modeling: Strategic Engineering and Economic Evaluation of
Materials Technologies.” JOM Journal of the Minerals Metals and Materials Society 59 (10). doi:10.1007/s11837-007-0126-0.
Fuchs, Erica R. H., Frank R. Field, Richard Roth, and Randolph E. Kirchain. 2008. “Strategic Materials Selection in the Automobile Body:
Economic Opportunities for Polymer Composite Design.” Composites Science and Technology 68 (9): 1989–2002. doi:10.1016/j.
compscitech.2008.01.015.
Grant, Carroll. 2006. “Automated Processes for Composite Aircraft Structure.” Edited by Faye Smith. Industrial Robot: An International
Journal 33 (2): 117–121. Emerald. doi:10.1108/01439910610651428.
Grimshaw, Michael N., Carroll G. Grant, and Jose Manuel Luna Diaz. 2001. “Advanced Technology Tape Laying for Affordable
Manufacturing of Large Composite Structures.” International SAMPE Symposium and Exhibition 46 (4): 2484–2494.
Hagnell, M. K., and M. Åkermo. 2015. “A Composite Cost Model for the Aeronautical Industry: Methodology and Case Study.”
Composites Part B: Engineering 79: 254–261. doi:10.1016/j.compositesb.2015.04.043.
Kaufmann, Markus, Dan Zenkert, and Christophe Mattei. 2008. “Cost Optimization of Composite Aircraft Structures Including Variable
Laminate Qualities.” Composites Science and Technology 68 (13): 2748–2754. doi:10.1016/j.compscitech.2008.05.024.
Kirchain, R. E. 2001. “Cost Modeling of Materials and Manufacturing Processes A2 – Buschow, K.H. Jürgen.” In Encyclopedia of
Materials: Science and Technology (2nd ed.), edited by Robert W Cahn, Merton C Flemings, Bernhard Ilschner, Edward J
Kramer, Subhash Mahajan, and Patrick Veyssière, 1718–1727. Oxford: Elsevier. doi:10.1016/B0-08-043152-6/00310-7.
Krachangphiphop, Pailin, Jessada Wannasin, and Napisphon Meemongkol. 2017. “Process-Based Cost Modelling for Gas Induced Semi-
Solid-Processed Below-Knee Prosthesis.” International Journal of Production Research: 1–8. Taylor & Francis. doi:10.1080/
00207543.2017.1364441.
Lin, Than, Jae-Woo Lee, and E. L. J. Bohez. 2012. “New Integrated Model to Estimate the Manufacturing Cost and Production System
Performance at the Conceptual Design Stage of Helicopter Blade Assembly.” International Journal of Production Research 50 (24):
7210–7228. doi:10.1080/00207543.2011.644818.
Lukaszewicz, Dirk H. J. A., Carwyn Ward, and Kevin D. Potter. 2012. “The Engineering Aspects of Automated Prepreg Layup: History,
Present and Future.” Composites Part B: Engineering 43 (3): 997–1009. doi:10.1016/j.compositesb.2011.12.003.
Marsh, George. 1996. “Aerospace Composites – The Story so Far.” Reinforced Plastics 40 (9): 44–48. doi:10.1016/S0034-3617(98)80192-
6.
Marsh, George. 2011. “Automating Aerospace Composites Production with Fibre Placement.” Reinforced Plastics 55 (3): 32–37. doi:10.
1016/S0034-3617(11)70075-3.
Mistry, J., and F. Levy-Neto. 1992. “The Behaviour of Repaired Composite Domes Subjected to External Pressure.” Composites 23 (4):
271–277. doi:10.1016/0010-4361(92)90188-Z.
Nadeau, Marie-Claude, Ashish Kar, Richard Roth, and Randolph Kirchain. 2010. “A Dynamic Process-Based Cost Modeling Approach to
Understand Learning Effects in Manufacturing.” International Journal of Production Economics 128 (1): 223–234. doi:10.1016/j.
ijpe.2010.07.016.
Rezaie, K., B. Ostadi, and S. A. Torabi. 2008. “Activity-Based Costing in Flexible Manufacturing Systems with a Case Study in a Forging
Industry.” International Journal of Production Research 46 (4). 1047–1069. Taylor & Francis Group. doi:10.1080/
00207540600988121.
Roy, R., P. Souchoroukov, and T. Griggs. 2008. “Function-Based Cost Estimating.” International Journal of Production Research 46 (10):
2621–2650. doi:10.1080/00207540601094440.
Schubel, P. J. 2012. “Cost Modelling in Polymer Composite Applications: Case Study – Analysis of Existing and Automated
Manufacturing Processes for a Large Wind Turbine Blade.” Composites Part B: Engineering 43 (3): 953–960. doi:10.1016/j.
compositesb.2011.11.036.
Singh, U. K., Manish Dwivedi, and Elsevier. 2009. “Manufacturing Processes.” In Manufacturing Processes, 1–60. Delhi: Pearson
Education.
Storck, Joakim. 2010. “Exploring Improvement Trajectories with Dynamic Process Cost Modelling: A Case from the Steel Industry.”
International Journal of Production Research 48 (12): 3493–3511. doi:10.1080/00207540902725288.
Umer, R., S. Rao, J. Zhou, Z. Guan, and W. J. Cantwell. 2016. “The Low Velocity Impact Response of Nano Modified Composites
Manufactured Using Automated Dry Fibre Placement.” Polymers and Polymer Composites 24 (4): 233–240. doi:10.1108/
01439910610651428.
1810 B. A. Soares et al.
Appendix

Table A1. Process input variables for the baseline test case.

Process variables ATL AFP

Machine acquisition € 1,000,000 1,000,000
Power consumption kWh 20 20
Space required m2 50 50
Set-up time h 2 2
Tooling cost € 30,000 30,000
Consumables cost €/m2 10 10
Material cost €/m2 24.3 33.0
Scrap disposal cost €/kg 3.0 3.0
Material areal density g/m2 300 300
Operators allocated nr 2 2
Reject rate Parts per million 10,200 10,200
Available time h/day 22.4 22.4
Exogenous variables
Working days per year Days/year 240
Wage operators (include benefits) €/hr 15
Unit energy cost €/kWh 0.08
Equipment life years 15
Annual maintenance cost % of annual equipment cost 10
Indirect workers/direct worker Nr 0.6
Building unit cost €/m2 1500
Building life years 30
Opportunity cost annual rate % 15