Available

Available online
online at
at www.sciencedirect.com
www.sciencedirect.com

ScienceDirect
ScienceDirect
Available online atonline
Available www.sciencedirect.com
at www.sciencedirect.com
Procedia
Procedia CIRP
CIRP 00
00 (2016)
(2016) 000–000
000–000
ScienceDirect
ScienceDirect www.elsevier.com/locate/procedia
www.elsevier.com/locate/procedia

Procedia CIRP 00 (2017)

Procedia 000–000
CIRP 75 (2018) 108–113
www.elsevier.com/locate/procedia

15th
15th CIRP
CIRP Conference
Conference on
on Computer
Computer Aided
Aided Tolerancing
Tolerancing –– CIRP
CIRP CAT
CAT 2018
2018

In-process
In-process measurement
28th CIRP of
measurement the
Design
of surface
surface quality
theConference, May 2018,for
quality aa novel
Nantes,
for Francefinishing
novel finishing process
process
for polymer additive manufacturing
A new methodologyfor to polymer additive
analyze the manufacturing
functional and physical architecture of
existing P.
Wahyudin
Wahyudin products
P. Syama
Syama*, for anLeach
*, Richard
Richard assembly
Leach
aa
oriented
,, Konstantin
Konstantin product
Rybalcenko
Rybalcenko
bb
family
,, Andrȇ
Andrȇ Gaiobidentification
Gaio b
,, Joseph
Joseph Crabtree
Crabtreeb
b

Paul StiefMetrology
aManufacturing
a *, Jean-Yves Dantan, Alain Etienne, Ali Siadat
Team, Faculty of Engineering, University of Nottingham, NG8 1BB, UK
Manufacturing
Metrology Team, Faculty of Engineering, University of Nottingham, NG8 1BB, UK
bAdditive
Manufacturing Technologies, Alison
Alison Business Centre, Sheffield, S2
S2 1AS, UK
b
École Nationale SupérieureAdditive
d’Arts Manufacturing
et Métiers, ArtsTechnologies,
et Métiers ParisTech,Business
LCFC EA Centre,
4495,Sheffield, 1AS,Fresnel,
4 Rue Augustin UK Metz 57078, France
** Corresponding author. Tel.: +44-7492316036; fax: +44-1157484818. E-mail address: wahyudin.syam@nottingham.ac.uk
Corresponding author. Tel.: +44-7492316036; fax: +44-1157484818. E-mail address: wahyudin.syam@nottingham.ac.uk
* Corresponding author. Tel.: +33 3 87 37 54 30; E-mail address: paul.stief@ensam.eu

