1st international conference on new research and...
TRANSCRIPT
1
1st International Conference on New Research and Development
in Technical and Natural Science,
ICNRDTNS
Radenci, Slovenia, 18.- 20. September 2019
2
The organizers of the 1st International Conference on New Research and Development in Technical and Natural Science, ICNRDTNS
are expecting you, our dear participants and guests at another traditional gathering of all of those who are interested in the use of
Mathematics, Physics, Chemistry, Biology, Medicine, Computer Science, Electrical Engineering, Mechanical Engineering, Civil and
Geotechnical Engineering and all related high technologies in the felds of Engineering and Science.
During the whole week from 18th until 20th of September 2019 in Radenci, Slovenia, the 1st convention ICNRDTNS offers you a
number of interesting events. The first bioclimate health resort in Slovenia, featuring springs of mineral and thermal Radenska water,
invites relaxation that will boost your health and well-being. Numerous regular guests are loyal to this health resort with a big heart.
The conference venue is the hotel Radin. Radenci Spa is located in the northeastern Slovenia, 5 kilometers from Gornja Radgona at
the Austrian border and 12 kilometers from Murska Subota. Legends say that the path for the mineral water in the Radenci Spa is
paved by the diligent elves. Karl Henn was listening to the underground ripples in 1833, when he visited Radenci for the first time.
After detailed water analysis, he came back to Radenci as an acknowledged doctor and filled a first bottle with Radenska mineral
water in 1869. It was later delivered to the imperial court in Vienna and to the pope in Rome. First guests visited Radenci almost 130
years ago, or in the year of 1882 to be more exact. Radenci Spa is known worldwide for its mineralized water. Mineral water
Radenska is sodium-calcium-hydrogen-carbonate mineral water and its CO2 concentration make it one of the most abundant mineral
waters in Europe. It has multiple therapeutic and beneficial effects on our body: stabilizes blood pressure, precipitates digestion,
neutralizes excessive gastric acid, lowers the uric acid values, increases urine excretion and strengthens the body and well being in
general. Mineral water helps with different heart and blood vessel diseases: arterial hypertension, stable angina pectoris, conditions
after suffering a heart attack, conditions after heart and blood vessels surgeries, obstructed peripheral arterial and vein circulation.
Mineral water is used in water-intake therapies and mineral baths. The temperature of the mineral bath is 30-33 degrees Celsius, it
lasts from 5 to 20 minutes, depending on the individual. A significant therapeutic factor in Radenci spa is also the sweet water mud.
Peloid compress has a soothing effect, alleviates the pain and has a positive effect on chronic inflammations. As one of the rare true
health resort towns, Radenci boasts as many as four natural healing factors: natural mineral water, thermal water, healing mud
(peloid) and beneficial climate with 250 sunny days per year. Radenci is know as oldest marathon. This is the oldest marathon, not
only in Slovenia but in all of south-eastern Europe. All tracks are officially measured by AIMS/IAAF. They are run on flat terrain
between fields and villages along the river Mura. As a participant describes the event: ―One of the last running events with heart and
soul‖ Three Hearts Marathon (Slovene: Maraton treh src) is a marathon, organised in Radenci in Slovenia. It has been taking place
since 1981 and attracts several thousand people each year. In addition to the marathon, a half marathon (21 km), recreative running
(10 km) and a course for juniors and teenagers are organised. The event was the Slovenian national championships race from 1992 to
1998 and has hosted the national race in even-numbered years since then, now sharing the honour with the Ljubljana Marathon.
The programme of ICNRDTNS includes plenary sessions, core dialogues, debates, discovery demos, knowledge exchange sessions,
knowledge factories, networking meet-ups, panel talks and poster presentations on specific topics and informal networking
opportunities in which practitioners share their experiences, ideas, new information and perspectives.
On behalf of the International Program Committee we express our sincere gratitude to our sponsors.
Come to magical, ancient and sunny Radin, participate in ICNRDTNS 2019 and be part of the ICNRDTNS events from September 18
– 20, 2019.
Dr. Matej Babič, Ph.D.
International Program Committee General Chair
3
Organized by Complex System Company, s.p.
Editor: Matej Babič, Ph.D.
Publisher: Complex System Company, s.p. Podgradje
Circulation: Elektronic edition
Date: Podgradje, 2019
Kataloţni zapis o publikaciji (CIP) pripravili v Narodni in univerzitetni knjiţnici v Ljubljani
COBISS.SI-ID=301710848
ISBN 978-961-290-461-6 (pdf)
icnrdtns.com
4
International Organizing Committee
1. Matej Babič, President, (Slovenia)
2. Branko Matovič, Director of Centre of Excellence-CEXTREME MAT, Institute for nuclear sciences Vinca, Belgrade
University, (Serbia)
3. Michele Calì, University of Catania, (Italy)
4. Abdulhamit Subasi, Professor ＆ Research Center Institute Director (RCI), IS Dept, Effat University of Saudi Arabia,
(Saudi Arabia)
5. Pavel Kovač, Head of Chair for Chip Removal Technologies University of Novi Sad, (Serbia)
6. Mileta Janjić, Vice Dean for S&R, University of Montenegro, (Montenegro)
International Scientific Committee
1. Carlos Pérez Bergmann, Federal University of Rio Grande do Sul, (Brasil)
2. Borislav Savković, University of Novi Sad, (Serbia)
3. Mehmet Can, International University of Sarajevo, (Bosnia and Herzegovina)
4. Saeed Qaisar, Effat University of Saudi Arabia, (Saudi Arabia)
5. Vesna Maksimović, Vinča Institute of Nuclear Science, (Serbia)
6. Aleksandra Zarubica, University of Niš, (Serbia)
7. Ivana Cvijović-Alagić, Vinča Institute of Nuclear Science, (Serbia)
8. Jelena Luković, Vinča Institute of Nuclear Science, (Serbia)
9. Michel Fillon, CNRS Director of Research, French National Center for Scientific Research, (France)
10. Božo Soldo, Head of Department of Civil Engineering, University North, (Croatia)
11. Cristiano Fragassa, Alma Mater Studiorum University of Bologna, Bologna, (Italy)
12. Kumar Dookhitram, University of Technology, (Mauritius)
13. Zlatko Botak, University North, (Croatia)
14. Alexander N. Herega, Deputy Director of Research and Production Center, Odessa (Ukraine)
15. Rita Ambu, University of Cagliari, Sardegna, (Italy)
16. Aleksandar Devečerski, Vinča Institute of Nuclear Science, (Serbia)
17. Gyula Varga, University of Miskolc, (Hungary)
18. Liliana-Laura Badita, INCD-Mechatronics & Measurement Technique, (Romania)
19. Zoran Jurkovic, University of Rijeka, (Croatia)
20. Pavel Beňo, Technical University in Zvolen, (Slovakia)
21. Elzbieta Bielecka, Military University of Technology, Warsaw, (Poland)
22. Ivan Liković, University of Novi Sad, (Serbia)
23. Manfred Zehn, Technical University Berlin, (Germany)
24. Dragan Marinković, Technical University Berlin, (Germany)
25. Jurij Mihelič, University of Ljubljana, (Slovenia)
26. Lenka Lhotská, Head of the Department of Natural Sciences, Faculty of Biomedical Engineering & Head of the Department
of Cognitive Systems and Neurosciences Czech Institute of Informatics, Robotics and Cybernetics, Czech Technical
University in Prague, (Czech republic)
27. Borbás Lajos, Edutus Egyetem (EDUTUS University), Budapest, (Hungary)
28. Luiz Moutinho, University of Suffolk, (England)
29. Dean Kinshuk, Dean of the College of Information at the University of North Texas, (USA)
30. Ravi Kumar, Laboratory for High Performance Ceramics (NAC207), Central XRD Laboratory (HSB138) Indian Institute of
Technology-Madras (IIT Madras), INDIA
5
Hotel Accommodation
The conference hotel is Radin (4*). The oldest mention of the town of Radenci dates back to 1436, when it was referred to as Radein.
Our hotel was named after this early variation. A four star Hotel Radin is the central hotel of the spa complex and it‘s connected with
Izvir hotel, thermal swimming pool part and health and wellness centre through a hallway. Most of the single-bed rooms in Radin
have French beds, and the specificity of the two-bed private rooms is that they have an extra bed.
The Hotel Radin is distinguished by a drinking lounge with a spring of natural mineral water. From the hotel, guests have direct
access to the thermal world, the health centre and the wellness centre. The hotel restaurant offers a selection of healthy dishes. Treat
yourself to a break at a health resort with a beneficial climate!
A number of hotel rooms in different price categories have been booked in Radenci with arrival September 18 and departure
September 20 for the congress in Hotel Radin. Additional nights can be confirmed if available.
The oldest mention of the town of Radenci dates back to 1436, when it was referred to as Radein. Our hotel was named after this early
variation. The Hotel Radin is distinguished by a drinking lounge with a spring of natural mineral water. From the hotel, guests have
direct access to the thermal world, the health centre and the wellness centre. The hotel restaurant offers a selection of healthy dishes.
Treat yourself to a break at a health resort with a beneficial climate!
6
TECHNOLOGIES (Big Data, Artificial Intelligence and Robotics, Smart Manufacturing,
Internet of Things/Emotions)
Luiz A M Moutinho Management and marketing futurecast, bioscience and neuroscience in marketing, futures research Univ. of Suffolk, England; Universidade Europeia and the Marketing School, Portugal, [email protected]
TECHNOLOGIES
(Big Data, Artificial Intelligence and Robotics, Smart Manufacturing, Internet of Things/Emotions)
Recent topics for possible presentations
7
Big Data
8
9
10
Artificial Intelligence and Robotics
11
12
13
Professor Luiz A M Moutinhomanagement and marketing futurecast,
biomarketing and neuroscience in marketing, futures researchUniversity of Suffolk, England, and University of South Pacific, Fiji
2018
FROM Reality Mining, On-Demand Society, Motion/Wearable Sensors, Digital Darwinism, the ME Web and Surface Computing
TO the Mobile Arcade, Sensorconomy, Predictive Intelligence, Beneficial Intelligence and Humarithms...
14
15
16
17
18
19
20
21
22
23
Smart Manufacturing
24
Internet of THINGS / ... of EMOTIONS
25
26
27
On the applicability of high-performance computing platforms: an overview
Jurij Mihelič, Tomaţ Dobravec, Uroš Čibej, Boštjan Slivnik
Faculty of Computer and Information Science, University of Ljubljana,
Ljubljana, Slovenia
{jurij.mihelic, tomaz.dobravec, uros.cibej, bostjan.slivnik}@fri.uni-lj.si
Abstract
To bypass physical limitations, high-performance computing (HPC) must resort to parallelism. Current HPC architectures implement
parallelism in many different ways, from multicore processors found even in most smartphones to large clusters of computing nodes
consisting of state-of-the-art processors supported with powerful general-purpose GPUs. Nevertheless, not every HPC architecture is
suitable for solving every problem; on the contrary, a problem can be solved on many different HPC architectures with varying degree
of efficiency. Nowadays many research and industrial organizations can afford a considerable amount of parallel computing
infrastructure on their own while aiming at a petaflop computational-power computer. This paper contains an overview of approaches
suitable for solving interdisciplinary problems, either research or industrial, using different kinds of HPC architectures.
Keywords – high-performance computing, processor, application, comparison
1. Introduction
Modern-day applications oftentimes require a lot of computational power. Various approaches exist to tackle such problems, one of
the solutions is to use high-performance computing (HPC) platforms that offer substantially more computational resources than a
typical off-the-shelf computer. Such platforms can be considered as a complementary technology to the well-known von Neumann
architecture as the latter is no longer being able to provide exponential improvements [1].
The common denominator of HPC platforms is the exploitation of parallelism, i.e., a division of work into tasks which are as
independent as possible. This division depends on at least two things, i.e., the problem at hand and the HPC platform itself. A task is
usually referred at as a thread which is described as a part of the program that can be executed independently; hence, a number of
threads can be executed simultaneously if the system consists of multiple processing units.
This paper offers an overview of four HPC platforms with example applications. The goal of the paper is to give a set of
reference points for practitioners that have similar (or same) applications. Often practitioners do not have the required knowledge
and/or skills to choose the most suitable platform and later also to exploit all its available power. The paper tries to give an insight into
these platforms and suggest the types of problems that are most suitable for each approach.
