Beruflich Dokumente
Kultur Dokumente
A thesis presented to
the faculty of
In partial fulfillment
Master of Science
Aditya Mahadik
August 2018
by
ADITYA MAHADIK
Dale Masel
Dennis Irwin
ABSTRACT
Cost estimation of additive manufacturing is helpful for designers and engineers at the
preliminary design stage to understand the cost impact of alternative designs. There are
several models of cost estimation of additive manufacturing, but most of them require
limited number of primary user input parameters. The break-down approach is used to
calculate the total cost which is the addition of the cost components: machine, material,
labor, and post-processing. These primary user input parameters are used to calculate
can be changed by the user depending on the availability of those part or process
based on time and cost. The build time is calculated by studying the machine activities
for preparation of a single layer on the selected process and multiplying it by the total
number of layers. The methodology incorporates the yield of the AM process, the
uncertainty in the number of hours of machine operation per day, the machine life, and
estimation tool (AMCET). The comparison of actual and estimated build time is
4
presented for FDM, SLA, and Polyjet. The actual cost of the parts manufactured using
Polyjet is compared with their estimated cost. The accuracy of the model lies within
average error of 5% for time estimation and 20% for cost estimation. This research
supports the quick and accurate cost estimation with little information for parts
DEDICATION
I dedicate my thesis to Industrial and Systems Engineering faculty and students at Ohio
University. The support received from everyone at Ohio University was great.
6
ACKNOWLEDGMENTS
I would like to express sincere gratitude to my advisor Dr. Dale Masel for his
continuous support, motivation, patience, and immense knowledge. His guidance helped
all the time during the research and writing this thesis. I could not imagine having a better
I would like to thank the rest of thesis committee: Dr. Dusan Sormaz, Dr. Tao Yuan,
and Dr. David Stowe for their feedback, comments, and encouragement.
work in lab and supporting with valuable information. I appreciate the support extended
research.
I thank my friend and colleague Aniruddha Joshi for his excellent collaboration. He
supported me greatly and was always willing to help. Also, I thank all my friends at Ohio
Most Importantly, I would like to thank my family and especially my wife Bhakti.
Due to her inspiration, support and affection I could pursue my masters and thesis
audaciously.
7
TABLE OF CONTENTS
Page
Abstract ............................................................................................................................... 3
Dedication ........................................................................................................................... 5
Acknowledgments............................................................................................................... 6
LIST OF FIGURES ............................................................................................................ 9
LIST OF TABLES ............................................................................................................ 10
CHAPTER 1: INTRODUCTION ..................................................................................... 11
1.1 Background ............................................................................................................. 11
1.2 Motivation ............................................................................................................... 13
1.3 Objective ................................................................................................................. 15
CHAPTER 2: LITERATURE REVIEW .......................................................................... 17
2.1 Cost Estimation Techniques ................................................................................... 17
2.1.1 Intuitive Techniques......................................................................................... 17
2.1.2 Analogical Techniques ..................................................................................... 19
2.1.3 Analytical Techniques...................................................................................... 21
2.1.4 Parametric Techniques ..................................................................................... 23
2.1.5 Software Cost Estimation Tools ...................................................................... 24
2.2 Methods of Additive Manufacturing ...................................................................... 26
2.2.1 Material Extrusion ........................................................................................... 26
2.2.2 Vat Photo-polymerization ................................................................................ 28
2.2.3 Material Jetting ................................................................................................ 28
2.3 Cost Estimation of Additive Manufacturing ........................................................... 29
2.3.1 Break-down Approach ..................................................................................... 30
2.3.2 Activity Based Approach ................................................................................. 33
2.3.3 Parametric Approach........................................................................................ 37
2.4 Time Estimation of Additive Manufacturing.......................................................... 38
CHAPTER 3: METHODOLOGY .................................................................................... 39
3.1 Approach ................................................................................................................. 39
3.2 Machine Cost .......................................................................................................... 43
8
LIST OF FIGURES
Page
LIST OF TABLES
Page
CHAPTER 1: INTRODUCTION
1.1 Background
Additive Manufacturing (AM) is the technique of producing parts directly from three-
dimensional Computer Aided Design (3D CAD) models by adding thin layers of material
over each other, resulting in the desired final geometry. AM simplifies the process of
producing complex parts because extra tools and fixtures are not required. AM requires a
basic understanding of the part geometry, the machines, and the materials to use the
technology. Because of this easy usage, AM technology is becoming widely popular. The
Moreover, AM is not only limited to prototyping or modelling, but also used for
Traditional manufacturing requires the analysis of the features to determine the order
of fabrication, the process requirements, and the needed tools and fixtures, whereas these
tasks can be avoided in AM. To convert the raw material into a finished good, traditional
manufacturing often requires the product to flow through different workstations, whereas
in AM the multiple workstations may not be required. Often, part(s) can be printed on a
can model the desired part geometry. Once the part model is complete, one can proceed
with printing that part. The 3D model needs to be converted into a file format called STL
machine. STL is a standard data transmission format which approximates the surface of
12
the solid model with triangles. Once the CAD file is converted to STL, necessary
manipulations can be made to ensure its feasibility for printing and generating the support
structure. After loading the file to the AM machine, required parameters can be set on the
machine to build the part. Depending on the design, shape, layer thickness, and various
configured parameters, the AM machine builds the part layer by layer. After the part is
built, it can be removed and treated with various post-processing activities, such as
cleaning, removing the support structure, and super-finishing based on the application
[1].
agreement between ISO and ASTM to create a common set of ISO/ASTM standards.
This standard describes the seven different categories of AM processes [3]. These seven
categories are mentioned and described further. (1) Binder jetting is the process in which
powder material is bound together with the deposition of adhesive particles wherever
bonding is needed. (2) Direct energy deposition is the process in which material is melted
and fused using an energy source such as laser or electron beam to deposit layers to form
an object. (3) In material extrusion processes, material is melted inside the chamber and
deposited through the nozzle to form an object in multiple layers. (4) In material jetting
the minute particles of the material are deposited over each other in the layers to form a
desired object. (5) Powder bed fusion uses thermal energy to selectively fuse the powder
present in the powder bed. (6) Sheet lamination is the process of forming an object from
sheets of material using adhesive. (7) Vat photo-polymerization cures the liquid
conventional manufacturing. Parts can be transferred digitally and printed near the
manufacturing facility where they are needed to reduce the transportation cost. The lead
time for AM products is relatively less compared to traditional manufacturing since the
final product is printed on a single machine. Less waste and rework enable AM to have
better environmental effects. Designers and engineers can create designs in an innovative
way so that cost-effective options can be explored. Further, designers and engineers can
make the prototypes to analyze the performance so that decisions for development and
processing steps, assembling activities, and reducing inventory. The AM technology has
healthcare, and dental [2], [5]. Due to these noteworthy applications of AM, it can be
1.2 Motivation
Cost Estimation (CE) is the process of predicting the cost of a product before it is
manufactured. The cost of the product can be estimated using techniques such as
quantitative models or historical data [4], [8]. An estimate of the product cost should be
known for making decisions, such as identifying business potential, performing the
break-even analysis, analyzing the contribution of resources, and value engineering (cost
reduction). To make the decision about whether to buy or manufacture the products, the
making design changes in the product, forecasting business prospects, and devising
methods of production. Costing can help sales and marketing divisions of a company to
because an overpriced quotation could result in the loss of an order and an underpriced
associated with the cost, such as the material, the labor, the machine, and the tooling.
These cost components contribute in different percentages to constitute the product cost.
the use of machine, or a variation in the materials. It is challenging for designers and
manufacturing technique if the cost associated with the cost components is not known.
network.
Studies have been conducted to analyze the cost contributors based on the energy
consumption, the production capacity, the time and the relevant activities involved [6],
[8]. Most of the existing models developed are dedicated to specific AM technologies
There are some models which estimate the cost for any AM technology, although
Additionally, the existing models have not combined the effect of factors such as the
15
speed of AM the machine, the life of the machine, the expected working hours, the
quality of the final product, the recycling rate of the material, and the manual activities
involved.
The estimated global revenue of AM goods was around $967 million in 2013. Out of
this, the US contributed $367 million, or 38% to the global production [2]. It is also
estimated that the revenue of the AM market will reach approximately $50 billion by
2030 and $100 billion by 2040. This indicates that there is an enormous potential for AM.
The growth in AM will occur if the technology and infrastructure supporting AM makes
progress [2]. Additionally, in the detailed analysis, it is identified that the potential
1.3 Objective
The objective of this thesis is to develop a cost estimation model for parts made
through AM using a break-down approach. The total cost is addition of the cost
components: the material, the machine, the labor, and the post-processing. The total cost
will be calculated by primary user parameters which are limited in number. If additional
part or process parameters are available, the user can enter the secondary input
Secondary input parameters are dependent on the values for the primary input
parameters. The cost component parameters are used to estimate the sub-component cost
for material, machine, labor and post-processing. The cost component parameters are
calculated, based on the primary and the secondary input parameters. By combining
16
minimum, most likely, and maximum values as a user input to determine expected
machine operation hours for estimating the machine cost per hour.