Abstract
Abstract
Abstract
In
In the
the last
last decade,
decade, there
there has
has been
been considerable
considerable growth
growth in in the
the production
production of of end-use
end-use polymer
polymer parts
parts and
and components
components using
using additive
additive manufacturing
manufacturing
methods.
Inmethods.
today’s AA wide
business range of polymers,
environment,
wide range of polymers, from
the trend Nylon-12
towards more
from Nylon-12 to thermoplastic
product variety
to thermoplastic polyurethane polymers,
and customization
polyurethane can
polymers,iscan be processed
unbroken. with
Due towith
be processed complex
thiscomplex geometry
development, tailored
the need
geometry of
tailored
to
to specific
agilespecific function.
function. However,
and reconfigurable However, due
due to
production the
the nature
tosystems
nature of
of the
emerged layer-by-layer
the to
layer-by-layer
cope with variousprocess used
used in
processproductsin additive
additive manufacturing,
manufacturing,
and product high
high
families. To roughness
roughness
design surfaces remain
remain on
surfacesproduction
and optimize on
the
the parts.
systems asTo
parts. To reduce
as tothe
reduce
well the roughness
roughness
choose of
of the
the optimalthe surfaces,
surfaces, aa proprietary
proprietary
product matches, post-processing
post-processing
product method,
method,
analysis methods aredeveloped
developed by
by Additive
mostManufacturing
Additive
needed. Indeed, of the known Technologies,
Manufacturing Technologies,
methods aim to is
is
applied
analyze to the
appliedatoproductsurfaces.
the surfaces. To
or oneTo monitor
monitor
product and control
and on
family control the finishing
the finishing
the physical of the surfaces,
level.ofDifferent
the surfaces, an in-process
an in-process
product surface detection
surface detection
families, however, may differinstrument
instrument has
largely inhas been
been
terms developed
ofdeveloped based
the numberbased
and
on
on machine
naturemachine vision
vision and
of components. and machine
This factlearning.
machine impedesThis
learning. This paper
paper presents
an efficient presents the
the machine
comparison machine learning
of approach
learning
and choice approach
appropriateand the
the effectiveness
andproduct
effectiveness of
of the
the instrument
instrument
family combinations for in-process
forproduction
for the in-process
measurement
measurement of
of the
the finished
finished surfaces.
surfaces.
system. A new methodology is proposed to analyze existing products in view of their functional and physical architecture. The aim is to cluster
©
© 2018 The Authors. Published by
by Elsevier B.V.
© 2018
these The
products
2018 Authors.
The Authors.
Peer-review under
Publishedoriented
in new assembly
Published
responsibility by
of
Elsevier
Elsevier
the
B.V.
product
B.V.families
Scientific Committee
for the optimization of existing assembly lines and the creation of future reconfigurable
of the 15th
Peer-review
assembly under
systems. responsibility
Based on Datum of the
Flow Scientific
Chain, Committee
the physical
Peer-review under responsibility of the Scientific Committee of the of the
15th15th
structure ofCIRP
CIRPCIRP
the Conference
Conference
products
Conference is on
on Computer
Computer
onanalyzed.
Computer
Aidedsubassemblies
AidedAided
Functional Tolerancing
Tolerancing
Tolerancing
- are
CIRP
- CIRP-CAT
CIRP CAT
CAT 2018.
2018.
identified,
2018. and
a functional analysis is performed. Moreover, a hybrid functional and physical architecture graph (HyFPAG) is the output which depicts the
Keywords: In-process
Keywords:between measurement,
In-processproduct
measurement, additive
additive manufacturing,
manufacturing, machine
machine learning,
learning, machine
machine vision
vision system planners and product designers. An illustrative
similarity families by providing design support to both, production
example of a nail-clipper is used to explain the proposed methodology. An industrial case study on two product families of steering columns of
thyssenkrupp Presta France is then carried out to give a first industrial evaluation of the proposed approach.
©1. Introduction
1.2017
Introduction
The Authors. Published by Elsevier B.V.
Peer-review under responsibility of the scientific committee of the 28th CIRP Design Conference 2018.
Additive
Additive manufacturing
manufacturing (AM)
(AM) has
has been
been extensively
extensively used
used for
for Manufacturing
Manufacturing Technologies
Technologies (AMT). (AMT). AMT AMT offers
offers aa proprietary
proprietary
the
the production
Keywords: of
of end-use
Assembly;
production polymer
Design method;
end-use polymer parts
Family during
during thethe last
identification
parts last decade.
decade. post-processing
post-processing machine for polymer AM parts
machine for polymer AM parts that
that can
can
With AM, a wide range of polymers, including
With AM, a wide range of polymers, including Nylon-12 and Nylon-12 and automate
automate the process to improve the quality of the surface
the process to improve the quality of the surface
thermoplastic
thermoplastic polyurethane
polyurethane polymers,
polymers, can can be be shaped
shaped into into texture.
texture. The
The automated