2. Shared memory multiprocessors
Shared memory multiprocessors are perhaps the simplest of all parallel and distributed systems. It's abstract model consists of a
number of independent processors all attached to a single memory (and an I/O system). Although the implementation details vary, the
memory, i.e., the data stored at any location in memory, is shared by all processors.
On the one hand, all multi-core systems, i.e., those based on multi-core CPUs, are shared-memory multiprocessors because
every core within a CPU run as an independent processor. The number of cores within a single CPU can exceed 16 and with the use
of hyperthreading, such a CPU can run 2 threads on each core virtually simultaneously. As just about all personal computers, either
laptops, desktops or servers, as well as most smartphones and video game consoles belong to this group, this makes shared memory
architecture the most prevalent parallel architecture today. On the other hand, specialized shared memory systems, e.g., Intel Xeon Phi
coprocessor with up to 72 cores capable of running 288 threads simultaneously, also belong to this group.
To fully utilize a shared memory multiprocessor, e.g., a multi-core platform, a program must be written so that the
computation is divided among several threads. In the ideal case, there are as many threads as the system can support and the
computation should be as evenly distributed among threads as possible. In a real application, this is, of course, hard to achieve.
As the shared memory architecture is relatively general (compared to other parallel architectures) it is suitable for a wide
range of problems: everything from Monte Carlo methods to solving differential equations numerically as well as discrete
mathematics problems and machine learning. However, in data-intensive applications the efficiency of a multi-core CPU might suffer
once all its cores start accessing the shared memory simultaneously.
28
The low-level approach to programming shared-memory systems is based on a C library implementing POSIX Threads (or
pthreads). The library provides functions for thread management and synchronization, and data exchange among threads. However, as
pthreads are implemented as a C library, many languages define their own implementations of threads that are supposed to be used
instead of pthreads. Java and Python, for instance, implement threads using their Thread classes, Go uses goroutines and channels,
Rust uses function spawn, etc.
While explicit threads are the appropriate approach when a program is divided into either a few long-lasting threads or a
number of virtually independent threads, they are ill-suited for most scientific and engineering applications. Hence, programming
frameworks like OpenMP (which enhances C/C++ and Fortran) are used instead [2]. OpenMP still supports explicit threads but adds a
support for parallel loops that execute in parallel using all available cores, and atomic operations that prevent data corruption in case
of data dependencies.
Furthermore, there are languages specially designed for parallel programming. Chapel, for instance, introduces domains that
allow large data structures, primarily multidimensional arrays, to be distributed effortlessly among threads. Naturally, it allows
parallel for loops (similar that OpenMP does) over these distributed data structures. Regardless of its high-level approach to
expressing parallelism in a program for a multi-core platform, programs written in Chappel are almost as fast as if they are written in
C and OpenMP once they are run on dozen cores or more.
3. General-purpose graphical processing units
A general-purpose graphics processing unit (GPU) is a many-core processor which enables very efficient simultaneous processing of
multiple tasks or threads. It consists of an integrated circuit in which a higher number of processors has been attached for enhanced
performance, reduced power consumption, and fast simultaneous execution of a selected algorithm on a set of input data. A GPU is
made of several streaming processors, organized as streaming multiprocessors, which are independent processing units. Each
streaming multiprocessor processes batches of blocks that are divided into groups of threads (warps), which are executed physically in
parallel [3]. A normal GPU execution performs several hundred or even thousands of threads at the same time.
One of the most famous and commonly used GPU platforms is called CUDA (Compute Unified Device Architecture),
created by NVIDIA Corp. CUDA also stands for a programming model used to solve problems that involve data-intensive computing
on a CUDA platform [4]. To solve problems with the CUDA platform an extension of the C programming language is provided. This
model treats GPU as a computational device operating as a coprocessor to the main CPU (host).
The execution of tasks on a GPU relies on the execution model which uses several threads to execute the same task on
different data. Threads that are combined into blocks (usually no more than 512 threads in one block) compose a computational grid.
Each thread of a grid executes a code (called a kernel), is synchronized and can communicate with other thread in the same block via
shared memory. Threads that are not in the same block cannot be synchronized and can only communicate by using common global
memory.
GPUs are used to execute data-parallel compute-intensive functions. A trivial but at the same time, a very eloquent example
of a problem suitable for GPU processing is a manipulation of an image (i.e., changing a color model or blurring image pixels), where
only a local image data is used to alter a pixel. A traditional (sequential) approach for solving this kind of problems is the use of a
loop over all the pixels, changing their information one-by-one. Using a GPU programming model this can be done much easier and
faster by defining a kernel, that is a code to alter a single pixel, and running this code in as many threads as the number of pixels that
need to be changed. Since GPU comes up with many processors, these threads will execute in parallel, each thread altering its pixel,
which will ensure an almost instant change of the image.
Due to its computational model, CUDA is not suitable for all applications. Its main drawback is in the memory usage which
makes it more suitable for applications with low memory bandwidth requirements. Promising problem areas for its efficeint use
include: linear algebra problems (clustering, data mining, finite element analysis, partial differential equations), spectral methods
(fluid dynamics, quantum mechanics, weather prediction), n-body methods (formation of galaxies, protein folding), structured and
unstructured grids (image processing, belief propagation), etc. [5]. Optimized usage of the global memory bandwidth (with
coalesced memory transactions) is a key feature of efficient algorithms that run on GPU. This can be easily achieved with some
frequently used algorithms like sorting arrays of data [6] or fast AES encryption [7].
29
4. Dataflow computing architecture
Another approach to high-performance computing is a dataflow architecture, where the availability of data is guiding the execution
and not vice-versa, as is the case with control-flow computers. A dataflow program can be viewed at as a directed graph, where nodes
represent operations and edges the flow of data between these operations. Various hardware implementations for dataflow graphs
exist [8], but in this overview we focus on the dataflow solution as provided by the Maxeler Technologies, Inc.
Their solution is a hybrid architecture consisting of both classical central processing unit (CPU) as well as dataflow engine
(DFE). Here, the former is responsible for the setup and supervision of the latter as well as for the processing of the data; however, the
main computation is performed in DFE which consists of an FPGA chip with additional memory and logic. A user solving the
problem must thus be able to program for both platforms. In particular, she must write a control-flow program in the C programming
language as well as a dataflow program in the Java-like programming language called MaxJ. Here, the MaxJ program consists of a
code for managing the streams of data and a code (called kernel) for manipulating these streams. See e.g. [8, 9, 10, 11] for details.
To efficiently perform the programming task one must consider several acceleration techniques such as data choreography,
i.e., the planning of data movement between computing elements in order to ensure that the data is available at the moment when it is
required, efficient pipelining, i.e., different dataflow graphs may conceptually represent the same algorithm, thus, considering
variations (e.g., switching the order of independent operations) may result in changes in performances, data precision, i.e., the
dataflow architecture is very flexible in defining numeric data types as the data precision can be precisely specified for almost any
operation, thus, reducing the precision may result in the increase of the performance, and different algorithms, i.e., algorithms deemed
successful for the classical von Neumann computer may perform poorly on dataflow computer. Solving the problems and
implementing the solutions may, therefore, be quite daunting for the user. However, there are also some clear advantages of the
dataflow technology one of them being a low energy consumption.
Finally, let us conclude this section by listing several applications of the dataflow paradigm. Matrix calculations such as
multiplication and exponentiation are straightforward to implement as well as operations on graphs represented with matrices such as
calculating graph walks, counting triangles and shortest paths [9]. The experimental evaluation also shows that the evaluation of
polynomials is also a promising application. An interesting application is also a simplex algorithm for solving linear programming
problems [11]. Nevertheless, dataflow approach was used in other applications such as bitonic sorting, bitcoin mining, fast Fourier
transform, neural networks, linear regression, ray tracing, classification, etc. See Maxeler‘s application gallery
(https://appgallery.maxeler.com) for many more examples.
5. Distributed memory systems
To avoid a restriction imposed by a shared memory system where processors must wait for each other if they try to access the system's
memory simultaneously, a distributed memory system contains a number of computing nodes each consisting of a processor and its
own local memory. All local memories together form the global memory of the system, yet each processor can directly access data
stored in its local memory. For accessing nonlocal data it must communicate with the computing node where the data is stored; the
implementation details may vary, but the access takes more time than if data is stored in a local memory. At the lower end, a
distributed memory system can be made by connecting several consumer-level PC using standard TCP/IP connections. At the higher
end, supercomputers use specialized interconnection networks, e.g., InfiniBand or Gemini, that allows a processor to access nonlocal
memories as well without interrupting other processors; this approach results in hybrid architectures.
Even though not every computational problem can be efficiently parallelized for a distribute memory system, using such
computer architecture can make solving most of them significantly faster. The range of such problems is indeed wide. Traditionally it
included intensive numerical computations, e.g., solving differential equations [12] and modeling natural phenomena [13], weather
and nukes, but nowadays it includes anything from combinatorics and big data as well.
6. Conclusions
The paper presents a set of HPC platforms that we deem as the currently most advanced and useful for practical applications. Due to a
an increased demand for high computational power in practice, an overview is a great starting point for any practitioner. In Table 1 we
gathered a few properties of each platform which are of particular interest when searching for a suitable solution to a practical
problem.
30
Price
Energy
consumption
Hardness of
adaptation
Class of
applications
Multicore low medium low broad
GPGPU medium medium medium specific
Dataflow high low high very specific
Clusters high high high broad
Table 1: Qualitative comparison of the parallel platforms
Multicore platforms have become the default in every personal computer, which means that their price is very low, it is easy
to adapt them due to the widespread availability of various tools, and also the class of applications suitable for this platform is very
broad. The next most widely available platforms are GPUs, since most of today's graphic cards are GPUs and they can be used for
general processing as well. The tools for their adaptation are more specific than for multiprocessors and their field of applications is
much narrower. Dataflow computers and clusters are much more specialized platforms, their price is much higher, and it is also much
harder to adapt these two technologies due to very specialized tools. One of the most significant differences between these two
platforms is the energy consumption. Dataflow computers are most widely known for their low power consumption, whereas clusters
require a substantial amount of energy to operate.
References
[1] Hennessy, J. L., Patterson, D. A. (2019). A new golden age for computer architecture, Communications of the ACM, Vol. 62, No.
2, Pages 48-60, http://doi.acm.org/10.1145/3282307.
[2] Trobec, R., Slivnik, B., Bulić, P., Robič, B. (2018). Introduction to parallel computing: from algorithms to programming on state-
of-the-art platforms, (Undergraduate topics in computer science), Springer.
[3] Andre, R. B., Trond, R. H., & Martin, L. S. (2013). Graphics processing unit (GPU) programming strategies and trends in GPU
computing. Journal of Parallel and Distributed Computing, 73(1), 4–13, Metaheuristics on GPUs.
[4] Ghorpade, J., Parande, J., Kulkarni, M., & Bawaskar, A. (2012). GPGPU processing in CUDA architecture. Advanced Computing:
An International Journal, 3(1), 105–120.
[5] Wu Feng, Vincent Hindriksen. (2013) The 13 application areas where OpenCL and CUDA can be used. StreamHPC. Available at
https://bit.ly/2Fou3wZ.
[6] Darko Boţidar, Tomaţ Dobravec. (2015) Comparison of parallel sorting algorithms. Available at
https://arxiv.org/abs/1511.03404.
[7] Kristian Zupan. (2015) Comparison of cryptographic algorithm implementations on CPU in GPU (master thesis, mentor T.
Dobravec, Ljubljana UL-FRI). Available at http://eprints.fri.uni-lj.si/2994/.
[8] Šilc, J., Robič, B., Ungerer, T. (1999) Processor Architecture: From Dataflow to Superscalar and Beyond, Springer.
[9] Mihelič, J., Čibej, U. (2019). Matrix-based Algorithms for DataFlow Computer Architecture, In: Milutinović, V., Kotlar, M.,
Exploring the DataFlow Supercomputing Paradigm: Example Algorithms for Selected Applications., Springer, Pages 91-131.
[10] Mihelič, J., Čibej, U. (2017). Dataflow Processing of Matrices and Vectors: Experimental Analysis, In: Informatics 2017:
proceedings, Pages 265-270.
[11] Čibej, U., Mihelič, J. (2017). Adaptation and evaluation of the simplex algorithm for a data-flow architecture. In: Hurson, A. R.,
Milutinović, V., Advances in computers, Vol. 106, Pages 63-105.
[12] Trobec, R., Orel, B., Slivnik, B. (1996) Coarse-grain parallelisation of multi-implicit Runge-Kutta methods. In: Numerical
analysis and its applications: 1st international workshop, WNAA '96. Berlin, 498-504.