The building time is estimated by considering the total number of layers and the time
to prepare a single layer. The single layer preparation time is estimated by finding four
critical activities. First, the time to lower the platform; second, the time to move the
machine (print) head; third, the time to deposit the layer, and finally, the time to fuse the
material to form a layer of an object. The specific model of time estimation is developed
for individual processes (FDM, SLA, and Polyjet) by analyzing activities undergone by
This section discusses the classification of the cost estimation techniques. There are
numerous classifications available for CE according to the approach type, the granularity
level, and the phase of applicability [7]. The approach type categorizes CE methods into
qualitative and quantitative which are commonly used in the literature [8], [9], and [10].
Granularity level classifies CE into top-down and bottoms-up approaches. The phase of
applicability classifies models into early prediction and late estimation based on the data
Niazi et al. [8] presented a hierarchical classification of product cost estimation (PCE),
Intuitive techniques are developed based on the knowledge of an expert to make rules,
decisions, and judgements to prepare a cost estimate. Depending on the experience, the
18
domain expert can predict the cost of the product with certain information. This technique
Case-based techniques generate the estimate based on the past designs which are
similar to the new design. This system uses the past design and makes necessary changes
to meet the requirements of the new design. This approach is useful to make an estimate
reasoning was applied by Ficko et al. [11] to develop an intelligent system to provide the
cost of the tool used for tool manufacturing. The cost was predicted by using similar
features from previous CAD models for which the cost was known. This technique can
only be used when similar past design data is available to generate an estimate.
making during the cost estimation process for evaluating design alternatives. This
and design functions to prepare rules or logical statements. This technique is further
classified as rule based, expert system, or fuzzy logic as shown in Figure 2 [8].
A rule-based system uses the process time and the cost calculation of the part based on
design and manufacturing constraints. This approach requires a set of experts in the area
and may be time consuming. The model developed by Gayretli et al. [12] calculated the
process time and the product cost depending on the user constraints such as
knowledge in the database to mimic the human expert. Venkatachalam et al. [13]
processes based on the design and the production constraints. Additionally, their system
could determine the manufacturing cost of the different processes. This system may give
A fuzzy logic system can handle the uncertainty involved in CE. The system is formed
by identifying a set of input variables and an output variable (cost) to develop the model.
The fuzzy model determines fuzzy inference by a defined set of rules and these
inferences can be used to predict the cost. Predicting complex features using fuzzy logic
Analogical techniques are part of qualitative techniques which are based on historical
data to derive an estimate. Analogical cost models are further sub-classified as regression
features such as height, length, width, volume, and weight to determine a linear
relationship with the cost of the product. This technique is mostly reliable when similar
cost data and necessary product parameters are available [8]. Ruffo et al. [15] considered
parts’ height, volume, and a bounding box to develop a build-time estimator using
models are applied where a relationship between input and output parameters is
unknown. The training set — that is, the number of examples observed from the system
— is used for learning. The build time for fused deposition modelling was estimated
using ANN by Di Angelo and Stefano [16] using parts’ geometrical data. Availability and
reliability of the data are the major limitations of this technique [8].
21
Analytical techniques are based on the summation of the components of the resources
that are utilized during production. The product cost is expressed in terms of operations
and activities involved during manufacturing the product [10]. Analytical techniques are
specification of the part along with the manufacturing processes. Singh [17] introduced a
model based on design tolerance with multiple objectives: to minimize the unit cost, to
minimize the quality loss, and to minimize the lead time. The cost-tolerance relations
were developed to determine the product cost. These models are useful to recognize cost-
Activity-based cost estimation is prepared from the cost incurred by the activities
the activities and estimating the cost elements. Activities are classified as machine-based,
labor-based, technical, and administration-service-based. The unit rates are used with
22
activity times to generate an aggregate cost estimate [18]. This model is widely used in
processes. This method is easy to implement because the cost of activities can be
determined effectively.
The break-down approach calculates the product cost by identifying the resources
consumed during manufacturing the product and then adding each to derive the unit cost.
Son [19] developed a cost model using the break-down approach which classified the
costs as well-structured costs and ill-structured costs. The well-structured costs included
production costs such as labor, material, depreciation, machine, tool, floor space, and
computer software. The ill-structured costs were further classified as the quality and the
flexibility cost. The quality cost included expenses incurred for inspection and prevention
of defects. The flexibility cost was addition of the cost incurred for setup, waiting, idle
time of the machine, and inventory carrying cost. The break-down approach is generally
easy to adopt although it requires detailed information of the resources consumed [8].
the part. The cost model is generally the function of time required to manufacture the
part. Jung [20] proposed a cost model which considers operation time, set-up time, and
𝑇
Manufacturing Cost = (𝑅𝑜 + 𝑅𝑚 ) [( 𝑄𝑠𝑢) 𝑇𝑜𝑡 + 𝑇𝑛𝑜 ] + 𝑀𝑐 + 𝐹𝑒 (1)
where,
Ro = Operator’s rate
23
Rm = Machine rate
Q = Batch size
Mc = Material cost
Fe = Factory expenses
Operation-based cost models can help to prepare alternate process plans to optimize
results; however, these models require exhaustive design and process planning data.
adding up the respective cost associated with each feature. The features can be design-
materials used, whereas process-related can be machining, casting, and molding. This
methodology allows the user to select the feature as either design or manufacturing so
that cost can be predicted [21]. This way, product cost can be estimated by recognizing
features and allocating the cost to generate an estimate. This method is effective to
determine the cost for higher features; however, it can be difficult to identify the cost for
functional or technical characteristics and the cost. The parametric approach considers the
Estimation Relationships (CERs) are developed to establish the cost model [22]. The
models can estimate the cost during the preliminary design stage using single or multiple
CERs. The estimates can be generated quickly during very early stages of design using
Duverlie and Castelain [10] estimated the cost of a diesel engine piston using the
parametric method and the case-based reasoning method. The cost estimation formulas
were determined using linear and second degree polynomial regression. The results
concluded that using the parametric method, the cost can be estimated very quickly, but
the reason for the root cause is unknown. This is because of the principle of regression
which gives the general trend. On the other hand, case-based reasoning allows the user to
consider the particular case and gives more accurate results [10]. The parametric method
is useful where the cost drivers (parameters) are known; however, it is ineffective where
Chan et al. [23] developed a model using big data analytic tools to estimate the
manufacturing cost of AM products based on similar products in the past. To estimate the
cost, the CAD model can be submitted online to the cost analytic service provider, who
has the database and the data mining tools. The 17 feature vectors were extracted from
the uploaded model based on the geometry of the part, the manufacturing process, and the
material. These feature vectors were used as an input to a predictive model to estimate the
cost. The predictive model was developed using machine learning algorithms and
The results revealed that the costs predicted using the big data analytic tool were
within maximum of 7% error. However, the availability of the relevant product and
production data are the major limitations of the data-driven model. Additionally, the
accuracy of the model is dependent on the quality and quantity of the data used for
developing the model. The developed model does not consider post-processing activities,
A hierarchical cost estimation tool was developed by Koonce et al. [24] to estimate the
cost of manufacturing the part at any stage of design. The developed tool supports design
time cost estimation using Federated Intelligent Product Environment (FIPER). This
research was the part of National Institute of Standard and Technology (NIST) and
Advanced Technology Program (ATP). The tool was developed with the capability to
During the early stage of design, an estimate can be generated by considering simple
parametric cost relations. However, in the later stage of design, an estimate can be
developed by considering design features or process plans. The FIPER cost estimation
tool was framed with a cost engine, a graphical user interface (GUI) and an element
builder. Elements (part, operation or inspection) are identified with costs in the FIPER
environment. These elements are coded as java classes and stored in the library. To
generate an estimate, elements are linked to design or process parameters and total cost is
estimated by addition of all the cost elements. The model was tested with 80 large
rotational jet engine parts and concluded that the accuracy of an estimate was improved
by 22% [24].
26
helps a manufacturer to estimate the cost so that design and manufacturing decisions can
solutions. The cost estimation model is a knowledge base and uses parametric cost
approach on time and motion studies, company project history, and cost relationship
SEER-MFG can generate a cost estimate directly from a 3D CAD model. The
software provides the cost break-down for labor, material, and tooling by considering
setup, operation, inspection, and rework. Further, the software enables users to select an
generate detailed estimate. Finally, SEER-MFG provides various charts, graphs, and
There are several techniques available for additive manufacturing and they are divided
in to seven categories according to ASTM 52900:2015 [3]. Each process has a unique
along with the working principle and steps involved during manufacturing.