automated processprocess results
results inin aa significant
significant increase
increase
complex
complex geometries
geometries designed
designed for
for specific
specific functions.
functions. in
in productivity
productivity due
due to
to a
a substantial
substantial reduction
reduction of
of the
the post-
post-
1. Introduction
However, due to the layer-by-layer nature of AM, the so- of the product
processing timerange
and and
an characteristicsofmanufactured
improvement the surface and/or
texture
However, due to the layer-by-layer nature of AM, the so- processing
assembled time and an improvement of the surface
in this system. In this context, the main challenge in texture
called
called “stair-case
“stair-case effect”
effect” on
on thethe surface
surface of aa produced
ofthe produced part
part quality.
quality. and analysis is now not only to cope with single
Due
occurs to the fast development in domain of modelling
occurs [1],
[1], causing
communication causing the part
and the an part
to
to have
ongoing
aa rough
havetrend rough
of of
surface.
surface. Various
digitization Various
and
An
An integrated
products, integrated
a limited
measurement
measurement
product range
system
system
or existing
for use
use with
forproduct with the
the
families,
surface
surface inclinations
inclinations due
due to
to the
the orientation
orientation of a
a part
part during
during proprietary
proprietary machine
machine is
is required
required to
to detect
detect surface
surface conditions
conditions and
and
digitalization,
processing manufacturing enterprisesstructures
are facing important but also to be able to analyze and to compare products to define
processing and and the
the excess
excess of of support
support structures increaseincrease the the to
to send
send feedback
feedback to to the
the machine
machine for for closed-loop
closed-loop control
control [3].
[3].
challenges
effect [2]. in today’s market environments: a continuing new
With product families. It can be observed that classical existing
effect [2]. With integrated measurement, a closed-loop feedback control
integrated measurement, a closed-loop feedback control
tendency
To towards texture
reduction AM of product development times and product families are regrouped in function of clients or features.
To reduce
reduce thethe texture of of AM polymer
polymer parts,
parts, post-processing
post-processing and
and aa significant
significant quality
quality improvement
improvement of of processed
processed parts
parts can
can
shortened
procedures product
can lifecycles.
be applied. In addition,
Manual there
and is an
laboriousincreasing
post- However,
be assembly oriented product families are hardlyhas to find.
procedures can be applied. Manual and laborious post- be obtained [4]. In this paper, an in-process instrument has been
obtained [4]. In this paper, an in-process instrument been
demand
processingof customization, being at theused. same time in a global On the product family level, products differ mainly in two
processing procedures
procedures are are commonly
commonly used. These These procedures
procedures developed
developed to to monitor
monitor and and control
control surfaces
surfaces processed
processed by by the
the
competition
are very timewith competitors
consuming and all over
reduce the
the world.
productivity This oftrend,
AM main characteristics:
post-processing (i)
machine. the number
The paper of components
describes a new and (ii) the
in-process
are very time consuming and reduce the productivity of AM post-processing machine. The paper describes a new in-process
which is inducing the development from macro to micro
processes. type of components
surface (e.g. mechanical, electrical, electronical).
processes. surface detection
detection solution
solution for for polymer
polymer AM AM parts.
parts.
markets,
A results in diminished lot sizes due to augmenting Classical methodologies considering mainly single products
A newnew automated
automated solution
solution for for the
the post-processing
post-processing of of This
This paper
paper isis structured
structured as as follows.
follows. Section
Section 2 2 describes
describes the the
product
polymer varieties (high-volume to low-volume production) [1]. or solitary, already existing product families analyze the
polymer AM AM partsparts has has beenbeen developed
developed by by Additive
Additive proprietary
proprietary post-processing
post-processing machinemachine including
including an an example
example of of aa
To cope with this augmenting variety as well as to be able to product structure on a physical level (components level) which
identify
2212-8271 possible optimization potentials in the existing causes difficulties regarding an efficient definition and
2212-8271 © © 2018
2018 The
The Authors.
Authors. Published
Published by by Elsevier
Elsevier B.V.
B.V.
production
Peer-review system,
under it is important
responsibility of the to have
Scientific a precise
Committee of knowledge
the 15th CIRP comparison
Conference
Peer-review under responsibility of the Scientific Committee of the 15th CIRP Conference on of different
on Computer
Computer Aided product
Aided Tolerancing
Tolerancing families.
-- CIRP
CIRP CAT
CAT 2018. Addressing this
2018.