[13] Trobec, R., Slivnik, B., Geršak, B., Gabrijelčič, T. (1998). Computer simulation and spatial modelling in heart surgery.
Computers in Biology and Medicine, 28(4), 393-403.
31
Jurij Mihelič is an Assistant Professor in Computer Science at University of Ljubljana, where he received Ph.D. in 2006. His
research interests include algorithm engineering, experimental algorithmics, combinatorial and graph problems, system software and
programming languages.
Tomaž Dobravec received B.Sc. (1996) in mathematics and PhD (2004) in computer science, both from the University of Ljubljana.
He is an Assistant Professor at the Faculty of Computer and Information Science, University of Ljubljana. His main research interests
are algorithm design, computer security and networks.
Uroš Čibej received his Ph.D. in computer science from the University of Ljubljana in 2007. He is with the Laboratory of
Algorithmics, where he pursues research in location problems, distributed systems, computational models and complexity, halting
probability, and graph algorithms.
Boštjan Slivnik is an Assistant Professor of Computer Science at the University of Ljubljana, where he received his Ph.D. in 2003.
His research interests include compilers and programming languages with the special focus on parsing algorithms and formal
languages, scheduling and distributed algorithms, and software engineering. He has been a member of the ACM since 1996.
32
ENGINEERING ECONOMIC ANALYSIS OF WATER, ELECTRICITY AND
ABRASIVES COSTS AND THEIR EFFECT ON THE PRICE OF ABRASIVE WATER
JET MACHINING
Ramiz Kurbegović, Mileta Janjić1, Milan Vukčević
Faculty of Mechanical Engineering, University of Montenegro, 81000, Podgorica, Montenegro
Abstract
Abrasive water jet machining is one of the non-conventional production technologies, and for its many advantages over
other cutting technologies, it is often used in the industry. However, to find a wider application and improve its
performance, it is necessary to perceive a number of input and output parameters and their impact on the machining
process.
The aim of this paper is to investigate the effect of water, electricity and abrasives costs, as most dominant operating
costs, on the price of abrasive waterjet machining and to find the economically acceptable variance of process parameters.
Major significant process factors affecting the price of abrasive waterjet machining were determined. Relationship
between the cost of abrasive waterjet machining and the costs of water, electricity, and abrasives has been formed.
Keywords: Abrasive Water Jet, Engineering economic analysis, Water jet lagging
1. Introduction
The principles on which the abrasive water jet machining process is based on is erosion. Some authors explain the process
of erosion as a kind of abrasive wear, at which abrasive particles and water jet repeatedly impact the surface, resulting in
flushing of the material from that surface [1,2].
The cutting front geometry of the workpiece machined by the abrasive water jet is influenced by machining parameters
such as traverse speed, operating pressure, abrasive flow rate, standoff distance, depth of cut and angle of cutting [1,3].
Defining the geometry of the cutting front, is in fact, the determination of the deviation - lagging, Ylag, of the abrasive
water jet from the vertical line. The line that defines the lagging of abrasive water jet is described by Zeng et al. [4] as a
parabola.
Analyzing and comparing costs represents the basic aspects of engineering practice [5]. Since there are a large number of
costs that affect the final price of the abrasive water jet machining, the analysis of the most dominant operating costs will
provide insight into the justification of the variation of certain process parameters. The cost of water, electricity, and
abrasive, represent the most dominant exploitation costs.
2. Experimental details
Experimental research was realized to define the influence of water, electricity and abrasives costs on the price of
abrasive water jet machining and to achieve better machining performance.
Samples presented by Marušić et al. [6], concerning the influence of the water pressure (𝑝), traverse speed (𝑣𝑔) and
abrasive flow (𝑚 𝑎 ) on the abrasive water jet lagging, were used for creating experimental variants for this work.
Machining parameters are shown in table 1.
The system used for machining the samples, is the product of PTV JETS, model 3,8/60 Classic, Czech Republic. Orifice
inner diameter was 0.254 mm. Inner diameter of the focusing tube was 1.02 mm with a length of 76 mm. Jet impact angle
1 - name of the presenter
33
was 90°. Abrasive material was Garnet mesh 80. Work piece material was stainless steel X5CrNi 18–10, thickness of 30
mm. Distance between focusing tube and material was 2 mm.
Maximum installed power of the system is 𝑃𝑚𝑎𝑥 = 49,4 𝑘𝑊. 3-phase asynchronous motor on the pump has a nominal
power of 37 kW. Other consumers, with an installed power of 𝑃𝑂𝐶 = 12,4 𝑘𝑊, are air compression machine,
water treatment system, abrasive supply system, oil cooling system and CNC (Computer Numerical Control) workstation
with automation. Maximum water flow of the system is 3,8 l/min (𝑄𝐴).
Table 1. Samples and its parameters [6]
Sample 4 Sample 5 Sample 10 Sample 11 Sample 13 Sample 17
𝑣4 = 40 𝑚𝑚
𝑚𝑖𝑛
𝑝 = 413 𝑀𝑃𝑎
𝑚 𝑎 = 400 𝑔
𝑚𝑖𝑛
𝑣5 = 50 𝑚𝑚
𝑚𝑖𝑛
𝑝 = 413 𝑀𝑃𝑎
𝑚 𝑎 = 400 𝑔
𝑚𝑖𝑛
𝑣 = 35 𝑚𝑚
𝑚𝑖𝑛
𝑝3 = 290 𝑀𝑃𝑎
𝑚 𝑎 = 400 𝑔
𝑚𝑖𝑛
𝑣 = 35 𝑚𝑚
𝑚𝑖𝑛
𝑝4 = 245 𝑀𝑃𝑎
𝑚 𝑎 = 400 𝑔
𝑚𝑖𝑛
𝑣 = 35 𝑚𝑚
𝑚𝑖𝑛
𝑝 = 413 𝑀𝑃𝑎
𝑚 𝑎 ,1 = 350 𝑔
𝑚𝑖𝑛
𝑣 = 35 𝑚𝑚
𝑚𝑖𝑛
𝑝 = 413 𝑀𝑃𝑎
𝑚 𝑎 ,5 = 150 𝑔
𝑚𝑖𝑛
Measurements for determining water jet lagging were performed in ten places (at the same distance) along with the
sample thickness using optical microscope.
For the description of the operation of centrifugal pumps the affinity laws given by Equations (1), (2) and (3) can be used.
They are useful for quick and precise enough analysis. 𝑄𝐴
𝑄𝐵=
𝑛𝐴
𝑛𝐵 (1)
𝑝𝐴
𝑝𝐵=
𝑛𝐴
𝑛𝐵
2
(2)
𝑃𝐴
𝑃𝐵=
𝑛𝐴
𝑛𝐵
3
(3)
where: Q – flow, p – pressure, P – power of the motor and n – speed of pump impeller.
Needed Strength of the pump (and motor) is described with equation (4).
𝑃 =𝑝 ∙ 𝑄
𝜂𝑈 (4)
2.1. Experimental Procedure
Variant A of this work will be sample 17 from table 1. For getting almost identical values of water jet lagging, as in
sample 17, linear interpolation was used on water pressure parameter on samples 10 and 11. Calculated value of water
pressure is 𝑝𝑥 = 265 𝑀𝑃𝑎 and that will be the Variant B of this work. Both variants and its parameters are presented in
table 2.
34
Table 2. Variants for the analysis
Variant A Variant B
𝑣 = 35 𝑚𝑚
𝑚𝑖𝑛
𝑝 = 413 𝑀𝑃𝑎
𝑚 𝑎 ,5 = 150 𝑔
𝑚𝑖𝑛
𝑣 = 35 𝑚𝑚
𝑚𝑖𝑛
𝑝𝑥 = 265 𝑀𝑃𝑎
𝑚 𝑎 = 400 𝑔
𝑚𝑖𝑛
Costs will be analyzed for a year of machining, with a straight line machining of 30 m/day, 20 working days/month.
Water price will be 𝑈𝑃𝑊 = 1 €/𝑚3 and price of abrasive Garnet # 80 will be 𝑈𝑃𝐴 = 400 1 €/𝑡. Electricity price will be
calculated in accordance with the Price List of Elekroprivreda Crne Gore A.D. Niksic [7] for a basic model consumer
with a single tariff meter connected to a 10 kV line, measuring the average 15-minute load, active and reactive power.
3. Results
According to the given length of machining (lDM) and the speed of cutting head (𝑣) we can get daily machining time
𝑡𝐷𝑀 = 857,14 𝑚𝑖𝑛/𝑑𝑎𝑦, which is same for both variants.
Using the 𝑡𝐷𝑀 , number of working days (𝑁𝑊𝐷), abrasive mass flow for variants A (𝑚 𝑎 ,5) and B (𝑚 𝑎 ) and price of
abrasive (𝑈𝑃𝐴), we can calculate abrasive grand total for A (𝑃𝐴𝐴) i B (𝑃𝐴𝐵):
𝑃𝐴𝐴 = 𝑡𝐷𝑀 ∙ 𝑁𝑊𝐷 ∙ 𝑚 𝑎 ,5 ∙ 𝑈𝑃𝐴 ∙ 10−6 ≅ 12.342,82 €/𝑦𝑒𝑎𝑟 (5)
𝑃𝐴𝐵 = 𝑡𝐷𝑀 ∙ 𝑁𝑊𝐷 ∙ 𝑚 𝑎 ∙ 𝑈𝑃𝐴 ∙ 10−6 ≅ 32.914,18 €/𝑦𝑒𝑎𝑟 (6)
Conditions for electricity cost calculation: 𝜂𝑃,𝐴 = 𝜂𝑃,𝐵 = 𝜂𝑃 = 80 %, 𝜂𝐸𝑀 ,𝐴 = 𝜂𝐸𝑀 ,𝐵 = 𝜂𝐸𝑀 = 93 %, 𝜂𝑃𝑁 ,𝐴 = 𝜂𝑃𝑁 ,𝐵 =𝜂𝑃𝑁 = 98 %, 𝜂𝐼,𝐴 = 𝜂𝐼,𝐵 = 𝜂𝐼 = 90 %, 𝑐𝑜𝑠𝜑𝐴 = 𝑐𝑜𝑠𝜑𝐵 = 𝑐𝑜𝑠𝜑 = 1, and 𝑛𝐴 = 1480 𝑟𝑝𝑚, where: 𝜂𝑃 , 𝜂𝐸𝑀 , 𝜂𝑃𝑁 , 𝜂𝐼 –
efficiency of pump, motor, pipe network and intensifier, 𝑐𝑜𝑠𝜑 – power factor, and 𝑛𝐴 – rotation speed of motor. Total
system efficiency (𝜂𝑇) is 65,621 %.
Using affinity laws (1), (2), (3), and equation (4), we can calculate pump impeller speed (𝑛𝐵 = 1186𝑟𝑝𝑚) and
flow for variant B (𝑄𝐵 = 3,05 𝑙/𝑚𝑖𝑛), required power for electric motor for variant A (𝑃𝐴 = 23,92 𝑘𝑊) and B (𝑃𝐵 =12,28 𝑘𝑊). Total required power of A (𝑃𝑇,𝐴 = 30,12 𝑘𝑊) and B (𝑃𝑇,𝐵 = 18,48 𝑘𝑊) variant will be a sum of required
powers of electric motors with a half of installed power of other consumers (𝑃𝑂𝐶). Price of the electricity, for variants A
and B, are shown in tabs 3, 4 and 5.
Table 3. Monthly price of the electricity for a variant A
Name Unit of
measure
Unity price Consumption Price
[€c/kWh] [kWh] [€]
Active electricity kWh 4,4932 8605,69 386,67
Reactive electricity kVArh 0,8986 0,00 0,00
Engaging transmission capacity kW 9,3632 30,12 282,02
Losses in the distribution system kWh 0,1113 8605,69 9,58
Compensation to the market operator monthly 0,0175 8605,69 150,60
Price w/o VAT (21%) 828,87
Tabela 4. Monthly price of the electricity for a variant B
Name Unit of
measure
Unity price Consumption Price
[€c/kWh] [kWh] [€]
Active electricity kWh 4,4932 5279,98 237,24
Reactive electricity kVArh 0,8986 0,00 0,00
Engaging transmission capacity kW 9,3632 18,48 173,03
35
Losses in the distribution system kWh 0,1113 5279,98 5,88
Compensation to the market operator monthly 0,0175 5279,98 92,40
Price w/o VAT (21%) 508,55
Table 5. Price of the electricity
Name Price
[€/year]
Price of the electricity for variant A (𝑃𝐸𝐴) 9.946,44
Price of the electricity for variant B (𝑃𝐸𝐵) 6.102,60
Total price of water consumption (𝑃𝑊𝐴) and (𝑃𝑊𝐵) is calculated using equations 7 and 8.