The machine is loaded with material in the form of spool. The material is simultaneously
27
drawn through the nozzle and heated in the liquefier chamber, to deposit layer by layer as
shown in Figure 5 [26]. The nozzle moves in a horizontal direction to deposit the
material, then the platform moves vertically down after depositing the layer. Since the
material is added through a nozzle, the pressure and speed must be kept constant to get
the semi-liquid phase from the nozzle and directly prints the part, wastage is minimum in
this process. The FDM process has restricted accuracy due to the use of filament and
nozzle radius dimension. Since the cross-sectional area needs to be filled by part material
which uses a vat of resin polymer to build the object layer by layer. An ultraviolet (UV)
laser is used to harden or cure the required resin to form the cross-section of an object.
mirror. The platform is moved downward after each layer is cured to bring a fresh resin
SLA gives high accuracy and surface finish. A wide range of materials are available
necessary [29].
jetted on the build platform through a nozzle in the form of drops or continuously as
shown in Figure 7 [30]. The platform is lowered down once material is deposited so that
29
a new layer can be deposited to build a part layer by layer. The deposited material is
This process achieves high accuracy due to minute droplet deposition. Also, polyjet
allows the use of multiple materials together for producing the part. Support material is
There are several models for additive manufacturing cost estimation which are based
costing model to determine the cost [38-42]. This approach seems to be pragmatic,
detailed, and provides better results. However, there are other models including
general CE for AM. Below the CE models developed for AM are discussed.
30
In 1998, Alexander et al. [33] developed a generic CE model for AM using the break-
down approach. The model calculated the total cost by adding the cost involved in pre-
The pre-build cost considered the time required for entering the parameters into the
AM software, positioning and scaling the part, generating the support structure, slicing
the part, and creating the road path by AM software. Build cost considered the cost of the
part including the support structure, which was the product of the time to build and the
cost of running the machine. Further, the model included the labor cost to remove the
support structure, finishing activities, and cost of extra material required during post
processing.
Alexander et al. also developed specific models for cost estimation of FDM and SLA.
They calculated the cost with different orientation to determine the lowest cost. The
model assumes that the activities of pre-build, build, and post-build are independent, but
these might be inter-dependent. This model has not accounted for the factors such as life
determine the total cost for producing the parts through AM. They made a comparison of
parts manufactured by AM with injection molding. The total cost was broken-down into
rate, platform build time, production volume, depreciation of machine, and equipment.
31
Labor cost was calculated by considering parameters such as build time, machine
operator cost per hour, setup time, and post processing time. Material cost was generated
by multiplying the weight of material including the support and the material cost per
kilogram. The cost of manufacturing the part by injection molding was obtained from a
quotation.
The results showed that AM technology is preferred for low volume production and
may compete with injection molding in terms of cost for relatively high volume. Results
also concluded that machine cost plays a major role in production cost. Hopkinson and
Dickens [34] considered machine uptime as 90%, which means around 46 weeks of
consumption and space rental were ignored during the cost estimation [6].
Yim and Rosen [35] developed a cost estimation model using the break-down
approach which integrated overall cost by addition of four sub-costs: purchase cost,
machine operation cost, material cost, and labor cost. Purchase cost was derived
considering build time. Operation cost was the product of build time and operation rate.
Build time was calculated by adding recoating time, material processing time, and delay
time. Material cost was estimated by considering part volume, material rate, support
structure factor, recycling factor, and number of parts. Labor cost was the product of
Cost estimates were generated for SLA, FDM, and Ink-jet to compare with the quoted
prices. The results exhibited that the estimated model predicted the cost with average
32
error of 18.2%. However, the models can be used for cost comparison and selection of
rejection rate. Also, it was assumed that machine up time was 95% during useful life
Piili et al. [36] used the break-down approach to determine the total cost and build
manufacturing cost was the addition of direct cost, indirect cost, and energy consumption
for selective laser sintering process. The build time was computed by software supported
by AM machine. They compared two scenarios for platform filling: first, creating a single
part at a time and second, completely filling the platform with 40 pieces.
The results showed that time per piece when manufacturing a single part was 529
minutes whereas for multiple pieces it was 102 minutes. Hence, it was evident that
manufacturing time was heavily dependent on recoating time (the time to move the print
head) of the AM machine. The recoating time remains constant even if the number of
pieces were changed. As observed by other researchers, Piili et al. [36] confirmed that
machine cost was the highest contributor in total cost, but could be reduced to 6% by
building multiple parts as compared with a single part. Therefore, optimal utilization of
the building platform is a very important factor that the user should consider while
multiple or mixed parts (with different geometry) was developed by Rickenbacher et al.
33
[37]. The cost model used the break-down approach developed by Alexander et al. [33]
including pre-build, build, and post build activities. Additionally, this model calculated
the total cost by adding the cost of preparing geometric data, the cost of building the job
assembly, the cost of machining setup, the cost of building up the part, the cost of
removing the part from the selective laser melting (SLM) machine and the cost to
separate parts from the substrate plate. To estimate the build time, Rickenbacher et al.
This cost model can be used to estimate the cost of manufacturing multiple parts with
similar or different part geometries to reduce setup time and manufacturing time. Due to
this, the total manufacturing cost per part can be reduced. The model by Rickenbacher et
al. [37] does not consider warm-up and cool-down time. This model considered a
uniform hourly rate of labor and does not consider the operator’s skill level. It also does
not consider the cost of energy for the SLM process. It can also be noted that the time
Lindemann et al. [38] used ‘time driven activity-based costing’ which considered
different influence factors based on utilization of resources. The model was developed by
considering four main processes: preparation of the part, production of the part, labor in
removing part and support structure, and post-processing to improve material properties.
The cost of manufacturing was estimated by adding fixed cost and machine operation
cost. Fixed cost included expenses incurred in gas consumption and labor cost. Machine
operation cost was calculated by multiplying the machine hourly rate and the time to
34
build the part. Material cost was calculated in similar way as calculated by Ruffo et al.
[43].
The results of the Lindemann et al. [38] model concluded that machine cost was the
largest contributor to the total cost followed by material cost. The third cost driver was
data preparation, for pre- and post-processing activities [6]. The various assumptions
made regarding build rate, utilization rate, machine investment cost, and material cost can
Baumers et al. [39] developed a CE model by considering direct cost, indirect cost,
and energy consumption for the SLS process using activity-based costing. The direct cost
was material cost, which was calculated by multiplying the material weight (including the
support structure) and the material rate. The indirect cost rate (which contains machine
cost, production labor, and administrative overheads) was multiplied by the build time.
The energy consumption estimate was multiplied by the rate of electricity to obtain the
They also used a regression model to estimate the time and energy consumed by a
single layer depending on the geometry of the part. The results confirmed that estimated
time and energy were within 10% of the error as compared with actual machine values.
The combined estimate of cost and energy consumption may be utilized to examine the
effect on different parts [39]. This model did not consider many factors such as build
preparation, machine cleaning, and post processing, which could increase the accuracy of
an estimate.
35
costing and implemented into a software tool. The model evaluated the process cost for
(cost drivers) for sample parts. They developed the model by considering recycling of
waste material, the support structure, the calculation of manufacturing time, the
complexity involved in product design, time for post-processing, and the integration of
identified seven main processes for the cost model which were: (1) design and planning;
(2) material processing; (3) machine preparation; (4) manufacturing; (5) post-processing;
(6) administration; and (7) sales and quality. The cost calculation tool took 77 different
The sensitivity analysis demonstrated three major findings: (1) the investment cost of
the machine played a major role; (2) optimization can be achieved in post-processing
activities for small bodies and high quantities; (3) economies of scale exist for small
products, whereas large products are independent from the ordered quantity. Costabile et
al. [6] mentioned that the cost model of Schröder et al. [39] included the sales and
administration cost; however, the activities involved in sales and administration were not
specified. AM cost model should include only manufacturing cost for cost accounting.
The break-even analysis provided by Atzeni et al. [41] compared selective laser
melting (SLM) with high-pressure die casting and 5-axis machining. An aeronautical
compared to traditional technologies. The analysis revealed that two pairs of aluminium
landing gear were produced within 2.5 days from the availability of the CAD model with
the SLM technique. However, mold and actual production of the part took several weeks
using 5-axis machining. The result of this study demonstrated that AM technology can be
Enhanced activity-based cost estimation approach was adopted by Baumers et al. [42]
on a blower component to compare the unit cost using a selective laser sintering (SLS)
technique against computer numerical control (CNC) machining and tungsten inert gas
(TIG) welding. Their framework integrated four aspects: the optimization of capacity
utilization; identification of ancillary process steps; build failure effect; and design
The basic model considered the direct cost as the product of material price and weight
of the part. Indirect cost was derived from the time to manufacture the blower component
and the cost of energy. The model included the labor cost incurred in setup separately.