2212-8271©©2017
2212-8271 2018The
The Authors.
Authors. Published
Published by Elsevier
by Elsevier B.V. B.V.
Peer-review
Peer-review under
under responsibility
responsibility of scientific
of the the Scientific Committee
committee of the
of the 28th 15th
CIRP CIRPConference
Design Conference on Computer Aided Tolerancing - CIRP CAT 2018.
2018.
10.1016/j.procir.2018.04.088
 Wahyudin P. Syam et al. / Procedia CIRP 75 (2018) 108–113
2 Wahyudin P. Syam et al./ Procedia CIRP 00 (2018) 000–000
surface condition of a polymer AM part before and after post-
processing. Section 3 describes the in-process surface detection
requirements, hardware and software developments,
instrument testing and sensitivity analysis of the detection
process. Finally, section 4 concludes the paper and describes
future work.
2. PostPro3D: a novel finishing process for polymer
additive manufacturing
Surface smoothing technology Postpro3D (see Fig. 1) is a
physical-chemical process partly based on the chemical
PUShTM method, developed at the University of Sheffield [5,6].
The technology can smooth a wide variety of polymers used in
AM, including Nylon-12, Nylon-11, Nylon-6, thermoplastic
polyurethane (TPU), thermoplastic elastomers, ULTEM 9085,
PMMA or the like. Postpro3D is a non-line-of-sight process
meaning it can smooth complex-to-reach internal cavities of
polymer components.
The PostPro3D smoothing process is highly controllable.
This capability allows for reproducible results and a desired
level of surface finish to be achieved, which effectively
becomes comparable to that of polymer components
manufactured using injection moulding (see Fig. 2). During the
process, the surface pores of the component are closed, which
in turn provides water tightness properties to the AM polymer
material.
Fig. 1. PostPro3D automatic smoothing machine.
Fig. 2. Surface of a TPU material smoothed to different controllable levels.
3. Development of in-process measurement for the
finishing process
We define “in-process measurement” as the use of an
instrument that measures an aspect of product quality, for
example, surface texture and/or geometrical characteristics,
109
during, before or after a manufacturing process. The instrument
is placed within the process cycle, either inside or outside the
process chamber. The results of the measurement can be used
to inspect the product or control the process.
3.1. In-process measurement requirement
An in-process measurement for the surface condition
detection has several requirements:
 The size of the instrument should be < (200 × 200 ×
200) mm to fit the end-effector of a collaborative
robot.
 The mass of the instrument should be < 2 kg to meet
the payload of a small collaborative articulated robot
arm.
 Stand-alone robust and fast software.
 Detection of surface condition within < 15 s.
 The cost of the system should be < 5 % of the cost of
the post-processing machine.
 The instrument should be flexible and portable.
 The instrument should be simple to be integrated into
the post-processing machine.
Two possible solutions for the development of the in-
process surface detection instrument are 3D surface
measurement and 2D machine vision methods. 3D surface
measurement is deemed to be not currently suitable for several
reasons:
 3D surface measurements still require longer
measurement times (typically in the order of one
minute) compared to the required detection time
(< 15 s).
 The cost of the system will be high due to the need for
a precision optical system and many systems require
a linear motion stage to scan through a focus position
so that a sequence of images can be captured.
For the above reasons, a solution based on 2D machine
vision was selected. Reasons of the selection are as follow:
 Quantitative surface analysis is not required. The
measurement involves image comparison of a
measured and a reference surface. The reference
surface is a surface considered as a pre-defined
surface with smooth surface finish.
 A low-cost instrument can be obtained with a 2D
machine vision method since the cost of imaging
sensors has been significantly reduced.