𝑃𝑊𝐴 = 𝑈𝑃𝑊 ∙ 𝑄𝐴 ∙ 60 ∙ 𝑡𝐷𝑀 ∙ 𝑁𝑊𝐷 ≅ 781,95 €/𝑦𝑒𝑎𝑟 (7)
𝑃𝑊𝐵 = 𝑈𝑃𝑊 ∙ 𝑄𝐵 ∙ 60 ∙ 𝑡𝐷𝑀 ∙ 𝑁𝑊𝐷 ≅ 627,62 €/𝑦𝑒𝑎𝑟 (8)
Total price of the machining for A (𝑀𝑃𝐴) and B (𝑀𝑃𝐵) is calculated using equations 9 and 10.
𝑀𝑃𝐴 = 𝑃𝐸𝐴 + 𝑃𝐴𝐴 + 𝑃𝑊𝐴 = 23.071,21 €/𝑦𝑒𝑎𝑟 (9)
𝑀𝑃𝐵 = 𝑃𝐸𝐵 + 𝑃𝐴𝐵 + 𝑃𝑊𝐵 = 39.644,40 €/𝑦𝑒𝑎𝑟 (10)
Relationship between total price of the machining for A and B is shown in equation 11. 𝐶𝐾𝐵
𝐶𝐾𝐴= 1,7183 (11)
Calculating the price difference, we can improve productivity of the variant A for the given difference. Improvement in
abrasive flow of the variant A (𝐼𝐴) will be:
𝐼𝐴 =(𝑀𝑃𝐵 − 𝑀𝑃𝐴)
𝑈𝑃𝐴 ∙ 𝑁𝑊𝐷 ∙ 𝑡𝐷𝑀≅ 201 𝑔/𝑚𝑖𝑛
Sample 13 have similar machining parameters for economically identical and improved variant A. To get the productivity
improvement, linear interpolation was carried on samples 4 and 5 (shown in table 1) to get traverse speed value. That
value is 𝑣𝐼 = 41,7 𝑚𝑚/𝑚𝑖𝑛.
Improvement in productivity of the abrasive water jet system will be: 𝑣𝐼
𝑣= 1,1914 (12)
4. Conclusions
Total price of the machining, using parameters from variant B, is 71,83% higher than a total price of the machining using
parameters from variant A, with a same productivity and quality if the machining.
Prices of water and electricity influences less on a total price of the machining than a price of abrasive. From this we can
conclude that the variance of abrasive flow, as most dominant cost, represents an economically sound approach, while
variance of pressure – lowering working pressure, represents an economically wrong approach.
Machining speed of variant A could be improved to 41,7 mm/min, for the same quality and price as in variant B, and
improve productivity of the system by 19,14%.
Acknowledgment
The paper is a result of support of my mentor and the Ministry of Science of Montenegro.
36
References
[1] Mombar, AW, Kovačević, R (1998). Principles of abrasive water jet machining, Springer, London.
[2] Arola, D, Ramulu, M (1993). Mechanism of material removal in abrasive waterjet machining, In: Proceedings of
the 7th Water Jet Conference, Washington, 46-64.
[3] Hasish, M (1984). A modeling study of metal cutting with abrasive waterjets, Journal of Engineering Materials
and Technology, 106, 88-100.
[4] Zeng, J, Heines R, Kim TJ (1991). Characterization of energy dissipation phenomena in abrasive water jet
cutting, In: Proceedings of the 6th American Water Jet Conference, St. Louis, 163-177.
[5] Vukčević, M (2012). Inţenjerska ekonomija, Univerzitet Crne Gore - Mašinski fakultet, Podgorica.
[6] Marušić, V, Baralić, J, Nedić, B, Rosandić, Ţ (2013). Effect of machining parameters on jet lagging in abrasive
water jet cutting, Technical Gazette, 20, 4, 677-682.
[7] Elektroprivreda Crne Gore A.D. Nikšić. Cijene za snabdijevanje krajnjih kupaca električne energije, from
https://www.epcg.com/sites/epcg.com/files/multimedia/
main_pages/files/2014/04/cijene_za_snabdijevanje_krajnjih_kupaca_elektricne_
energije_1405_2019.pdf, 03.08.2019.
37
Machine Learning in Medical Imaging
Martin Bobák1, Ladislav Hluchý
1, Mara Graziani
2, Henning Müller
2
1 Institute of Informatics, Slovak Academy of Sciences
Bratislava, Slovakia
{martin.bobak, ladislav.hluchy}@savba.sk
2 University of Applied Sciences Western Switzerland (HES-SO)
Sierre, Switzerland
{mara.graziani,henning.mueller}@hevs.ch
Abstract
The paper describes the machine learning in medical imaging which represent one of the exascale services prepared in the PROCESS
project. It also presents the architecture capable to handle such services with quantitative analysis performed at two computing sites.
Keywords – machine learning, deep learning, exascale architecture, medical imaging, high-performance computing, cloud
computing.
1. Introduction
Digital histopathology is the automatic analysis of a biopsy or surgical tissue specimens that are captured by a high-resolution scanner
and stored as Whole Slide Images (WSIs). WSIs are usually stored in a multi-resolution pyramid structure, where each layer contains
down-sampled versions of the original image. The amount of information in WSIs is large, since it includes tissue that is not relevant
for cancer diagnosis (e.g. background, healthy tissue, etc.) For this reason, machine learning and deep learning models are built to
detect Regions of Interest (ROIs) within WSIs. ROIs are portions in the WSI where the cancer is visible and therefore contain relevant
information to train the network.
2. Use case description
Figure 1 shows a data pre-processing pipeline. As a first step the raw WSIs are analysed at a very low resolution, and tissue is filtered
from the background. Based on physician‘s annotations, tumor regions are isolated. These regions represent ROIs that are used to
perform network training. From the normal tissue and from tumor ROIs, patches are extracted at a higher level of magnification.
Higher resolution highlights qualitative features of the nuclei which are essential for cancer detection. For instance, recent research
has shown that performance of classifiers improves with higher resolution patches.
Figure 1: WSIs preprocessing pipeline - Normal tissue and tumor tissue masks are extracted and high-resolution patches are sampled
from the selected regions.
38
The use case application is organized in three layers. Layer I implements the extraction of patches of dimensions 224x224 pixels from
the gigapixel slides of breast lymph node tissue. Patches are randomly sampled from the slide, in which areas of tumour were
annotated by a physician. Patches belonging to a tumorous region are assigned a ‗tumor‘ label (a Boolean variable equals to true). The
extracted data are stored in an intermediate dataset with the corresponding labels. Layer II loads the intermediate dataset of patches
and labels and trains a state-of-the-art deep convolutional network to classify the two patch types. Different models can be chosen by
a configuration parameter. Layer III focuses on network robustness and interpretability. A summary of the use case application layers
can be found in Napaka! Vira sklicevanja ni bilo mogoče najti..
Layer I:
Data pre-processing and patch
extraction
Layer II:
Local and distributed training
Layer III:
Performance boosting and
interpretability
Creation of normal and tumour
tissue masks from the physician‘s
annotations
State of the art deep
convolutional networks
currently implemented:
Resnet50, Resnet101,
InceptionV3
Generation of intermediate
visualizations
Random sampling of high-
resolution patches and labelling
Local training on single and
multiple GPUs Feature importance analysis
Intermediate storage of the patches
on H5DS Training on HPC clusters Perturbation robustness analysis
Distributed training
Table 1: Camnet - Interchangeable networks for Camelyon.
3. PROCESS platform
The reference exascale architecture (see Figure 2) is divided into the following parts (from top to bottom):
Users of the exascale scientific applications (in yellow) - the exascale system has to support functionalities required by its
user communities. The best way is to build it on containerization. All of the applications are stored in a containerized
repository that is available to use communities.
Virtualization layer (in blue) - is situated between the containerized application repository and platform infrastructure
managers. Interoperability of data and computing infrastructures is the key and critical requirement of the exascale systems.
Data management (in green) - could be divided into two main groups: distributed data federation and metadata. The
metadata module has to be federated and distributed as well as the management system for the data infrastructure itself. At
this level of the infrastructure, the system architect has to be careful whether the component will be containerized, or not.
Micro-services serve as adapters and connectors to infrastructural services. They are integrated into a containerized micro-
infrastructure, which is customized according to requirements coming from a use case and connecting them to a distributed
virtual file system.
Computing Management (in red) - this part of the infrastructure is related to scheduling and monitoring computing
resources. Two kinds of resources are supported, namely: high-performance computing (HPC) resources, and cloud
resources. HPC manager is based on a queuing approach. The manager of the cloud resources is based on the
REpresentational State Transfer (REST) Application Programming Interface (API). Both types of resources are often
enriched by support from high-throughput resources or accelerated resources.
39
Figure 2:Reference exascale architecture.
4. Results
Experiments were computed on UvA (University of Amsterdam) and AGH (Akademia Górniczo-Hutnicza im. Stanisława Staszica w
Krakowie) computing sites for Layer I and Layer II of the use case software. The use case application handling of the resources and
access to CPU and GPU memory was intensively optimized. Table 1 and Table 2 illustrate the current status of the application layers I
and II. First order statistics of computational requirements for Layer I are reported in Table 1. First order statistics for the time
requirements of Layer II are reported in Table 2. Performance of Layer II is also reported in Table 2 in the form of model accuracy.
Location
Hes-so UvA AGH
Patch sampling time [s/patch] 0.41 - -
Data loading time [ms/patch] 2.0 1.5 0.5
Table 1: Time baselines of the first layer: Data Preprocessing and Patch Extraction.
Resource (Location) Hes-so UvA AGH
Training accuracy 96,91±0,45 96,1±0,24 84.3
Validation accuracy 85,6±6,2 83,1±0,14 93.7
Training time [s/epoch] 2440,80 1203,68 17277
10 epochs time [h] 7 h 3.34 h 47 h
Table 2: Performance baselines for model 1 (ResNet50) of the second layer: Interchangeable Network Model.
40
The performance evaluation highlights very efficient data extraction and loading at AGH, with only 0.5 ms to load a single patch. The
high-performance GPUs available at UVA, by contrast, provide fast computations on a single GPU, halving the computational time of
the model training (from 7 hours to 3.34 hours).
5. Conclusion
The paper presents the medical use case of the PROCESS project that is focused on exascale learning on medical image data. The
requirements coming from the use case are handled by the exascale reference architecture. It is based on containerization and virtual
machines supported by an exascale capable distributed virtual file system, and computing manager. The last section presents
experimental results that were performed at two computing sites and show their advantages and disadvantages within the use case
scenario.
Acknowledgement
This work is supported by the "PROviding Computing solutions for ExaScale ChallengeS" (PROCESS) project that has received
funding from the European Union‘s Horizon 2020 research and innovation programme under grant agreement No 777533, the project
APVV-17-0619 (U-COMP) "Urgent Computing for Exascale Data", and the project VEGA 2/0167/16 "Methods and algorithms for
the semantic processing of Big Data in distributed computing environment"
Martin Bobák (Dr.) is a research scientist at the Institute of Informatics of the Slovak Academy of Sciences. He received Ph.D.
degree in Applied Informatics from the Institute of Informatics of the Slovak Academy of Sciences in 2017, and M.Sc. degree in
Computer Science from Faculty of Mathematics, Physics and Informatics, Comenius University in Bratislava (Slovakia) in 2013. His
research interests are cloud computing, algorithms and data structures. He is (co-)author of several scientific papers and has
participated in international (European Union‘s Horizon 2020 research and innovation programme), and national research projects as
a key person and scientific coordinator. He is a reviewer for international scientific conferences and journals, and a teaching assistant
at Faculty of Informatics and Information Technologies, Slovak University of Technology in Bratislava (Slovakia).
Ladislav Hluchý (Assoc. Prof.) is the Head of the Parallel and Distributed Information Processing department, the vice-director of
the Institute of Informatics, Slovak Academy of Sciences (IISAS). He received M.Sc. and Ph.D. degrees, both in computer science.