Further, the model integrated capacity utilization with the computational build volume
packaging approach to optimize the build capacity. Build failure effect (i.e., failure of
single layer preparation) was incorporated in the model by considering a constant discrete
probability of 0.025%. This probability tree model was integrated in the costing model.
Energy savings arising from design adaptation of AM was modelled in terms of cost,
The results revealed that the largest cost component was indirect cost which
contributed 37%. The indirect cost included machine cost, wire erosion, and overheads.
37
The second contributor of cost was risk of build failure which was 26%, indicating that
process instability can affect AM. Share of cost was due to labor component, post-
manufacturing, the cost saving of around 37.5% was reported. The probability of build
part [42].
Ruffo et al. [43] developed a model based on a parametric and engineering approach
and constructed cost estimation relationships for SLS. The total cost was the addition of
direct cost (material consumed during manufacturing) and indirect cost (time to build the
part). This model improved on the model by Hopkinson and Dickens [34] by considering
machine utilization as 57% which was more realistic. They also considered the labor cost
incurred for making the product and the material recycling limit (thermal treatment of
powder). By considering this model, the break-even point between AM and injection
molding changed from 14,000 parts (as calculated by Hopkinson and Dickens), to 15,500
parts. The significance of material cost reduced to 33% from 78% for the SLS compared
with the old model. This model also concluded that machine investment and its
maintenance contributed 38%, changing from the previous perception of 24%. However,
the model developed by Ruffo et al. ignored the possible machine failure, post-processing
The analytical method was used by Zhang and Bernard [44] to create a generic model
for estimating build time of a part produced by SLS. The total time was computed by
adding machine preparation time, layer drawing time, layer preparation time, and
operation time. The build time was estimated using the height of building parts and the
layer thickness. Zhang and Bernard [44] concluded that producing multiple parts with
different heights could improve the production rate and build time distribution. However,
the authors did not validate this model in an actual production environment, hence
An artificial neural network (ANN) technique was used by Di Angelo et al. [45] to
estimate the manufacturing time using FDM for six test cases which had complex
geometries and required a support structure. The build time estimator system (BTES) was
created based on back-propagation ANN with two hidden layers. The model was
developed using 75 nodes with each hidden layer and output node used to estimate the
manufacturing time. The model was programmed using the MATLAB Neural Network
toolbox [46]. To estimate the manufacturing time, the driving factors were evaluated for
selected models using the STL file. These driving factors were used to train the ANN
using back-propagation training rules. The manufacturing time estimated using this
model was within an error of 20%. The error can be minimized by including the detailed
CHAPTER 3: METHODOLOGY
3.1 Approach
The cost estimation model developed in this research has a hierarchical structure, as
shown in Figure 8. The total cost of the part is broken-down into four main components:
The total cost of the part is calculated by adding machine cost, material cost, labor
where,
Ct = Total cost
AM. The first level consists of the primary user input parameters and the AM technique.
These are minimum inputs required in the system. The second level is the secondary
input parameters which are already configured in the model with the default values. The
values in the secondary level are dependent on the parameters entered by the user in the
first level. However, if the user has information about these secondary parameters, the
The third level shows cost component parameters, which are calculated from the
primary and secondary parameters. These cost component parameters are required to
calculate sub-component cost parameters. The fourth level forms the sub-component
costs calculated from cost component parameters. The fifth level consists of component
cost estimates which are calculated by adding sub-component costs. The final level has
the total cost estimate which is formed by addition of component cost estimates to
Optional
Secondary
Level 2 Parameters
Cost -
Component
Level 3 Parameters
Sub-
Component
Level 4 s Cost
Estimate
Component
Level 5 Cost
Estimate
example, the machine cost calculation is explained below considering different levels.
1. Level 1 consists of the primary user input parameters, which are machine type and
expected machine operation hours per day. These parameters are required from the
thickness, and machine speed which will be selected automatically depending the type
of machine specified in the primary input parameters. For level 1 and 2 there are
additional primary and secondary parameters configured in the model; however, only
3. Level 3 contains component cost parameters which include machine hourly rate,
machine operation cost, machine-built rate, and others calculated based on primary
and secondary parameters. For example, machine hourly rate is calculated using
4. Level 4 consists of the sub-component cost estimates which are machine operation
cost and repair and maintenance cost. Machine operation cost is estimated by
5. Level 5 is the addition of all sub-component costs to generate the machine cost
estimate.
6. Finally, all component cost estimates are added as shown in equation 2 to get the total
cost estimate.
way.
43
The machine cost consists of the cost incurred for owning, operating, and maintaining
the machine used for manufacturing the part. Hence, the machine cost is calculated by the
addition of the sub-component cost parameters, which are machine operation cost and
repair and maintenance cost incurred for the AM machine as represented in Figure 8. The
where,
The machine operation cost is calculated by multiplying the hourly operating cost and
the time required to manufacture the part. Manufacturing the part requires several
activities to be performed on the machine. During these activities, the machine is engaged
and is not available to start on a new part. Hence, the time to perform all those activities
should be considered to estimate the time required to manufacture the part. These
activities are loading material in the machine, setup activities on the machine, part
where,
Tc = Cleaning time
Machine hourly operating cost is used to calculate the machine operation cost.
Machine hourly operating cost is calculated by dividing the equivalent annual purchase
cost of the machine by number of hours operated per year. The machine hourly operating
𝐶𝐸𝐴𝐶
𝐶𝑚𝑟 = (5)
12 ∙ 𝑁𝑑𝑎𝑦𝑠 ∙ 𝐸[𝐻]
where,
The number of working days in a month is a secondary input and the user can edit this
value depending on the number of days for which the machine is expected to operate.
Further, the annual cost of owning is divided by months, number of working days, and
A single value for machine operation hours per day may not be accurate and might
introduce an error. This value is important because the number of machine operation
hours is directly proportional to the machine operation cost. The triangular distribution is
used in this research to determine an accurate estimate and deal with the uncertainty
The triangular distribution is useful since it is easy to estimate the minimum, most
likely, and maximum values that define the distribution. The expected machine operation
hours are calculated using minimum, most likely, and maximum values as an primary
user input in the model. The expected machine operating hours is the mean of all three
where,
The equivalent annual purchase cost of the machine is calculated using the purchase cost
(present worth) of the machine for a specific time-period considering machine life and fixed
rate of interest [48]. The equivalent annual purchase cost is calculated using equation 7.
𝑟 ∙ (𝑃𝑉)
𝐶𝐸𝐴𝐶 = (7)
1−(1+𝑟)−𝑛
where,
r = Rate of interest
It is assumed that the cost incurred for repair and maintenance of the machine is
proportional to the time for which the machine is operating for producing the part. Hence,
machine repair and maintenance cost are calculated by considering the time to perform
46
all those activities for manufacturing the part – loading material in the machine, setup
Therefore, repair and maintenance cost of the machine is calculated using equation 8.
where,
The repairs and maintenance are calculated by considering a certain percentage of the
cost of owning, because it is assumed that repairs and maintenance are proportional to the
equivalent annual purchase cost of the machine. Hence, to simplify the calculation it is
assumed that the annual repairs and maintenance are a fixed percentage of the equivalent
If the cost of repairs and maintenance is not readily available in terms of the
percentage of the cost of owning machine, the user may have to convert it into a
percentage to enter model. It is assumed that the value of annual cost of repairs and
This variable is configured as a secondary input parameter and can be edited by the user
to generate accurate estimate on the availability of the data. Hence the unit cost of repairs
𝑥 ∙ 𝐶𝐸𝐴𝐶
𝐶𝑟𝑟 = 𝑥 ∙ 𝐶𝑚𝑟 = (9)
12 ∙ 𝑁𝑑𝑎𝑦𝑠 ∙ 𝐸[𝐻]
where,
The material cost is calculated by considering the primary inputs given by the user and
default values of secondary input parameters. The primary material parameters include
type of material, part volume, and dimensions. The dimensions of the part include length,
width, and height of the part which specifies the bounding box of the part in X, Y, and Z
planes, respectively. These primary material input parameters are used to estimate the
material cost of the part. The secondary input parameters consist of material density,
support structure factor, rejection rate, and recycling rate. Secondary material parameters
Material cost of the part is estimated using the type of material being manufactured
and the mass of material required including the support structure. Hence, part material
cost is calculated by the addition of the cost of material and the cost of support structure
where,
Product material cost and support structure material cost are calculated by considering
the mass of material consumed and the unit cost of material. Part material cost is
where,
where,
The mass of material required for the part is estimated by using the value provided by
the user for the volume of the part. The total quantity of material consumed for
manufacturing the part is calculated by multiplying volume of the part by the density of
where,
There are two types of support material required for most AM parts. One is the base
support material which is required on the platform so that the part can be manufactured
on the base support structure. This base support material enables easy removal of the part
from the base plate without damaging the part. The other is the part support material,
the part. The part support material and base support material can be the same as the part
49
being manufactured or can be different [49]. These sacrificial support materials are
The mass of base support material depends on the base area of the part, the type of
AM technique, the layer thickness and the material used for manufacturing. Prior to
manufacturing of the actual part, a constant number of layers of base support material is
deposited. The mass of base support material is calculated using equation 14.
where,
Tl = Layer thickness
The base area of the part is calculated by multiplying length and width of the part
The mass of the part support material varies with part design. This is accounted for by
the support factor. The support factor is the ratio of the part support volume to the part
volume. If the support factor is more than one, this means the volume of support material
is greater than the volume of part material; however, if the support factor value is less
The mass of part support material depends on the volume of the part, the density of
support material, and the support factor. The mass of part support material is calculated
where,
Sf = Support factor
The support factor is a secondary user input parameter and the value can be edited by
the user. The density of support material depends on the type of support material.