 The 2D machine vision measurement capability can
be optimised by utilising a machine learning method
to improve the detection capability of various surface
textures.
3.2. Instrument development
The instrument design is shown in Fig. 3. The design
complies with the requirements for low-mass, flexibility and
portability. With the instrument, a small area of a surface can
be captured and magnified for further analysis. In Fig. 1 (top)
the dimensions of the instrument are (203 × 121 ×
84) mm.
110 Wahyudin P. Syam et al. / Procedia CIRP 75 (2018) 108–113
Author name / Procedia CIRP 00 (2018) 000–000
Fig. 3. The 3D solid model of the in-process surface detection instrument.
The instrument is made from illumination and microscope
modules. The illumination module consists of (see Fig. 3
bottom) a white light light emitting diode (LED) and a diffuser
lens. The LED has a total power output of 250 mW with an
intensity of 3 mW/cm2 . The LED’s emission has a spectrum
of 400 nm to 700 nm . To improve the cross-sectional
intensity distribution of the light from the LED, the diffuser
lens, with a transmission spectrum of 380 nm to 1100 nm, is
used. With the diffuser lens, the LED will have a uniform
intensity across the field of view of the microscope’s objective
lens [7].
The microscope module consists of (see Fig.3 bottom) a
camera with a complementary metal-oxide semiconductor
(CMOS) sensor, a beam splitter, a tube lens, and objective
lenses with 4× and 10× magnifications. The tube lens is used to
transform a parallel ray reflected from the measured surface
through the objective lens into an image on the CMOS sensor.
Both the beam splitter and the tube lens have transmission
spectra of 400 nm to 700 nm. The CMOS sensor has a pixel
density of (1280 × 1024) pixels with a frame rate of 30 fps.
3.3. Software development
There have been many studies about surface texture analysis
using 2D machine vision methods. Examples are the use of
machine vision methods to estimate surface texture parameters
from turning [8], milling [9] and grinding [10] processes. The
studies commonly capture and process an image, calculate
some image parameters specific to the features of interest from
3
a process and statistically correlate the calculated image
parameters to certain texture parameters.
Fig. 4. The developed software. (a) The measurement module and (b) the
machine learning module.
Most of the existing studies only focus on a specific type of
surface and are based on regression analysis. With regression
analysis, the correlation of a specific image condition to
specific texture parameters is only limited to a specific surface
from a specific process.
In this paper, a general classifier of various different surface
types from different AM processes and different polymer
materials is developed. The classifier is based on an
unsupervised machine learning approach using principal
component analysis (PCA) [11]. The goal of PCA is that high
dimension data, in this case the (1280 × 1024) dimensions of
the image size obtained from the CMOS sensor, can be reduced
to a lower number of dimensions that still contains the
important surface texture information.
To improve the computation efficiency of the PCA, a total
of 54 image parameters are calculated from the captured image
of a surface, obtained from the developed instrument. By
calculating these parameters, a first step in data reduction is
applied to speed up the classification process using PCA. The
54 image parameters are divided into two categories: colour-
related and texture-related parameters. The parameters are:
 A total of 17 colour-related parameters.
 A total of 37 texture-related parameters.
The colour-related parameters are obtained from the
calculation of statistical parameters of the colour (RGB and
HSV) of a 3-channel image, and the histogram entropy of the
3-channel image [11]. Meanwhile, the texture-related
parameters are obtained from the calculation of statistical
 Wahyudin P. Syam et al. / Procedia CIRP 75 (2018) 108–113 111
4 Wahyudin P. Syam et al./ Procedia CIRP 00 (2018) 000–000