He is RD Project Manager, Work-package leader in a number of 4FP, 5FP and 6FP projects, as well as in Slovak RD projects (VEGA,
APVT, SPVV). He is a member of IEEE, ERCIM, SRCIM, and EuroMicro consortiums, the editor-in-chief of the journal Computing
and Informatics. He is also (co-)author of scientific books and numerous scientific papers, contributions and invited lectures at
international scientific conferences and workshops. He also gives lectures at Slovak University of Technology and is supervisor and
consultant for Ph.D., master and bachelor studies.
Mara Graziani (M.Phil.) is currently a PhD student at the Computer Science faculty at the University of Geneva (Switzerland). Her
research focus is on interpreting Deep Learning for medical applications. She completed the Masters of Philosophy in Machine
Learning, Speech and Language Technology at the University of Cambridge (UK) in 2017. She has a BEng. in Information
Technology Engineering at La Sapienza Univerisity of Rome.
Henning Müller (Prof.) studied medical informatics at the University of Heidelberg, Germany, then worked at Daimler-Benz
research in Portland, OR, USA. From 1998-2002 he worked on his PhD degree at the University of Geneva, Switzerland with a
research stay at Monash University, Melbourne, Australia. Since 2002, Henning has been working for the medical informatics service
at the University Hospital of Geneva. Since 2007, he has been a full professor at the HES-SO Valais and since 2011; he is responsible
for the eHealth unit of the school. Since 2014, he is also professor at the medical faculty of the University of Geneva. In 2015/2016 he
was on sabbatical at the Martinos Center, part of Harvard Medical School in Boston, MA, USA to focus on research. Henning is
coordinator of the ExaMode EU project, was coordinator of the Khresmoi EU project, scientific coordinator of the VISCERAL EU
project and is initiator of the ImageCLEF benchmark that has run medical tasks since 2004. He has authored over 400 scientific
papers with more than 12,000 citations and is in the editorial board of several journals.
41
Controllable synthesis of multidoped ceria nanopowders for various applications
Branko Matovic, Snezana Boskovic
University of Belgrade, Institute for nuclear sciences Vinca
Belgrade, 11000, Serbia, [email protected]
Abstract
Development of new technologies for the synthesis of nanoparticles and nanostructured materials, which are profitable for industrial
production and environmental safety are the subject of the great number of studies, nowadays. These materials have new and specific
physical properties and find application in almost all spheres of human life. In nanoparticle materials besides appearance of new
properties significant changes in existing physical characteristics compared to microcrystalline ones take place, which in some cases
can differ to several orders of magnitude. These specific changes in properties of the nanomaterial are observed in the change of
magnetic, mechanical and optical characteristics, phase relations, electrical conductivity, etc. In this presentation, self propagating
room temperature syntheis has been applied for controllable synthesis of nanostructured CeO2 powders with fluorite-type structure.
Powder properties such as, crystallite and particle size, their thermal stability as well as, lattice parameters have been studied. Crystal
structure of fluorites, point defects, specific features and properties connected to it are discussed. In addition, the innovative method of
nanopowders synthesis, and properties of ceria based materials will be discussed through: crystal structure and defect chemistry,
synthesis of nanostructured solid solutions, hot consolidation of ceria nanopowders, some key properties of nanostructured ceria.
Keywords – ceria solid solutions, crystal structure, lattice defects, nanopowders synthesis, densification, properties
1. Introduction
Ceria (CeO2) is a technologically important material for many fascinating reasons. The capability of ceria to change oxidation state in
relative ease way affects the local structure and functionality of ceria 1. Ceria is also of great importance for its wide applications as
a promoter in three-way catalysts (TWCs) for the elimination of toxic auto-exhaust gases, fuel cells, low-temperature water–gas shift
reaction, oxygen permeation membrane systems, oxygen sensors, electrochromic thin-film application, ultraviolet absorbent, glass-
polishing materials as well as its role in environmental chemistry, and medicine 1. Thus, the excellent oxygen storage behavior
results from the balance between reduced and oxidized states of Ce ion, i.e., Ce+3
, Ce+4
and from increased oxygen transport capacity.
However, mentioned properties are strongly dependent on the structural features. Therefore, for the design of ceria based materials
with high oxygen storage and transport capacity it is important to know how to increase the number of structural defects (oxygen
vacancies) and to maintain at the same time a fluorite-type crystal structure. There are two possibilities to obtain ceria-based oxide
material as an oxygen storage component, either by promotion of Ce4+
reduction into Ce3+
or to chemically dope ceria with other
transition or rare-earth element 2. Reduction of Ce4+
to Ce3+
in the ceria lattice should produce similar benefit.
However, the success of many promising technologies is entirely dependent on the development of powder synthesis techniques. The
great variety of methods for nanosized powder synthesis had been published in the literature. The methods include hydrothermal
synthesis, sol–gel, homogeneous precipitation, different combustion or flame-synthesis, self propagating room temperature reaction
(SPRT), salt assisted aerosol decomposition, sonochemical and microwave heating, microemulsion method, gas condensation, and so
on. The challenge for these new synthesis techniques is to preserve high powder activity while attaining the desired complex
composition. The produced powders must be clean, nanosized, with precise stoichiometry and with homogeneously distributed dopant
cations throughout the batch. The applied method should give reproducible powder properties, high yield and must be time and energy
effective. Therefore, studies on low-cost and time effective method is one main emphasis of ongoing research efforts.
In this paper we applied self-propagating room temperatures synthesis to produce single phase and highly pure nanosized powders
with precise stoichiometry in order to examine properties and behavior of nanocrystalline ceria during annealing and thus to get
indications on its potentional for diverse application.
42
2. Synthesis
2.1 Self-propagating room temperature process
Self-propagating room temperature synthesis (SPRT) is one of the methods, that is cost- and time effective. It is based on high
exothermicity of the reaction between metal nitrites and sodium hydroxide. Many reactions of this type need to be heated to elevated
temperatures to start propagating. From that point on, the reactions are developing by themselves, and come to an end extremely fast.
Synthesis of pure ceria was performed by self-propagating method at room temperature. i.e. there is no need to heat up the reacting
mixture. It was however, shown that ceria solid solutions can be obtained by this method, as well. By self-propagating reaction at
room temperature solid solutions of ceria containing two or tree valence cations, as well as, co-doped with both cations and even
multiply doped ceria powders were prepared. SPRT procedure as mentioned above, is based on the self-propagating room temperature
reaction between metal nitrates and sodium hydroxide, and in the case of doped ceria the reaction can be written as:
2 1 − 𝑥 𝐶𝑒 𝑁𝑂3 3 ⋅ 6𝐻2𝑂 + 𝑥𝑀𝑒 𝑁𝑂3 3 ⋅ 6𝐻2𝑂 + 6𝑁𝑎𝑂𝐻 + 1 2 − 𝑦 𝑂2 → 2𝐶𝑒1−𝑥𝑀𝑒𝑥𝑂2−𝑦 + 6𝑁𝑎𝑁𝑂3 + 15𝐻2𝑂
(1)
Practically the experiments can be performed with cerium nitrate and sodium hydroxide as reactants which were homogenized by hand
mixing By introducing mechanical energy into the system, instead of the thermal one, the reaction proceeds at room temperature in air
and could be observed by bare eyes. The first step to observe is the release of crystalline water from nitrates (of loosely bound two
molecules) at about 50οC. The release of water makes the hand homogenization of reactants easier. With further mixing, the rest of
crystalline water is released. Exposition to air of the homogenized mixture is necessary to bring the reaction (1) to the end. These
results indicate that the reaction does not seem to need much energy to start propagating (Fig.1(a)).
0 100 200 300 400 500 600
-0.30
-0.25
-0.20
-0.15
-0.10
-0.05
0.00
-0.2
-0.1
0.0
0.1
0.2
0.3
0.4
0.551oC
170oC
T
(°C)
d(
m/m
)/d
T
T (°C)
(a) (b)
20 30 40 50 60 70 80
Inte
nsity (
a.u
.)
2 Degree
11
1
20
0
22
0
31
12
22
40
0
33
1
2
3
4
1
42
0
20 30 40 50 60 70
4%
3%
2%
Inte
nsity (
a.u
.)
2Degree
1%
11
1
20
0 22
0
31
1
40
0
33
1
Ag
2O
(2
00
)
(c) (d)
43
10 20 30 40 50 60 70 80
Inte
nsi
ty (
a.u
.)
2Degree
3%
6%
9%
12%
15%
20 40 60
40
0
22
2
20
0
Ce0,9Bi0,1O2-
Ce0,5Bi0,5O2-
Ce0,6Bi0,4O2-
Ce0,7Bi0,3O2-
Ce0,8Bi0,2O2-
CeO2-
31
1
22
0
2θ Degree
Inte
zite
t (r
.j.)
11
1
(e) (f)
Fig. 1: (a) DTA and DTG patterns of Ce(NO3)3.6H2O-NaOH mixture; (b) FESEM image of as-synthesized ceria powder at room
temperature; (c) Room temperature XRD patterns of ceria after the heat treatment at different temperature for 15 min (1: 25C, 2:
400C, 3: 600C, 4: 800C); (d) Room temperature XRD patterns of Ag doped ceria synthesized by SPRT; (e) Room temperature
XRD patterns of Sr doped ceria, Ce1-xSrxO2- synthesized by SPRT; (f) X-ray patterns of Ce1-xBixO2- solid solutions for 0.10 ≤ x ≤
0.50.
The typical morphology of obtained solid solution Fig.1(b), depict very small isometric particles. It can be roughly estimated that the
size of individual particles about is about 5 nm. X-ray diffraction patterns for the CeO2 at different annealing temperatures are shown
in Fig. 1(c). XRD analysis reveals that all peaks for each starting sample were significantly broadened indicating small crystallite size
and/or strain. It is especially true for powder that is not annealed. It exhibits very diffuse diffraction lines, however, annealed powders
depict sharpened diffraction lines resulting from increased crystallite size 2.
Powder annealed at 800 C is, as obvious, very well crystallized. Monovalent cations are not usually studied as dpants for ceria
because of the large difference in their valence state. For exmple silver forms a stable solid solution with ceria in concentration 1-2
atom%. However, applying SPRT method, a higher solubility is obtained which could be attributed to the nanometric nature of
obtained powders due to used manufacturing method. The ceria peak intensity and the shape of (111) plane reflection in the Ag doped
CeO2 samples decreased and broadened, respectively, with increasing Ag doping (Fig.1(d)). In case of the sample doped with 4 wt%
of Ag ion, intensity of peaks increased and a new peak is observed indicating that the limit of Ag solubility in ceria crystal lattice is in
the range of 3 to 4 wt%. The peak at 2 37.8C can be attributed to Ag2O phase, (JSPDS-ICDD 41-1104). As a divalent doping
cation model, a Sr2+
is taken as an example. Since Sr2+
as dopant ion, with biger cation size compared to cerium ((Ce4+
= 0.097; Sr2+
=
0.126 nm), the solubility in ceria fluorite structure is not very high. XRD spectra of Ce1−xSrxO2−δ samples, presented in Fig.1(e),
revealed that obtained powders are single phase with the fluorite crystal structure. Main diffraction peaks in each sample were
broadened indicating small crystallite size and/or strain 3. No new peaks are observed for the sample doped untill 15 wt.% of Sr
(Fig.1(e)). As a trivalent cation Bi3+
was used. Single phase Ce1-xBixO2- solid solutions with the x=0.1-0.5 were successfully
synthesized, as well. It can be concluded from Figs.8 and 9 that by applying SPRT procedure monophase solid samples of Ce1-xBixO2-
with maximum Bi concentration of 50 at % can be prepared. Very wide reflections with low intensity indicate low crystallinity as
well as, small crystallite size.
Conclusion
A number of solid solutions of ceria doped with diferen tvalent cations with pricese stochiometry (Ce1-xAgxO2-; Ce1-xSrxO2- and Ce1-
xBixO2- ) was obtained by SPRT method. It was found that the particle size lies in the nanometric range, less than 4 nm.
Acknowledgement
Financial support from the Serbian Education and Science Ministry in the Framework of project No. 45012 is gratefully
acknowledged.
44
References
[1] Boskovic, S, Matovic, B, (2019) Nanostructured Solid Solutions of the Fluorite Type Crystal Structure, ed: van Aten M. Fluorite
Structure, Chemistry and Applications, Nova Science Publishers, Inc. New York, 1-111.
[2] Matovic, B., Nikolic, D., Labus, N., Ilic,S., Maksimovic, V., Lukovic, J., Bucevac, D., Preparation and properties of porous,
biomorphic, ceria ceramics for immobilization of Sr isotopes, Ceramics International (2013) 9645-9649.