The value of the support factor varies depending on part geometric dimensions,
value of support, the factor linear equations were developed with part geometrical
parameters. To develop linear equations the support factor was considered the dependent
The part geometrical parameters were the ratio of part volume and bounding box
(V/XYZ); the area of the base of the part (XY); the breadth of the part (Y); and the height
To test if there is a relationship between any of the part geometric parameters and the
support factor, hypothesis testing was used. The significance of the linear regression
model for deriving the support factor was validated using hypothesis testing. It was
assumed that the error component is distributed normally in the linear regression model
and α = 0.05 [50]. To test the hypothesis whether slope equals zero, the following
Null Hypothesis, 𝐻0 : 𝛽1 = 0
Alternate Hypothesis, 𝐻1 : 𝛽1 ≠ 0
51
If we fail to reject the null hypothesis then the average value of Sf will be used for the
model; otherwise the linear equation will be used to calculate the support factor. To test
the hypotheses, the t-Test statistic T0 was used, which is calculated using equation 16.
̂1 −0
𝛽
𝑇0 = (16)
√(𝜎2 ⁄𝑆𝑥𝑥 )
where,
T0 = t-Test statistic
σ = Residual error
The value of T0 was compared to check whether it lay within the critical region using
equation 17.
The linear equation was developed with different parameters using two datasets to
predict the value of the support factor. Table 1 represents the parameters — x values —
used to construct the linear equation, the number of observations, the t-Test statistics, and
Hypothesis
Dataset Parameter Observations T0 Tcritical
test
1 Dataset 1 Ratio (V/XYZ) 11 -1.08 ±2.26
2 Dataset 1 XY 11 0.64 ±2.26
3 Dataset 1 Y 11 0.48 ±2.26
4 Dataset 1 Z 11 1.70 ±2.26
5 Dataset 2 Ratio (V/XYZ) 20 -1.40 ±3.197
52
Using t-Test it is observed that the value of T0 lies between the critical values for all
the parameters, implying that we fail to reject the null hypothesis. So, there is no
Labor cost is calculated by considering the time required for the operator to perform
manual activities. Labor cost is the addition of the cost to set up the machine, the cost to
manufacture the part, and the cost to remove the part. For each activity in manufacturing
a part, labor cost is calculated by estimating the time spent by the operator and the
operator (labor) rate which is the hourly cost of the operator. Labor rate is the primary
user input entered in the model to estimate the labor cost. The time spent by the operator
during setting up the machine, manufacturing the parts, and removing the parts are
configured as secondary user input parameters. The labor cost is calculated as shown in
equation 18.
where,
The cost incurred in setup activities is estimated by multiplying the time for the setup
activities and the hourly labor rate. Setup activities include loading of material into the
machine, setting parameters on the machine, generating the STL file (including necessary
modifications required in the STL file), and adjusting the machine before starting. The
times required for these activities are configured as secondary user input parameters in
the model and are calculated based on the primary input parameters, which are type of
AM technique and machine type. The setup cost is calculated using equation 19.
where,
During manufacturing the part, it may not be necessary for the operator to be present.
The operator may only inspect the part at a regular interval to ensure the process is
progressing correctly. Hence, labor cost during manufacturing is calculated by the time
spent by the operator during manufacturing the part, using equation 20.
where,
of visits required during manufacturing and time spent per visit. Time spent by operator
𝑇𝑚𝑜 = 𝑁𝑣 ∙ 𝑇𝑣 (21)
where,
It is considered that the time spent by the operator during manufacturing a batch, other
than attending the machine, is the auxiliary time (Tn) i.e. time spent on other duties.
Figure 9 shows time spent by the operator during a single visit (Tv) and auxiliary time
(Tn).
It can be seen from Figure 9 that the time spent by the operator during a single visit
(Tv) and the auxiliary time (Tn) make a cycle, which repeats depending on the part
manufacturing time. Hence, the number of visits is calculated by dividing the part
manufacturing time by the time spent during a single visit and auxiliary time, using
equation 22.
55
𝑇𝑚 +
𝑁𝑣 = (𝑇 ) (22)
𝑣 + 𝑇𝑛
where,
Using an equation 22, the number visits will be rounded down up the next integer
considering first visit is accounted. The time spent by the operator during a single visit
and operator non-productive time are the secondary user input in the model.
The cost of part removal is calculated by multiplying the time required to remove the
𝐶𝑝𝑟 = 𝑇𝑟 ∙ 𝑂𝑟 (23)
where,
The time required for part removal is configured as a secondary user input parameter.
Post-processing is the activity performed to complete the part after it is removed from
the platform. These are removing the support structure, cleaning, curing, polishing,
considering the material and the labor time consumed for these activities, shown in
equation 24.
where,
and polishing the part might consume additional material. The amount and the type of
material consumed for post-processing varies. These materials are considered to calculate
where,
Post-processing labor cost is estimated by considering the time required during each
post-processing activity and the operator hourly cost. During different post-processing
activities, the operator hourly cost might vary depending upon skill and knowledge
where,
The cost of materials and operator hourly cost for post-processing are primary user
inputs entered in the model. The amount of material and time required by the operator are
The total cost calculated using equation 2 does not account for defects during additive
manufacturing processes. Some of the manufactured parts may not meet expected quality
standards and may be rejected as defective parts. The cost of defective parts has to be
borne by the acceptable parts and this is accounted in the model by calculating the
yielded cost.
The yield factor (Yf) is defined as the ratio of the number of acceptable parts to the
total number of parts manufactured. Yield factor is used to calculate the yielded total
cost. The total cost is divided by the yield factor to determine the yielded cost —
effective cost per good (acceptable) part after manufacturing [51]. The yielded total cost
where,
Ct = Total cost
Yf = Yield factor
The yield factor affects the total cost inversely. If the number of defective parts
produced during manufacturing is low, the yield factor is high. Therefore, a higher yield
factor will generate a lower cost estimate. The yield factor is configured in the model as a
secondary input parameter, so the user may edit the value corresponding to a specific AM
machine. This enables the user to account for defects of additive manufacturing in cost
estimation.
58
The time required to manufacture a part is estimated by breaking down the processes
for additive manufacturing. The part manufacturing time (Tm) is estimated by determining
the time required to build each layer, then summing these times to get the total time.
The number of layers for the part depends on the layer thickness of the process and
height of the part being manufactured. The layer thickness of the part is determined by
the type of processes and machine specification. The number of layers in the part is
ℎ𝑝
𝑛𝑝 = (28)
𝑇𝑙
where,
Tl = Layer thickness
For manufacturing each layer, the process is broken down into four activities. First,
the machine platform moves down a distance equivalent to the layer thickness. Second,
the machine (printer) head moves in the X-Y plane before deposition of material based on
the part’s geometry. Third, the material is deposited. Finally, the material is fused to form
a layer of the object. These steps may vary; two of the steps might occur simultaneously,
dependent on the specific AM technique. The following section explains the time
In FDM, the material is drawn through a nozzle to deposit a layer on the build
platform. The time required to deposit layer k is calculated by dividing the volume of
layer k by the deposition rate. After deposition of layer k, the material is fused (cured).
The time required to fuse the material of layer k is calculated by dividing the volume of
Material deposition and fusion activities take place simultaneously. The time to
complete the deposition and fusion depends on the part’s geometry, machine
specification, and material characteristics. After the deposition, a layer may not be fused
(cured) even if the deposition is complete. In this case next layer can only be deposited
after the fusion of the previous layer is complete. In other cases, the deposition time will
be more than the fusion time and deposition time will be considered for determining the
time for a single layer. Hence, the time required for the longer activity is considered —
the maximum of the deposition time and the fusion time — to estimate the time to
After deposition and fusion of a layer, the platform is lowered to deposit a new layer.