parameters of blobs of an image, local edge descriptors and

binary local patterns [13].
Local edge descriptors [12] and image colour parameters are
part of the multimedia content description interface (MPEG-7)
[13]. Binary local pattern parameters are image parameters that
are calculated from analysis of many small sub-regions in an
image [14,15].
The PCA classification of the images of different surface
conditions are calculated from the 54 image parameters. With
this strategy, the calculation of the PCA classification is more
efficient compared to the calculation of the PCA from all the
images’ pixels.
The software is implemented in the C/C++ programming
language as stand-alone software that controls the instrument
and process images to detect surface conditions. The image
processing uses the OpenCV robust image processing library
[16] and the graphical user interface (GUI) uses the Qt4 library
[17]. Fig. 4 shows the developed software with the two main Fig. 5. One example for each type of simulated images with speckle patterns.
modules: measurement and machine learning. The software is
able to control a collaborative robot in order to move the 3.4. Instrument and software testing
microscope and find the focus position with respect to a
surface. Instrument and software testing was carried out to study
The measurement module provides functionality to control their effectiveness for surface detection. Two stages of testing
the collaborative robot and to detect a surface condition and were implemented: testing with simulated images and testing
compare to a reference surface. The detection process is carried with real TPU surface images.
out based on machine learning approach that learns distinctive
image properties data from a reference surface. Based on the
learning process, a surface can be detected and classified as
similar or dissimilar to the reference surface.
The machine learning module provides functionality to
control the collaborative robot, to adjust camera settings and to
train the software with a specific reference surface. The camera
settings can be adjusted to find an optimal surface colour. An
auto-exposure algorithm [18] and a white-balancing algorithm
[19] are implemented to optimise the colour adjustment. This
module allows the setting of the number of training data and
the number of reduced dimensions from 2 to 54.
The machine learning process is as follows. An image is
taken from the CMOS sensor according to a number of training
images 𝑛𝑛 that are set by the user. The 54 image parameters are Fig. 6. Separation of each type of simulated images in PC space (2D view).
calculated. By extracting the parameters, the first data
reduction is applied to increase training efficiency so that only The testing with simulated images uses generated images
hundred number of images are required to effectively conduct with simulated features. The simulated features are speckles
the machine learning process. A mean of the total 54 with different size and density to represent different features on
parameters is calculated and a difference matrix between the 54 a surface. Four types of simulated images with speckle features
parameters of each image and the mean parameters is derived. are generated, namely Type 1, Type 2, Type 3 and Type 4 (see
A 54 × 𝑛𝑛 training matrix is constructed. Finally, the PCA Fig. 5). The method to generate the speckle images can be
method is applied to the training matrix. The eigenvectors and found elsewhere [20]. A total of 100 images are generated for
eigenvalues of the trained data are stored in a file. The file can each type. Type 1 images, representing an un-processed
be recalled when a specific surface detection is to be carried surface, have the largest size of speckle patterns with the lowest
out. density. In contrast, Type 4 images, representing a processed
A similarity value is calculated between the reference surface, have the smallest size of speckle patterns with the
surface and the measured surface to decide whether the two highest density.
surfaces are similar or not. The value is calculated as the A Type 4 simulated image is selected as a reference surface.
Euclidean distance between the reference surface and the A total of 100 images are used for training. The trained data are
measured surface in their principle component (PC) space. used to detect a different type of simulated image with respect
With the calculation of the similarity value, subjectivity for to the reference image. The PCA uses three dimensions of PC
determining a specific surface texture condition can be space for the surface detection.
eliminated.
112 Wahyudin P. Syam et al. / Procedia CIRP 75 (2018) 108–113
Author name / Procedia CIRP 00 (2018) 000–000 5

Fig. 6 shows the separation plot, considering only two PCs,

from each image type in PC space. From Fig. 6, it can be
observed that the different types of surface can be classified
into different groups. The Type 4 surfaces can be largely
separated from the other types. Calculated similarity values
will be significantly smaller for Type 4 compared to other
values for other types. Table 1 shows the calculated similarity
values for the four types of surfaces compared to the reference
surface (Type 4). From table 1, by setting a threshold, Type 4
surfaces can be detected.

Fig. 7. The measurement of the TPU surfaces.