[3] Matović, B., Dukic, J., Babić, B., Bučevac, D.,Docevic-Mitrovic, Z., Radovic, M.,, Bosković, S., Synthesis, calcination and
characterization of Nanosized ceria powders by self-propagating room temperature method, Ceramics International 39 (2013) 5007-
5012.
45
Platinum and Rhenium Modified Sulfated Zirconia in n-Alkanes Conversion: The
Promising Petroleum Industry Process
Aleksandra Zarubica1, Jelisaveta Hajdukovic
1, Aleksandra Krstic
1, Marjan Randjelovic
1,
Branko Matovic2
1Faculty of Science and Mathematics, Department of Chemistry, University of Nis,
18000 Nis, Serbia
E-mail: [email protected]
2 Institute of Nuclear Science Vinca, Materials Science Laboratory, University of Belgrade,
11000 Belgrade, Serbia
E-mail: [email protected]
Abstract
The aim of the investigation was the modification of sulfated zirconia (SZ)-based catalyst by incorporation of one and/or two metals
(platinum and rhenium) into catalysts with the purpose to correlate their physicochemical and resulted catalytic performances of
modified and unmodified SZ-based catalysts. The modification of bare sulfated zirconia by metallic components-functions increased
the primary isomerization efficiency (activity, selectivity and stability) of the SZ-based catalyst. This was affected by the
improvement of catayst textural properties (specific surface area, pore volume and pore size distribution), stabilized structural
properties (crystallite size below critical value and dominant active tetragonal crystal phase), and surface properties (higher total
acidity and presence of two type of active acidic sites). The conversion of n-hexane increased in the following order of tested
catalysts: SZ < Pt-Re/SZ < Pt/SZ. Among these catalysts, mono- and bimetallic promoted/modified SZ-based catalysts underwent a
bifunctional reaction route (over metallic and acidic active sites), while unmodified sulfated zirconia took part in the oxidative
dehydrogenation reaction during isomerization of n-hexane and n-pentane exhibiting only the initial activity.
Keywords – n-alkane conversion, modification, petroleum industry, platinum, process, rhenium, sulphated zirconia
1. Introduction
The modern world-wide environmental requests on the so-called ―clean and/or reformulated fuels‖ characterized with high octane
number (ON) have forced on petroleum industry to produce new (re)formulated fuels with the appropriate environmental friendly
boosters as constituents. Namely, most of conventionally used ON boosters (i.e. leaded gasoline, MTBE, benzene, and aromatic
compounds) are hazardous substances.
The selective isomerization of middle-chain n-alkanes (C5-C7) as simple and cost effective petroleum industry process presents a
promising solution in producing branched hydrocarbons as boosters for the increase of the ON of the gasoline pool. Specifically,
branched alkanes have higher ON than their linear counterparts (n-C6 and n-C7 are characterized with RON 25 and 0, respectively,
while the isomerization process would supply converted hydrocarbons with increased RON to 74 and 45, correspondingly) [1, 2].
Conventional isomerization process was performed using liquid catalysts in homogeneous petroleum industry process (sulfuric acid or
hydrogen chloride or fluoride). However, these liquid acid catalysts have exhibited severe corrosion problems in industrial plants and
health risks. Unconventional petroleum technologies introduced heterogeneous catalysts in the process of hydrocarbons conversion
over Pt/chlorinated alumina and/or Pt/zeolites [2, 3]. Unfortunately, the usage of the first catalytic system has caused environmental
problems due to application of hazardous promoter (chloride), and additionally caused economical drawbacks since being highly
sensitive to so-called catalyst poisons (i.e. coke accumulation). The application of the second catalytic system resulted with lower
catalyst efficiency and required reaction runs at higher reaction temperatures [2, 3].
Appropriately, the design of new effective and environmentally friendly catalyst with both adequate nano-structure and catalytically
active sites (acid and metal ones) is postulated as challenging task for the near future petroleum industry. Sulfated zirconia (SZ) as
solid acid has attracted significant attention due to its surface acidity and activity in the isomerization of middle-chain n-alkanes at
46
mild and economical process conditions [4]. Though, a fast deactivation of SZ and short working cycle create it unpractical. The
incorporation of transition metals as SZ catalyst promoters may be possible solution resulting in final catalysts characterized with
higher efficiency than unmodified SZ in n-butane isomerization [5-7].
Platinum supported SZ is previously reported to prevent fast deactivation of sulfated zirconia thus enhancing catalyst stability and
total efficiency [7]. Chlorinated alumina promoted with both platinum and rhenium is known as widely applied catalyst in the
petroleum industry for alkylation and reforming processes [8]. Despite the fact that the role of platinum in Pt-Re/Al2O3 catalytic
system is more or less recognized [8, 9], a role of rhenium is still a subject of investigation.
The surface acidity of SZ, the nature of its catalytically active sites and a formation of typical acidic active sites are subjects of
debates. To the best of authors' knowledge, n-hexane and n-pentane isomerization has not been investigated over SZ catalyst modified
with both rhenium and platinum, and modified by grafting rhenium into SZ-based catalyst. Moreover, the exact roles of platinum and
rhenium along with the unknown effect of particular metal (platinum or rhenium) incorporated in bimetallic and monometallic
catalysts have not been explained.
2. Experimental section
2.1. Catalysts preparation
Zirconia-based catalysts were synthetized by using somewhat modified sol-gel preparation method [10, 11] with use of 70% solution
of zirconium (IV)-isopropoxide in 2-propanol (Sigma Aldrich, Co.). The catalyst preparation over alchoxide hydrolysis and
condensation has been carried out at pH 9.0. Further sulfating procedure was performed by incipient wetness technique of
impregnation of the zirconium(IV)-hydroxide with 0.5 M sulfuric acid in order to achieve nominal sulfates content of 5 wt.%
followed with the activation at 550 °C in dynamic air conditions (synthetic air flow, 20 ml/min) for 4h. Aqueous solutions of hexa-
chloroplatinic acid or ammonium-perrhenate were applied as metal promoters‘ precursors. The nominal content of platinum was 0.5
wt. % in mono- and bi-metallic catalysts. Similarly, the nominal amount of rhenium in each catalyst – mono- and bimetallic catalyst
was also 0.5 wt. %. Supplementary catalyst activation was realized at 550 °C in dynamic air conditions (synthetic air flow, 20 ml/min)
for next 4h.
2.2. Characterization of catalysts
The textural properties (specific surface areas calculated by the (Brunauer-Emmett-Teller) BET method, the average pore diameter,
the pore size distribution) of the catalysts were resulted from N2 adsorption-desorption isotherms obtained by using a sorptometer
Micromeritics ASAP 2010. The average pore diameter was determined as the BJH desorption mean pore diameter [10, 12]. The pore
size distribution curves were additionally plotted by using the the Kelvin equation [13] and an adequate computer software.
X-ray diffraction (XRD) analysis was applied for the determination of crystal structure of the zirconium oxide-based catalysts and
realized by using X-ray diffractometer APD-1700. Crystallite size and volume fraction of particular crystal phase (monoclinic and/or
tetragonal) were calculated by using Scherer's equation.
Surface properties of the catalysts were studied by Fourier Transformance Infrared Spectroscopy (FTIR) analysis of the catalyst
surface pre-adsorbed with pyridine. The adsorption of pyridine on pre-evacuated catalysts was followed by additional final evacuation
in order to remove physisorbed pyridine preceded the FTIR measurements. The FTIR spectra of the catalysts were scanned using Win
Bomem Easy FTIR spectrophotometer in the wavelenghts range from 4000 to 400 cm-1
.
Transmission electron microscopy (TEM) characterization was performed using a TEM/EDX microscope CM20, Philips, Lab 6
operated at 200 KeV. Electron micrographs were recorded at magnification of 600.000x.
Electron paramagnetic resonance (EPR) spectra in X-band (9.5 GHz) were recorded with the cw-spectrometer ELEXSYS 500-
10/12 (Bruker Co.) using a microwave power of 6.3 mW, a modulation frequency of 100 kHz and a modulation amplitude of 0.5 mT.
2.3. Test reaction – n-alkanes conversion
Catalytic efficiency in the selected catalytic run–test reaction of n-hexane and/or n-pentane isomerization was studied using a fixed
bed quartz micro-reactor under atmospheric pressure and partial n-hexane pressure of 60.5 mbar. The working conditions were:
47
reaction temperature 250 °C at atmospheric pressure using hydrogen as carrier gas, and space velocity of 6 ·10-2
mmol n-C6/gCAT ·
min. The fresh catalyst loading was activated by a reducing procedure in hydrogen flow (20 ml/min) at 350 °C during 4 h preceding
the reaction.
The analysis of the reaction products was done by gas chromatograph (GC) using the HP Series 5890 II equipped with a flame
ionization detector (FID). Products of the isomerization were separated on PONA GC-capillary column (30 m). The mixture of
products consisted of hydrocarbons beginning with methane to benzene.
3. Results and discussion
The textural characteristics (BET surface area, average pore diameter and pore size distribution, BJH cumulative desorption pore
volume) of all catalysts (monometallic and bimetallic SZ - based samples) are shown in Table 1 and Fig. 1. These properties are in
accordance with the used synthesis method of the catalysts, type of the modification/promoting with metal, content of metal doped,
and an order of metal incorporation in bimetallic catalysts.
Table 1
The general textural characteristic of all the prepared and tested catalysts was a bimodal pore size distribution with a dominant
fraction of meso-pores (Fig. 1) that was expected bearing in mind the origin of the SZ-based catalyst and its preparation method
applied. Namely, the origin of these zirconia-based catalysts from an alkoxide precursor using the somewhat improved sol-gel
procedure determined textural features of the final catalysts over memory effects [14-16]. It is evident that modification of pure SZ
with Pt, and both Pt and Re, highly affects the BET surface areas by increasing them (Table 1). The incorporation of platinum
attributes to additional texture stabilization.
Fig. 1
The nitrogen isotherms for all catalysts exhibited typical s-shape behaviour of type-IV with a type-H1 desorption hysteresis (not
shown). The pore size distribution, preferably in the meso-pores range, is changed slightly and shifted towards smaller meso-pore
average diameter when platinum was incorporated in bare SZ (not shown). Bimodal pore size distribution and appropriate meso-pores
fractions are known as responsible for the significant positive effect on the primary features of SZ-based catalysts – activity,
selectivity, stability and catalyst-life time [11].
In the case of bimetallic SZ-based catalysts (incorporation of two metals, Pt and Re), the order of particular metal significantly
affected the final textural properties of the catalysts.
Bimetallic platinum-rhenium modified SZ-based catalyst (Pt-Re-SZ) showed improved textural properties characterized by favourable
characteristics when compared to the unmodified SZ catalyst (Table 1). This catalyst possessed almost the equal pore volume and
similar specific surface area as the monometallic one. The specific surface area increases in the following order: SZ < Pt-Re-SZ ~ Pt-
SZ. On the other side, in case when Re was the first incorporated in SZ followed by a step of activation, and then Pt, which preceded
the final catalyst activation, textural properties of such a catalyst (Re-Pt-SZ) were reduced by diminishing the specific surface area,
and causing the existing of larger pore diameter, Table 1). We suppose that Re in this case is captured in pores, filling the most of the
pore volume, and simultaneously blocking closer contact and interaction of Pt with SZ support. It seems that both metals (Pt and Re)
form an ―egg-white/egg-yolk‖ model structure with platinum being protected in deeper layers as it was recently reported for similar
catalytic systems [17-19]. In this model, we suppose that rhenium exhibits the so-called ―skin effect‖ partially covering the active
metal, platinum, and protecting platinum from potential catalytic poisons from a feed that may be observed during the catalytic run
[19].
Based on the results of the XRD analysis of M-modified SZ support and bare SZ catalyst, a comparison obviously indicates that
platinum affected the stabilization of the crystal structure of the primary SZ catalyst (Table 2). Namely, the incorporation of platinum
caused a significant increase of the fraction of the catalytically active tetragonal crystal phase as well as a decrease of its crystallite
size when compared to the unmodified SZ catalyst [20-22]. In the Pt-SZ catalyst an oxide species PtO is registered (2θ = 34.65°) (not
shown), therefore, the authors recommend the oxidation state +2 of Pt as the active state during the reaction run. In addition, Pt-Zr
interaction (2θ = 41.00°) is possible, which is probably responsible for the optimal active metal-support interaction, moreover,
affecting higher dispersion of metal promoter together with improved resistance to the sintering process [17,23].