The time required to lower the machine platform is calculated by dividing the layer
The time required to manufacture a part along with the part support structure using
where,
60
Tl = Layer thickness
The ratio of mass of part support material and density of support material is volume of
𝑉𝑠 = 𝑆𝑓 ∙ 𝑉𝑝 (30)
where,
Hence from equation 31 for any layer k we can derive equation 31.
By substituting the value of Vsk from equation 32 we get, time to manufacture a part
along with part support structure for FDM, as shown equation 32.
𝑛𝑝
𝑉𝑝𝑘 ∙ (1+𝑆𝑓 ) 𝑉𝑝𝑘 ∙ (1+𝑆𝑓 ) 𝑇
(𝑇𝑝 )𝐹𝐷𝑀 = ∑ [𝑚𝑎𝑥 ( 𝑅𝑑
,
𝑅𝑓
) + 𝑆𝑙 ] (32)
𝑘=1 𝑙
and the layer thickness. The deposition rate of material (Rd), the fusion rate of material
(Rf), layer thickness (Tl), and platform lowering speed (Sl) are the secondary user input
61
parameters, configured in the model and calculated based on the primary user input
parameters.
In the FDM process, base support material deposition requires an additional time
which is not accounted for by equation 32. Hence the additional time required to deposit
where,
The times calculated using equation 32 and 33 are added to get the total time required
In SLA, a UV laser scans the resin to form the cross-sectional area of an object by
curing the resin. The time to cure layer k by UV light is calculated by dividing the cross-
sectional area of layer k by the scanning speed of the laser. After scanning the resin, the
platform is moved down to bring a fresh coat of resin material on the cured material. The
time required to lower the platform is calculated by dividing the layer thickness by the
platform lowering speed. A fresh coat of resin is spread across the platform by moving
the blade or roller. The time required to move a blade across the platform is calculated by
The time required to manufacture the part along with part support structure using SLA
where,
The volume of layer k is calculated by multiplying the area of layer k and the layer
thickness. Hence layer volumes can be calculated as shown in equations 35 and 36.
By substituting the value of Vpk and Vsk from equation 36 and 37 into equation 32;
equation 38 is derived.
By dividing Tl from both the sides in equation 37 we can get equation 38.
By substituting the value of Ask from equation 38 into equation 34, the time to
manufacture the part along with part support structure using SLA is derived as shown in
equation 39.
63
𝑛𝑝
𝐴𝑠𝑘 ∙ (1+𝑆𝑓 ) 𝑇 𝐿
(𝑇𝑝 )𝑆𝐿𝐴 = ∑ [ + 𝑆𝑙 + 𝑆 ] (39)
𝑘=1 𝑆𝑆 𝑙 𝑚
The scanning speed of the laser, the length of the platform, and the blade or roller
speed are secondary user input parameters, configured in the model and calculated based
In SLA, the base support material is formed by curing the resin, which takes additional
time, which is not considered by equation 39. The additional time required to
and breadth of the part. The times calculated using equation 39 and 40 are added to get
the total time to manufacture a part along with the support structure.
3.7.3 Polyjet
addition of the times of the activities of Polyjet machine. The machine (printer) head
moves across the cross-sectional area of the layer based on the part’s geometry.
The time required to move the Polyjet machine (printer) head is calculated by
considering the maximum length (movement in the x-direction) and maximum breadth
(movement the in y-direction) of the parts while building on the machine platform.
64
The maximum length travelled by the machine head will be dependent on the number
and size of parts being manufactured on the build platform. The total length travelled by
the machine head for manufacturing a single layer will be dependent on the length from
the default position of printer head, the gap between the parts, the maximum length, and
the maximum breadth of all the parts as shown in Figure 10. The machine head
movement is calculated by dividing the length travelled by the machine head by the speed
After a layer of material is deposited, the platform moves down to deposit a new layer
of the part. The time required to move the platform is calculated by dividing the layer
The build time to manufacture the part using Polyjet technology is calculated using
equation 41.
𝑛𝑝 +𝑛𝑠
𝐿 𝑇
(𝑇𝑝 )𝑃𝐽𝑇 = ∑ [( 𝑘 ) + 𝑙 ] (41)
𝑘=1 𝑆ℎ 𝑆 𝑙
where,
The speed of the Polyjet machine (printer) head movement, the deposition rate of
material, the layer thickness, and the platform lowering speed are the secondary user
input parameters, configured in the model and selected based on the primary user input
parameters. The time required to construct the base support material is dependent on the
number of additional layers of base support structure (ns). The part support material is
65
deposited along with the part material. Therefore, the time required to generate the part
equation 42.
The length travelled by the machine head is explained in the example below. Consider
three parts with different geometries and dimensions being manufactured on a Polyjet
machine as shown Figure 10. The machine head is shown at the default position on the
The distance travelled by the machine head in the x-direction on every pass is
equivalent to the length of the platform (Lx). To deposit the material, the machine head
moves from the default position, P0, towards the end of platform, shown as position P1.
After reaching P1, the print head comes back to the default position P0. Deposition and
The total distance travelled by the machine head in the y-direction to complete a single
layer depends on the maximum breadth of a part and the breadth of the print head (Hy).
On the second pass, the machine head moves in the y-direction a distance equivalent to
breadth of the print head (Hy) then towards position P2. After reaching P2, the print head
comes back to the default position P0 in a rectilinear manner. The distance travelled on
the second pass is calculated by addition of the two times the breadth of the print head
(Hy) and the length of the platform (Lx). During this movement, deposition and curing are
The machine head moves sequentially to visit positions P1, P2, P3, and P4 from the
default position P0 to complete the deposition of a single layer. The number of passes is
where,
The maximum breadth of a part will be used for calculation as the user needs to enter
maximum part dimensions in the primary input parameters. The dimension of the print
head in the y-direction (breadth) will be automatically selected from the database
The total distance travelled by the machine head for depositing a single layer and
where,
𝑁𝐻 +1
∑𝑁 𝐻
𝑛=1 𝑛 = 𝑁𝐻 ∙ ( ) (44)
2
𝑁𝐻 +1
𝐿𝑘 = 2 ∙ 𝑁𝐻 [𝐻𝑦 ∙ ( ) + 2 ∙ (𝐿𝑐 + 𝐿𝑥 )] (45)
2
Using equation 45, the total length travelled by the print head can be calculated and
will be discussed. The results generated by AMCET will be validated by comparing with
Section 4.1 explains the architecture and features of AMCET. In section 4.2, the
validation of build time and cost are presented for Polyjet with actual manufactured parts.
Sections 4.3 and 4.4 discusses the validation of build time estimation for SLA and FDM
techniques respectively.
In section 4.5, the results of two different parts manufactured using Polyjet, SLA, and
FDM techniques are presented and compared with estimates generated by AMCET. The
actual build time and cost of each part are compared with estimates of AMCET.
AMCET consists of three sections: a primary user input screen, multiple secondary
Figure 11 shows the primary user input screen. At the top of the screen, the method of
manufacturing can be selected by the user. Then, the user can enter primary input
parameters based on the part design. These primary user input parameters are organized
The user can select the method of manufacturing out of the three AM techniques to
generate an estimate. After selecting the AM technique, the user can select the specific
AM machine and enter the primary machine input parameters. Further, the user can select
the specific material required from the AMCET database. Based on the part geometry the
user can enter the volume and the bounding box dimensions of the part into the material
parameters.
Labor rate and visit interval can be entered as primary labor input parameters as
shown in Figure 11. Further, based on the specific post-processing required on the part,
the user can enter the rate and quantity of the material. The “Configure” button provided
on each area of the primary user input parameters enables the user to view the secondary
Figure 12 shows the secondary parameters for the machine. Based on the selection of
primary user input parameters, the secondary machine parameters are configured.
As shown in Figure 12, the machine life, working days, machine platform
dimensions, repair and maintenance costs, and machine speeds are entered automatically
from the database depending on the specific machine type selected. To change the
configured parameters, the user can enter a different value for any parameter. As shown
in Figure 12, the 21 working days was configured in the model as the default. However,
the user has entered 20 working days which is now the input for calculation of the
machine cost.
Similarly, there are secondary material, labor, and post-processing parameters where
Figure 13 shows the screenshot of the output area where total cost is displayed along
with a chart of cost. The total cost and yielded cost of the part are displayed along with
the cost of machine, material, labor, post-processing material, and post-processing labor.
The output screen provides the cost break-down of total cost. AMCET enables the
user to generate the cost of identical parts (batch) together and the cost of a single part.