The testing with the TPU surface images uses five types of
surfaces, namely: Type 1, Type 2, Type 3, Type 4 and Type 5
that represent 0 % (unprocessed), 25 %, 50 %, 75 % and 100 %
(fully processed) post-processed surfaces, respectively. A total
of 100 images for each type of surface are captured. The 100
images are captured from different areas that cover the entire Fig. 8. The images of the five types of TPU surfaces.
TPU surfaces so that the images cover various types of features
on each surface type. Fig. 7 shows a measurement process for
one of the TPU surfaces.
Fig. 8 shows one of the images of the measured surface for
each type of surface. From Fig. 8, the Type 1 (unprocessed)
surface has high roughness and Type 5 (fully processed) has
low roughness. The reference surface is a surface from Type 5.

Table 1. Similarity values, with respect to a reference surface, for the tested
simulated and TPU surfaces. σ is a standard error.
Image Similarity value (mean ± σ) ×103
type Type 1 Type 2 Type 3 Type 4 Type 5 Ref.
Simula- 23797. 20117. 5219.6 8.7 ± Type
-
ted 6 ± 327 5 ± 234 ± 65 0.09 4
Image Similarity value (mean ± σ) ×106
type Type 1 Type 2 Type 3 Type 4 Type 5 Ref.
40.1 ± 176.2 ± 348.9 ± 48.2 ± 9.8 ± Type
TPU
0.3 1.9 3.4 0.8 0.3 5 Fig. 9. Separation of each type of TPU images in PC space (2D view).