Table 2
48
In bimetallic catalysts, the metal promoter impact on the fraction of the tetragonal active crystal phase is crucial one, and it is of
smaller effect on the crystallite size. The effect regarding the crystal phase composition and crystallite size (below critical value) is
completely comparable in Pt-Re-SZ as in the monometallic catalyst, Pt-SZ (Table 2). In the case of Pt-Re-SZ catalyst formated in the
previously mentioned ―egg-white/egg-yolk‖ model, a highly enhanced amount of ZrO2 (111) tetragonal crystal phase was observed.
The effect of rhenium on platinum dispersion may be an additional reason that the Pt-Re-SZ catalyst has the highest amount of the
tetragonal crystal phase and the smallest crystallite size among all catalysts (Table 2).
Surface properties of the catalysts, investigated by FTIR analysis (Fig. 2), demonstrated the dominant presence of BAS (at 1540 cm-1
),
and two types of LAS (at 1490 and 1610 cm-1
) in the monometallic Pt-SZ catalyst (the most active one). These facts are in line with
its high total acidity as a promising characteristic related to the superior efficiency in the isomerization of n-hexane [21, 24].
Fig. 2
2D micrographs of the selected locations on the Pt-SZ catalyst and crystallographic orientations of zirconium oxide were registered by
using the HRTEM (not shown). The dominant crystallographic orientation of zirconia is <010> characterized with a presence of so-
called ―short atomic steps‖ [23, 25]. Two or three concurrent particular orientations are presented at an atomic level. Such orientations
included the following ones: {101} – an orientation as the crystal lattice continuation in a regular direction; {100} – roughly shaped
plane, and finally {001} – a plane characterized with the most deformations that may include most Schottky vacancies [23, 25] and/or
so called ―stairs structure‖ at the edges.
Fig. 3
Crystallites with high Miller indices characterize the mentioned crystal planes potentially presenting very active adsorption sites [25,
26]. The presence of these sites in the Pt-SZ catalyst was also proved by the XRD analysis.
The HRTEM micrograph of Pt-Re-SZ catalyst (Fig. 3) presented that the modification of bare SZ with Pt and Re retains the primary
grain of ZrO2 in size as in the Pt-SZ catalyst, and the same predominant orientation of crystallographic planes, too. Pt grains in size
are almost comparable to the same ones in the monometallic catalyst Pt-SZ. We suggest that all these characteristics would result in a
synergistic effect on the final catalytic performances in the test reaction.
The EPR spectrum of the as-synthesized Pt-SZ catalyst and the pure SZ support are completely comparable (not shown). However, in
case when rhenium is incorporated, an additional multi-line spectrum was registered in the middle field range, which was weak in the
case of the catalyst Re-SZ and the most intense for the catalyst Pt-Re-SZ (not shown). The observed EPR-active valence states of
rhenium in our catalysts are Re6+
(S = ½) and Re4+
(S = 3/2) (Table 3). There are only very few literature data on spin Hamiltonian
parameters of paramagnetic Re species that differ extremely depending on the matrix (Table 3) [27, 28].
Table 3
Subsequently, the registered additional signals in the Pt-Re-SZ catalyst should result from Re6+
the coordination of which is slightly
modified, probably in the vicinity of Pt. This suggests that already in the as-synthesized state, Re and Pt are in close contact.
It is widely known that the Pt-Re-oxide support catalyst is a bifunctional one where noble metal acts as a centre for chemisorption and
activation of reactant transformation, and the support additionally supports the spill-over mechanism. Specifically, platinum as the
active noble metal may improve this mechanism in hydrogenation/dehydrogenation of alkenes/alkanes in the n-alkane isomerisation
reaction course [29, 30].
All the discussed properties (textural, structural, surface, morphological) have an impact on the catalyst efficiency in n-hexane and n-
pentane isomerization reaction. Pt-promoted SZ catalyst confirmed an initial activity at as low as 170 °C, and reached a significant
yield at 200/250 °C (Table 4). In the case of bimetallic catalysts (Pt-Re-SZ and Re-Pt-SZ), better catalytic performances were
achieved after reduction at temperatures higher than 350 °C. The catalyst Re-Pt-SZ with an opposite order of metal loading showed
somewhat lower activity/selectivity probably due to poorer physicochemical properties determined by the order of particular metal
incorporation, and activation steps.
Table 4
49
It is important to underline that the increased reduction temperature applied over Pt-Re-SZ catalyst contributed to the catalytic
characteristics almost half as much of those achieved for the Pt-SZ catalyst under the same operating working conditions.
4. Conclusions
Based on the obtained results of this investigation, it can be concluded that the modification of pure sulfated zirconia with platinum
significantly improves its physico-chemical properties and consequently its catalytic performances. Bimetallic promotion of pure
sulfated zirconia by both platinum and rhenium assists the attainment of almost half of the conversion obtained over platinum
promoted-modified sulfated zirconia under the same reaction conditions. Mono- and bimetallic promoted/modified sulfated zirconia
with selected noble metals, platinum and rhenium, undergo a bifunctional reaction course over metallic and acidic active sites, while
unmodified bare sulfated zirconia catalyzes conversion of reactants in the oxidative dehydrogenation reaction exhibiting only short-
time initial activity.
Acknowledgment
The authors wish to thank to the Ministry of Education, Science and Technological Development of the Republic of Serbia (Project III
45012), Serbian Academy of Sciences and Arts, and DAAD fellowship (A/10/05029; Section: 324) for financial supports.
References
[1] Jiménez, C., Romero, F.J., Roldán R. et al. (2003), Hydroisomerization of a hydrocarbon feed containing n-hexane, n-heptane and
cyclohexane on zeolite-supported platinum catalysts, Appl. Catal. A: Gen. 249, 1, 175-185,
https://doi.org/10.1016/S0926-860X(03)00177-7.
[2] Weyda, H., Köhler, E. (2003), Modern refining concepts-an update on naphtha-isomerization to modern gasoline manufacture,
Catal. Today 81, 1, 51-55,
https://doi.org/10.1016/S0920-5861(03)00101-9.
[3] Corma, A. (1995), Inorganic Solid Acids and Their Use in Acid-Catalyzed Hydrocarbon Reactions, Chem. Rev. 95, 3, 559-614,
https://doi.org/10.1021/cr00035a006.
[4] Yadav, G. D., Nair, J. J. (1991), Sulfated Zirconia and Its Modified Versions as Promising Catalysts for Industrial Processes,
Micropor. Mesopor. Mater. 33, 1, 1-48,
https://doi.org/10.1016/S1387-1811(99)00147-X.
[5] Yu, G.X., Zhou, X.L., Tang C. et al. (2008), A comparative study of the synthesis approaches and catalytic behaviors of Pt/SO42-
/ZrO2-Al2O3 catalysts for n-hexane hydroisomerization, Catal. Commun. 9, 8, 1770-1774,
https://doi.org/10.1016/j.catcom.2008.02.007.
[6] Garcia, E.A., Rueda, E.H., Rouco, A.J. (2001), Sulfated zirconia catalysts promoted with Fe and Mn: Mn effect in the Fe
dispersion, Appl. Catal. A: Gen. 210, 1-2, 363-370,
https://doi.org/10.1016/S0926-860X(00)00827-9.
[7] Kimura, T. (2003), Development of Pt/ SO42-
/ZrO2 catalyst for isomerization of light naphtha, Catal. Today 81, 1, 57-63,
https://doi.org/10.1016/S0920-5861(03)00102-0.
[8] Gurvey, Y.V., Ivanova, I.I., Lunin V.V. et al. (2007), Characterization of metal segregation in Pt–Re/Al2O3 reforming catalysts,
Appl. Catal. A: Gen. 329, 16-21, https://doi.org/10.1016/j.apcata.2007.06.015.
[9] Pieck, C.L., Gonzales, M.B., Parera, J.M. (2001), Total metallic dispersion of sulfided Pt-Re/Al2O3 naphtha reforming catalysts,
Appl. Catal. A: Gen. 205, 1-2, 305-312,
https://doi.org/10.1016/S0926-860X(00)00675-X.
[10] Breitkopf, C., Garsuch, A., Papp, H. (2005), Structure- activity relationships for sulfated zirconias- comparison of mesoporous
samples based on organic precursors, Appl. Catal. A: Gen. 296, 2, 148-156,
https://doi.org/10.1016/j.apcata.2005.07.058.
50
[11] Zarubica, A.R., Miljkovic, M.N., Kiss E.E. et al. (2007), Benefits of mesopores in sulfated zirconia catalyst, React. Kinet. Catal.
Lett. 90, 1, 145-150,
https://doi.org/10.1007/s11144-007-5051-2.
[12] Barrett, E.P., Joyner, L.G., Halenda, P.P. (1951), The Determination of Pore Volume and Area Distributions in Porous
Substances. I. Computations from Nitrogen Isotherms, J. Am. Chem. Soc. 73, 1, 373-380,
https://doi.org/10.1021/ja01145a126.
[13] Kelvin, L. (1871), Hydrokinetic solutions and observations, Phil. Mag. 42, 4, 362-377,
https://doi.org/10.1080/14786447108640585.
[14] Boskovic, G., Zarubica, A., Putanov, P. (2007), Precursor affected properties of nanostructured sulfated zirconia: morphological,
textural and structural correlations, J. Optoelect. Adv. Mater. 9, 7, 2251-2257.
[15] Boskovic, G., Zarubica, A., Kovacevic M. et al. (2008), Precursor memory effect determining structural properties of sulfated
zirconia, J. Therm. Anal. Cal. 91, 3, 849-854, https://doi.org/10.1007/s10973-007-8343-x.
[16] Zarubica, A., Putanov, P., Boskovic, G. (2007), Content of sulfates and their stability: Key factors determining the catalytic
activity of sulfated zirconia catalysts, J. Serb. Chem. Soc. 72, 7, 679-686,
https://doi.org/10.2298/JSC0707679Z.
[17] Zarubica, A., The influence of metal promoters from 4th
and 5th
d-transition series elements on catalytic properties of sulphated
zirconia in reaction of straight hydrocarbons isomerization, Ph.D Thesis, (2008), Univ. Novi Sad, Novi Sad.
[18] Zarubica, A., Putanov, P., Kostic D. et al. (2010), An impact of Re on Pt-Re/SO4 - ZrO2 catalyst for n-hexane isomerization, J.
Optoelectr. Adv. Mater. 12, 7, 1573-1576.
[19] Bošković, G., Vlajnić, G., Kiš E. et al. (1994), Geometric Factors in K and Al Promoting of the Fe/MgO Fischer-Tropsch
Catalyst, Ind. Eng. Chem. Res. 33, 9, 2090-2095. https://doi.org/10.1021/ie00033a010.
[20] Fottinger, K., Kinger, G., Vinek, H. (2004), In situ IR investigation of n-hexane isomerization over Pt containing sulfated
zirconia, Appl. Catal. A: Gen. 266, 2 195-202, https://doi.org/10.1016/j.apcata.2004.02.020.
[21] Blekkan, E.A., Johnesen, K.A., Loften, T. (2005), Isomerization of light alkanes: preparation and characterization of platinum
promoted sulfated zirconia catalysts, React. Kinet. Catal. Lett. 86, 1, 149-155,
http://doi.org/10.1007/s11144-005-0306-2.
[22] Zarubica, A., Boskovic, G., Putanov P. et al. (2010), A comparative study of physico-chemical and catalytic characterization of
M-modified SZ catalysts (M=Pt, Nb or Re) in n-hexane isomerization, J. Optoelect. Adv. Mater. 12, 5, 1183-1188.
[23] Zarubica, A., Randjelovic, M., Momcilovic, M., Radulovic, N., Putanov P. (2013), n-hydrocarbons conversions over metal-
modified solid acid catalysts, Russian Journal of Physical Chemistry A, 87, 13, 2166-2175,
http://doi.org/10.1134/S0036024413130281.
[24] Zarubica, A., Randjelovic, M., Momcilovic, M., Putanov P. (2012), 11th
International Conference on Fundamental and Applied
Aspects of Physical Chemistry, Proceedings, Volume 1, Belgrade, 139.
[25] Benaissa, M., Santiesteban, J.G., Diaz G. et al. (1996), Interaction of Sulfate Groups with the Surface of Zirconia: An HRTEM
Characterization Study, J. Catal. 161, 2, 694-703, https://doi.org/10.1006/jcat.1996.0231.