The usage (e.g., kilogram of material) and rates (e.g., $/kg) of each cost component are
displayed by AMCET which helps the user to understand the consumption and unit cost
of each cost component. The graphical representation of total cost break-down enables a
comparing with actual data of parts manufactured using the Polyjet technique. The parts
were manufactured at the Ohio University Innovation Center using Stratasys Objet 350
72
Connex 3D printer [52]. The build time, material cost, machine cost, and labor cost are
estimated using AMCET and compared with actual values of the manufactured parts.
technique.
Part 10 07:09
01:06 06:45
Part 9
00:52
Part 8 00:3500:31
03:18
Part 7 03:13
Part 6 01:10
Part 5 00:09 01:11
00:08 00:46
Part 4 00:56
Part 3 00:16
00:18
Part 2 07:19
Part 1 06:30 06:42
06:37
00:00 01:00 02:00 03:00 04:00 05:00 06:00 07:00 08:00
Build Time (H:M)
Estimates by AMCET Actual Build Time
By comparing the build times, it is noted that the estimates generated by AMCET have
mean absolute error of 20.9%. Some reasons for the errors include variations in machine
head speed, complex geometry of the part, and error in the estimate of the support
structure volume. For example, Part 4 is under-estimated because the volume support
structure considered for estimating a time is lower than actual support structure required
for manufacturing the part. Because the volume of the support structure is
Table 2 displays the actual and estimated build time of the ten parts manufactured at
the Ohio University Innovation Center using Stratasys Objet 350 Connex 3D printer and
The average error and standard deviation are considered because they help
understanding how well the estimates compare to the actual values. The average error
should be close to zero to ensure low average error. The standard deviation measures the
variations in the errors. When average error is near to 0%, a low standard deviation
confirms that errors are mostly close to 0%, whereas a high value of standard deviation
The average error generated by AMCET for build time estimation of Polyjet is 1.3%
with standard deviation of 13.2%, showing that AMCET can estimate the manufacturing
time of the parts using Polyjet technique with reasonable accuracy with acceptable
variation.
Figures 15 and 16 represent the cost components of the parts manufactured at the Ohio
University Innovation Center. The parts are divided into two groups (large parts and
$0 $50 $100 $150 $200 $250 $300 $350 $400 $450 $500
Material Cost Machine cost Labor Cost
Figure 15 compares the cost components for actual and estimated cost for the four
large parts. The estimated total cost is higher for all the four parts. The material, machine,
and labor cost of most of the parts are estimated higher than the actual cost.
75
Figure 16 compares the cost components for actual and estimated cost for the six small
parts. The estimated material cost is higher for four of the parts and lower for two of the
parts. The estimated machine cost is higher for all the parts, whereas labor cost is equal
Table 3 represents the values of cost components of the ten parts manufactured with
Machine Cost
Material Cost
Labor Cost
Labor Cost
Labor Cost
Error in
Error in
Error in
Machine
Machine
Material
Material
Cost
Cost
Cost
Cost
1 $83.00 $102.6 $40.00 $85.74 $110.0 $52.00 3.3% 7.1% 30.0%
Material costs estimated using AMCET have some variations as compared with the
actual cost. One reason is the difference in estimated and actual mass of support material.
AMCET considers the support factor and part volume to calculate the mass of support
material. However, the actual material (resin) consumption for support structure is
determined by the software used for Polyjet machine. Hence, constant value of support
factor can lead to underestimation of actual mass of support material. For all the ten parts
where mass of support material is overestimated has higher material estimated cost.
However, where mass of support material is underestimated has lower material cost.
Although material cost is estimated with an average error of less than 1%, the standard
Machine cost estimated using AMCET is higher than actual machine cost for most of
the parts with average error of 17.4% and standard deviation of 11.7%. The approach
used in determining the actual machine cost has considered the straight-line depreciation
which is the reason for over-estimation of the machine cost. If the interest rate in
AMCET is reduced to 0% from 10% then actual and estimated hourly machine operating
cost are nearly equal. Moreover, machine cost estimate uses various parameters such as
AMCET estimates the labor cost higher than actual for most of the cases. This is due
to the estimates are calculated by adding part loading time, setup time, operator time
spent during manufacturing, part removing time, and machine cleaning time. Whereas
only the machine time is used for calculating the actual labor cost. Hence additional labor
times will lead to higher estimation of labor cost. The average error for labor cost is
78
18.8% which shows reasonable accuracy and standard deviation of 20.7% indicate wider
range of estimates.
The build times of 12 parts manufactured using SLA were compared with the AMCET
estimate to validate the accuracy of SLA time estimation. The parts were downloaded
from the Thingiverse website [53] in the form of .STL files and detailed information is
provided in Appendix A. The SLA machine build time was estimated for these 12 parts using
PreForm Software developed by Formlabs [54]. Figure 17 represents the comparison of the
03:18
12 03:24
02:22
11 02:40
00:38
10 00:45
9 05:20
04:48
02:08
8 02:35
7 01:58
02:13
6 03:06
03:08
5 07:42
06:51
4 04:05
03:43
02:15
3
02:25
04:06
2 04:11
04:10
1
04:26
00:00 01:00 02:00 03:00 04:00 05:00 06:00 07:00 08:00 09:00
The build time estimated by AMCET for SLA parts has some variation as compared with
machine estimated time. The reason for this may be due to machine parameters (laser
scanning speed), complex part geometries, and error in estimation of the volume of
support structure. Also, due to the complex geometry of the part, the time required to
scan each layer could vary, although equation 40 assumes a constant time for each layer.
The estimates generated using AMCET take limited inputs from the user and are still able to
Table 4 represents the analysis of build time using SLA for Pre-Form and AMCET.
The time estimated by AMCET for manufacturing parts using SLA has an average error of
-3.4%, which indicates the accuracy of time estimation. Moreover, the standard deviation of
10.1%, demonstrates that estimated values are close to machine estimates. Hence AMCET
can be used to predict the build time of parts for SLA technique.
80
The build time data was collected for six parts manufactured separately at the Ohio
University ETM Department using FDM technology on Makerbot Replicator Z18 [55]
3D printer. The comparison of actual build time by machine and build time estimated by
01:35
6 01:54
Part Names
5 03:58
04:34
4 01:42
01:22
3 00:49
00:45
2 16:05
13:59
00:00 03:00 06:00 09:00 12:00 15:00 18:00
Build Time (H:M)
AMCET Estimate H:M Actual time H:M
The actual build time and estimated build time have a few variations. This may be due
to variation in the machine parameters such as a change in deposition rate and print head
speed during printing for complex part geometry, whereas equation 33 considers a
constant deposition rate and print head speed. Also, the difference in the volume of the
support structure can also introduce an error for estimation of the build time, because the
The actual build time and estimated build time of six parts using AMCET are
tabulated in Table 5.
81
Actual AMCET
Part
Estimate Estimate Error Analysis
Numbers
H:M H:M
1 9:34 9:05 -5.0%
2 13:59 16:05 15.0%
3 0:45 0:49 8.8%
4 1:22 1:42 24.3%
5 4:34 3:58 -13.1%
6 1:54 1:35 -16.6%
Average 2.2%
Standard Deviation 16.4%
Error analysis shows that the build time estimated with an average error of 2.2%
confirms the accuracy of time estimation of AMCET for FDM. The standard deviation of
16.4% indicates that the build time estimates are clustered around the mean.
The estimates generated using AMCET were evaluated by manufacturing two parts
using the three different techniques. The purpose of manufacturing two parts with three
different AM technique is to compare estimated and actual build time and total cost.
Figure 19(a) shows the CAD model of Part A and Figure 19(b) shows the CAD model
of Part B. Both parts were manufactured using Polyjet, SLA, and FDM.
These two parts were manufactured using Polyjet technique at the Ohio University
Innovation Center using Stratasys Objet 350 Connex 3D printer. For SLA, the parts were
manufactured using Formalab’s Form 2 printer and the FDM parts were manufactured
using the Makerbot Replicator Z18 printer at the Ohio University ETM Department.
Figure 20 shows the build time comparison for AMCET estimates and actually
06:00 Machine
Time
04:49
05:00
04:30 AMCET
Build Time (H:M)
00:00
Polyjet FDM SLA Polyjet FDM SLA
Part A Part B
The graph shows that the actual time required to manufacture Part A and Part B is
least for FDM, followed by Polyjet and SLA. AMCET estimates also show that FDM is
AMCET can be used to compare the build time estimates of various AM techniques
for the same part and select the technique which produces parts in the least amount of
time. This will enable the user to decide which technique can produce the maximum
output.
83
Figure 21 represents the total cost comparison of the two parts manufactured using
three AM techniques.