A total of 100 Type 5 TPU images are used for training with
PCA. Three PCs from the training data are used to calculate the
similarity of the measured surfaces with the reference surface.
Fig. 9 shows the separation plot, considering only two PCs, of
each TPU image type in PC space. From Fig. 9, the Type 5 TPU
surfaces can be separated from the other types of TPU surface.
The type 4 surfaces are grouped close to the Type 5 as can be
qualitatively observed from the image (Fig. 6) that the Type 4
surface is similar to the Type 5 surface.
Table 1 shows the calculated similarity values for the five
types of TPU surfaces compared to the Type 5 surface as the
reference. From table 1, by setting a threshold, Type 5 surface
can be detected. The detection time was around 2 s.
3.5. Sensitivity analysis
The pixel detector on the CMOS sensor has noise so that the
intensity value of a pixel for each detector will vary over time.
Sensitivity analyses were carried out to investigate the degree
of the variation of the pixel intensity on the detector over time
and to quantitatively analyse the effect of the pixel intensity
variation to the similarity value.
The analysis of the intensity variation over time is carried
out by analysing a single intensity value of a pixel on the
detector. A Nylon-12 surface was used for the analysis. The
sampling frequency of the detector was set to 15 fps. A total of
100 pixels were sampled over a period of 6.6 s. The sampling
period is considered sufficient, since it is larger than 2 s for a
 Wahyudin P. Syam et al. / Procedia CIRP 75 (2018) 108–113
6 Wahyudin P. Syam et al./ Procedia CIRP 00 (2018) 000–000
surface detection time. Fig. 10 shows the pixel intensity
variation over 6.6 s. The results of the variation analysis shows
that the standard deviation of the pixel intensity is 2 pixel unit.
Fig. 10. The pixel variation over a period of 6.6 s (100 values).
Fig. 11. The sensitivity of similarity value over the level of pixel variation.
The analysis of the similarity value is carried out by
analysing the similarity value of a Nylon-12 surface with
respect to increasing value of the pixel variation of the image
of the Nylon-12 surface. The intensity value of the pixels of the
image are perturbed by a Gaussian noise with a standard
deviation ranging from 0 to 100 pixel unit.
Fig. 11 shows the results of the sensitivity analysis of
similarity value. From Fig. 11, it can be observed that the
similarity value is still stable until the pixel noise is more than
30 pixel values. From this result, a surface detection is
considered robust, since the pixel intensity variation is only
within 2 pixel unit value.
4. Conclusion
In this paper, the development of an in-process surface
detection instrument for an automated post-processing machine
for polymer AM parts has been presented. The instrument can
significantly benefit the machine by providing closed-loop
process control and removing subjectivity of surface quality
verification. The instrument is equipped with software to
control the instrument and classify a measured surface based
113
on machine vision and machine learning. Future work will fully
integrate the instrument into the post-processing machine.
References
[1] Moroni G, Syam WP, Petrò S. Toward early estimation of part accuracy in
additive manufacturing. Procedia CIRP 2014;21:300-305.
[2] Moroni G, Syam WP, Petrò S. Functionality-based part orientation for
additive manufacturing. Procedia CIRP 2015;36:217-222.
[3] Schmitt R, Moenning F. Ensure success with inline-metrology. XVIII
Imeko World Congress, September 17-22, 2006, Rio De Jeneiro, Brazil.
[4] Kunzmann H, Pfeifer T, Schmitt R, Schwenke H, Weckenmann A.
Productive metrology – adding value to manufacture. Ann. CIRP
2005;54:155-168.
[5] Ellis A, Brown R, Hopkinson N. The effect of build orientation and surface
modification on mechanical properties of high speed sintered parts Surf.
Topogr.: Metrol. Prop. 2015;3:034005.
[6] Crane NB, Ni Q, Ellis A, Hopkinson N. Impact of chemical finishing on
laser-sintered nylon 12 materials. Add. Manuf. 2017;13:149-155.
[7] Pedrotti FL, Pedrotti LS. Introduction to optics: 2nd edition. USA: Prentice
Hall International; 1987.
[8] Lee BY, Tarng YS. Surface roughness inspection by computer vision in
turning operations. Int. J. Mach. Tool. Manuf. 2001;9:1251-1263.
[9] Kumar R, Kulashekar P, Dhanasekar B, Ramamoorthy B. Application of
digital image magnification for surface roughness evaluation using
machine vision. Int. J. Mach. Tool. Manuf. 2005;45:228-234.
[10] Huaian YI, Jian L, Enhui LU, Peng AO. Measuring grinding surface
roughness based on the sharpness evaluation of colour images. Meas. Sci.
Technol. 2016;27:025404.
[11] Bishop CM. Pattern recognition and machine learning. Singapore:
Springer; 2006.
[12] Manjunath BS, Ohm JR, Vasudevan V, Yamada A. Color and texture
descriptors. IEEE Trans. Circ. Systm. Vid. Technol. 2001;11:703-715.
[13] Park DK, Jeon YS, Won CS. Efficient use of local edge histogram
descriptor. In proc. ACM Workshop on Multimedia. 2000; 51-54.
[14] Pietikainen M, Ojala T, Xu Z. Rotation-invariant texture classification
using feature distributions. Pattern recognition. 2000;33:43-52.
[15] Ojala T, Pietikainen M, Maenpaa T. Multiresolution gray-scale and
rotation invariant texture classification with local binary patterns. IEEE
Trans. Patter. Anlys. Intelli. 2002;24:971-987.
[16] Bradski G, kaehler A. Learning OpenCV. USA: O’reilly; 2008.
[17] Blanchette J, Summerfield M. C++ GUI programming with Qt 4. USA:
Prentice Hall; 2006.
[18] Liang J, Qin Y, Hong Z. An auto-exposure algorithm for detecting high
contrast lighting conditions. In ASICON 2007, ASIC, 2007: 725-728.
[19] Huo JY, Chang YL, Wang J, Wei XX. Robust automatic white balance
algorithm using gray color points in image. IEEE Trans. Consum. Electron.
2006;52:541-546.
[20] Duncan DD, Kirkpatrick SJ, Wang RK. Statistics of local speckle
contrast. J. Opt. Soc. Am. A 2008;25:9-15.