$120
$100.82 Actual Cost
$100
$76.52 AMCET
$80
Total Cost
Estimate
$60
$42.22
$40 $32.36 $37.35 $31.31
$27.14 $27.94 $22.89
$18.26$18.34
$20 $15.34
$-
Polyjet FDM SLA Polyjet FDM SLA
Part A Part B
The graph shows the actual total cost of Part A and Part B are highest for the Polyjet
technique followed by SLA and FDM. The estimates generated by AMCET also confirm
that the estimated total cost is highest for Polyjet followed by SLA and FDM for both the
parts.
AMCET also enables designers and engineers to compare the cost of three AM
techniques for various designs by entering the user input parameters. The user can
identify the least expensive AM technique and select the process for manfuacturing.
84
CHAPTER 5: CONCLUSION
components: machine, material, labor, and post-processing. This specific approach has
not been explored in the past research of AM cost estimation. AMCET was developed by
integrating important factors from the cost point of view which have not been considered
in the previous research for AM cost estimation. These factors include the yield of AM
process, the uncertainty in machine operation hours, the machine life, the support
AMCET has a similar approach of cost estimation for three AM technologies although
it has a process-specific approach of calculating the build time for each AM technique.
This approach helps in calculating the cost for different AM technologies by changing the
The estimates generated by AMCET have some errors but are accurate enough for
time and cost estimation purposes. The results section presents the comparison of actual
build time versus AMCET estimates for three AM technologies and cost comparison for
Polyjet. These results validate the correctness of the AMCET methodology and the
modeling approach. The user can make use of AMCET to estimate the time and cost of
various parts at the early stages of designs. Moreover, AMCET helps selecting the
AMCET can help designers to understand the changes in the cost of a product in the
preliminary design stage for AM technologies with little information. AMCET can help
85
AMCET also supports researchers and engineers with a quick cost estimation by
requiring limited primary user input. However, they can provide additional information
into the secondary user input parameters to increase the accuracy of the estimate.
Companies planning to invest in additive manufacturing can make the use of AMCET
to determine manufacturing cost and decide the suitable process. AMCET can help
boundary box dimension. Unlike other additive manufacturing cost estimation software
where loading of an .STL file is mandatory to generate an estimate, AMCET can generate
an estimate by entering the geometric parameters. This will allow engineering companies
to protect their design by not having to share it outside an organization until the design is
complete.
available. AMCET can help manufacturing companies to reduce cost by providing the
potential in future.
techniques. It can be extended for additional techniques such as binder jetting, direct
energy deposition, powder bed fusion, and sheet lamination [3]. The process-specific
part. The specific data on post-processing in terms of material consumed and labor hours
Further, the accuracy of time and cost estimation can be improved if the required
amount of support structure is known accurately. The selection of the support factor can
be improved based on the part design and specific AM technique to generate more
precise estimates.
AMCET is configured with a limited database of the machines and the materials of
options of machines and materials, enabling the user to generate an estimate with
AMCET can be developed in Visual Basic or Java based program to increase the
flexibility of the user interface. This can also enable the user to extract the dimensional
inputs directly from a CAD model instead of entering them manually. The advanced
program-based tool could generate multiple result files which can be used for
comparison.
87
REFERENCES
[1] Gibson, I., Rosen, D. W., and Stucker, B., “Introduction and Basic Principles,” in
Additive Manufacturing Technologies (Rapid Prototyping to Direct Digital
Manufacturing), Springer, 2010, what is additive manufacturing, page 1-5.
[2] Douglas S. Thomas and Stanley W. Gilbert “Costs and Cost Effectiveness of Additive
Manufacturing” - A Literature Review and Discussion.
[4] A.M.K. Esawi and M.F. Ashby, “Cost estimates to guide pre-selection of processes”,
Materials and Design Volume 24, Issue 8, Pages 605-616, June 2003.
[5] “Ronald Berger, Additive Manufacturing a game changer for the manufacturing
industry?”, Ronald Berger, Strategy Consultant, Munich, November 2013.
[6] Costabile, G., M. Fera, F. Fruggiero, A. Lambiase, and D. Pham. “Cost Models of
Additive Manufacturing: A Literature Review.” International Journal of Industrial
Engineering Computations, 2017, 263–83. DOI: 10.5267/j.ijiec.2016.9.001.
[9] R. Roy, S. Kelvesjo, S. Forsberg, and C. Rush, “Quantitative and qualitative cost
estimating for engineering design,” Journal of Engineering Design, vol. 12, no. 2, pp.
147–162, 2001.
[10] P. Duverlie and J. M. Castelain, Cost Estimation During Design Step: Parametric
Method versus Case Based Reasoning Method, vol. 15. 1999.
[15] M. Ruffo, C. Tuck, and R. Hague, “Empirical laser sintering time estimator for
Duraform PA,” International Journal of Production Research, vol. 44, no. 23, pp. 5131–
5146, Dec. 2006.
[16] L. Di Angelo and P. Di Stefano, “A neural network-based build time estimator for
layer manufactured objects,” The International Journal of Advanced Manufacturing
Technology, vol. 57, no. 1–4, pp. 215–224, Nov. 2011.
[20] J.-Y. Jung, “Manufacturing cost estimation for machined parts based on
manufacturing features,” Journal of Intelligent Manufacturing, vol. 13, no. 4, pp. 227–
238, 2002.
[21] D. Sormaz, P. Gannon, and S. Pulugurta, “Methodology for feature modelling and
cost estimation of large cast parts,” in IIE Annual Conference. Proceedings, 2013, p.
2349.
[22] Parametric Estimating Handbook, 4th ed., International Society of Parametric Analysts.
Vienna., VA. Virginia, 2008, 1-1-1-5.
[23] S. L. Chan, Y. Lu, and Y. Wang, “Data-driven cost estimation for additive
manufacturing in cybermanufacturing,” Journal of Manufacturing Systems, vol. 46, pp.
115–126, Jan. 2018.
[25] SEER for manufacturing: Parts, Process, and Assembly, [Online]. Available FTP:
http://galorath.com/wp-content/uploads/2014/08/SEERforManufacturing2.pdfSEER
[27] Chee Kai Chua, Kah Fai Leong, and Chu Sing Lim, “Solid-Based Rapid Prototyping
Systems”, in Rapid Prototyping: Principles and Applications, 3rd edition, World Scientific
Publishing Company, Singapore, 2010, page 143-145.
[29] Chee Kai Chua, Kah Fai Leong, and Chu Sing Lim, “Liquid-Based Rapid
Prototyping Systems”, in Rapid Prototyping: Principles and Applications, 3rd edition,
World Scientific Publishing Company, Singapore, 2010, page 46-47.
[31] Liu Q, Orme M, “High precision solder droplet printing technology and the state-of-
the-art,” Journal of material processing technology 115 (2001) 271-283.
[32] Chee Kai Chua, Kah Fai Leong, Chu Sing Lim, “Liquid-Based Rapid Prototyping
Systems”, in Rapid Prototyping: Principles and Applications, 3rd edition, World Scientific
Publishing Company, Singapore, 2010, page 58-59.
[33] P. Alexander, S. Allen, and D. Dutta, “Part orientation and build cost determination
in layered manufacturing,” Computer-Aided Design, vol. 30, no. 5, pp. 343–356, 1998.
[35] S. Yim and D. Rosen, “Build time and cost models for additive manufacturing
process selection,” Volume, vol. 2, p. 32nd, 2012.
[38] C. Lindemann, U. Jahnke, M. Moi, and R. Koch, “Analyzing product lifecycle costs
for a better understanding of cost drivers in additive manufacturing,” in 23th Annual
International Solid Freeform Fabrication Symposium–An Additive Manufacturing
Conference. Austin Texas USA 6th-8th August 2012.
[43] M Ruffo, C Tuck, and R Hague “Cost estimation for rapid manufacturing – laser
sintering production for low to medium volumes” May 2006.
[44] Y.Zhang and A. Bernard, “Generic build time estimation model for parts produced
by SLS” in High Value Manufacturing, Nantes, France, 2014, ISBN 978-1-138-00137-4.
[45] L. Di Angelo and P. Di Stefano, “A neural network-based build time estimator for
layer manufactured objects,” The International Journal of Advanced Manufacturing
Technology, vol. 57, no. 1–4, pp. 215–224, Nov. 2011.
[46] H. Demuth and M. Beale, “Neural network toolbox,” For Use with MATLAB. The
MathWorks Inc, vol. 2000, 1992.
[51] D. V. Becker and P. Sandborn, “Using yielded cost as a metric for modeling
manufacturing processes,” in Proceedings of the ASME 2001 Design Engineering
Technical Conferences and Computers and Information in Engineering Conference,
2001.
The list of 12 part along with length (x), breath (y), height (z) and part volume is
listed in Table 6.
Part volume
Part Name x Inch y Inch z Inch
Number Inch3
!
!
Thesis and Dissertation Services