[26] Somorjai, G.A. (1981), Chemistry in Two Dimensions: Surfaces, 1st ed., Cornell Univ. Press, Cornell, New York.
[27] Pilbrow, J.R. (1990), Transition Ion Electron Paramagnetic Resonance, Clarendon Press, Oxford.
[28] Yao, H.C., Shelef, M. (1976), Surface interactions in the system Re-Al2O3, J. Catal. 44, 3, 392-403,
https://doi.org/10.1016/0021-9517(76)90416-4.
[29] Azzam, K.G., Babich, I.V., Sechan K. et al. (2008), Role of Re in Pt–Re/TiO2 catalyst for water gas shift reaction: A mechanistic
and kinetic study, Appl. Catal. B: Environ. 80, 1-2, 129-140,
https://doi.org/10.1016/j.apcatb.2007.11.015.
[30] Olympiou, G., Kalamaras, C., Zeinnalipour-Yazdi C. et al. (2007), Mechanistic aspects of the water–gas shift reaction on
alumina-supported noble metal catalysts: In situ DRIFTS and SSITKA-mass spectrometry studies, Catal. Today 127, 1-4, 304-318,
https://doi.org/10.1016/j.cattod.2007.05.002.
51
Tables
Catalyst SBET,
m2/g
d, nm Vp, cm3/g
SZ 73 7.2 0.15
Pt-SZ 116 4.6 0.14
Re-SZ 65 8.3 0.15
Pt-Re-SZ 101 5.4 0.14
Re-Pt-SZ 82 7.5 0.17
Table 1: BET specific surface areas, average pore diameters, BJH pore volumes
Catalyst Phase composition X, % d, nm
SZ T/M 86.7 / 13.3 11.7 / 12.1
Pt-SZ T/M 93.9 / 6.1 8.2 / 16.4
Re-SZ T/M 60.8 / 39.2 9.2 / 11.7
Pt-Re-SZ T/M 94.8 / 5.2 8.2 / 10.9
Re-Pt-SZ T/M 78.2 / 21.8 12.7 / 23.4
Table 2: Crystal phase composition, volume fraction of T/M crystal phases, crystallite size of zirconia
Matrix G A / G
Re4+ MoO3 1.617, 1,606, 1,667 330, 687, 370
Re4+ SnO2 1.626, 1.593, 1.81 682, 461, 399
Re4+ (NH4)PtCl6 1.817, 1.815 399, 395
Re4+ Al2O3 2.25 780
Re6+ CaWO4 ||1.85, 1.716 ||42, 324
Table 3: g and A tensor values of Re6+
and Re4+
Catalyst T X (%) S to
i-C6 (%)
Y (%)
SZ 250 14.0 33.9/40.9 /
Pt-SZ 175/250 7.2/74.1 95.1/34.8 6.8/25.8
Re-SZ 250 6.6 96.0 6.4
52
Pt-Re-SZ 250 34.3 94.4 32.3
Re-Pt-SZ 250 29.4 95.5 28.0
Table 4: Catalytic performances of monometallic and bimetallic SZ-based catalysts
Figures
a)
b)
Fig. 1: Curves of pore size distributions for catalysts: a) SZ, b) Re-Pt-SZ
a)
b)
Fig. 2: FTIR spectra of selected catalysts: a) Pt-SZ after pre-adsorption of pyridine, b) Pt-Re-SZ after pre-adsorption of pyridine
53
Fig. 3: HRTEM micrograph of the catalyst Pt-Re-SZ [17, 23]
54
Mathematical method using in robot laser hardening
Matej Babič
Faculty of Information Studies, Novo mesto, Slovenia, [email protected]
Abstract
This paper describes the roughness of surface of the robotically laser hardening microstructures with different parameters. We
hardened patterns pattern with different speed and different temperature. At the end it is presented the geometry of surface of the
robotic laser hardening. Characterization of surface topography is important in applications involving friction, lubrication, and wear.
In general, it has been found that friction increases with average roughness. Roughness parameters are, therefore, important in
applications such as automobile brake linings, floor surfaces, and tires. The effect of roughness on lubrication has also been studied to
determine its impact on issues regarding lubrication of sliding surfaces, compliant surfaces, and roller bearing fatigue. Finally, some
researchers have found a correlation between initial roughness of sliding surfaces and their wear rate. Such correlations have been
used to predict failure time of contact surfaces.
1. Introduction
The study of laser technology involves the study of principles of light energy in physics. Laser is an acronym for Light Amplification
by Stimulated Emission of Radiation. Laser hardening [1-3] is a metal surface treatment process complementary to conventional
flame and induction hardening processes. A high-power laser beam is used to heat a metal surface rapidly and selectively to produce
hardened case depths of up to 1,5mm with the hardness values of up to 65 HRc. It is used exclusively on ferrous materials suitable for
hardening including, steels and cast iron with a carbon content of more than 0.2 percent. Laser beam hardening is employed to locally
improve the surface properties of components. Use of this treatment can increase wear and fatigue resistance in parts of steel and cast
iron. Through a locally restricted heat treatment arises a minimum heat input, thereby minimised distortion. The associated high
heating and cooling rates result in fine microstructures with good mechanical properties. During the laser skin hardening, the material
(carbonaceous material) is heated up for a short time above austenitizing temperature and is transubstantiated by fast cooling down
into the martensitic structure. Different tool steels are widely used in industrial applications due to good performance, a wide range of
mechanical properties, machinability and wear cheapness. With the laser remelting surface of the material, we can significantly
improve their wear properties. Heat is generated by absorbing the laser radiation on the surface and the material is quenched by heat
transportation inside. The surface may not melt up. Robot laser hardening have many advantages:
laser is source of energy with outstanding characteristics (contactless methos, controlled input of energy, high capacity, constant
process, precise positioning)
• lower costs for additional machining
• no use of cooling agents or chemicals
• high flexibility
• the process can be automated and integrated in the production process
• superior wear resistance of hardened surface
• selective hardening of complex geometrical shapes
2. Materials preparation and method
55
We made samples of a standard label on the materials according to DIN standard 1.7225. The specimen test section had a cylindrical
form dimension 25×10mm (Fig 1). Tool steel was forged with the laser at different speeds and different power. So we changed two
parameters speed v ∈ [2, 5] mm / s with steps of 1 mm / s and temperature T ∈ [1000, 1400] °C to 50 °C steps. After hardening were
we cutting the specimen test into smaller parts (Step 4).
Fig. 1: Material preparation step 1 and step 2
Fig. 2: Material preparation step 3 and step 4
In all these attempts we have made recordings microstructure.Microstructure of specimens was observed with an field emission
scanning electron microscope JSM-7600F JEOL company. Irregular surface texture with a few breaks which are represented by black
islands (Fig. 2). We find relationship between parameters of Temperature and porosity. Our study was limited on tool steel standard
label of DIN standard GGG 60 (Fig. 1). Chemical composition of the material contained 0.38 to 0.45 C%, max 0.4% Si, 0.6-0.9 Mn%
P% max 0.025 max 0.035 S% and 0.15-0.3% Mo.
To investigate the possibility of application of fractal analysis to heat-treated surface, we have examined the relation between surface
roughness, hardness and fractal dimensions depending on various parameters of robot laser cell. In fractal geometry is the key
parameter fractal dimension [4], D, which should be determined complexity microstructure of robot laser hardened specimens. The
relation among fractal dimension D, volume V and length L can be indicated as follows with a concept similar to Eq. (1).
V~LD (1)
56
Fractal dimension were determined using the box couting method.
Fig. 3: Calculation fractal dimension of Image
3. Results and discussion
Figure 4-11 present roughness robot laser hardened materials with different parameter of robot laser cell.
57
In table 1 is presented parameters of hardened specimens wich impact on hardness. We mark specimens with P1 to P11. Parameter X1
present parameter of temperature [°C], X2 present speed of hardening [mm/s], X3 present fractal dimension and X4 present base
hardness (hardness before hardending), x5 is measured hardness of robot laser hardened specimens and x6 is measured roughness of
robot laser hardened specimens.
Specimen x1 x2 x3 x4 X5 X6
P1 1000 2 1,9135 34,5 60 201
P2 1000 3 1,9595 33,5 58,7 171
P3 1000 4 1,9474 34,5 56 109
P4 1000 5 1,9384 33,5 55,5 76
P5 1400 2 1,9225 34,5 58 1320
P6 1400 3 1,9784 33,5 57,8 992
P7 1400 4 1,954 34 58,1 553
P8 1400 5 1,9776 34,5 58,1 652
Table 1: parameter of robot laser hardened specimens
4. Conclussion
We observed microstructures of robot laser hardened patterns, where we discovered roughness surface. We interested for investigate
roughness and hardness surface of robot laser hardened patterns with different parameters. We will know how parameters of robot
laser hardened cell impact on roughness and hardness of hardened surface. In the future we will discover how parameters of robot
laser hardened cell impact on roughness and hardness of hardened surface in two-beams laser hardening.
58
References
[1] Akinay, Y., Hayat, F. The influence of the heat treatment on mechanical and microstructure properties of Fe-Mn-C high-
manganese steel. Metalic Materials. vol. 54 (2016), no. 2, pp. 91 – 96.
[2] Amini, K., Akhibarizadeh, A., Javadpour, S. Investigating the effect of the deep cryogenic heat treatment on the corrosion
behavior of the 1.2080 tool steel. Metalic Materials. vol. 54 (2016), no. 5, pp. 331 – 338.
[3] Barhoumi, H., Souissi, S., ben Amar, M., Elhalouani, F. Investigation of the microstructure and mechanical properties of squeeze
cast Al-11%Si alloy heat treated. Metalic Materials. vol. 54 (2016), no. 4, pp. 249 – 256.
[4] Yu, B. M., Cheng, P. A fractal permeability model for bi-dispersed porous media, International journal of
Heat Mass Transfer 45 (2002) 2983–2993.
59
Fractographic Examination of the Multilayer Aluminum Composites
Ivana Cvijovic-Alagic
Institute of Nuclear Sciences ―Vinča‖, University of Belgrade, Mike Petrovića Alasa 12-14, 11000 Belgrade, Serbia, [email protected]
Abstract
New trends in the aeronautical and military applications are focused onthe processing of the structural components with
enhanced mechanical andtribo-corrosive characteristics through the easy-to-apply andcost-effective production route. In
that purpose, the Al2O3 particlereinforced aluminum matrix composites are produced during this studyusing the non-
expensive household aluminum foils. The aluminum matrixcomposites are manufactured by hot-rolling of the non-
anodized oralternately sandwiched non-anodized and anodized foils. Obtainedmultilayer composites are subjected to
microstructural examination andmechanical testing in order to determine the influence of thecomposite’s
composition on their fracture mechanisms. Detailmicrostructural analysis and fractographic examinations after tensileand
three-point-bending testing were undertaken using the light opticalmicroscopy (LOM), scanning electron microscopy
(SEM) and X-raydiffraction (XRD) analysis. Results of the qualitative fractographyrevealed that the ductile fracture
features prevail in the overallfracture mode of the multilayer composite consisted of only non-anodizedfoils, while the
fracture surface of the composite obtained by thehot-rolling of the alternate anodized and non-anodized foils
showedpredominant brittle fracture with the delamination features.Quantitative fractography allowed more detailed
insight into thecomposite failure process and depicted critical parameters which led tothe fracture.
60
Consideration of morphology, structure, and composition in electrodeposited powders of
Ni-Co system
Vesna M. Maksimović1, Nebojša D. Nikolić
2
1Department of Materials Science, CEXTREME LAB, Institute of Nuclear Sciences ―Vinča‖, University of Belgrade, Mike Petrovića
Alasa 12-14, 11000 Belgrade, Serbia 2ICTM-Department of Electrochemistry, University of Belgrade, Njegoševa 12, P. O. Box 473, 11001 Belgrade, Serbia,
Abstract
Electrochemically deposited alloys of iron-group metals, whether in a form of powders or coatings are very important magnetic
materials and also are known as good catalysts for hydrogen evolution. Electrodeposition of alloy is more complicated than for pure
metals, due to co-deposition of at least two metals and formation of various crystallographic structures arising from the phase
diagram. Based on the binary phase diagram Ni-Co, nickel and cobalt build series of solid solutions in the whole range of
concentrations. In the case of Ni-Co alloy powders, aside from co-deposition of these two metals, there is hydrogen evolution as a
parallel reaction in the whole range current densities and potentials. Nickel-cobalt (Ni-Co) alloy powders were produced by a
galvanostatic regime of electrolysis using sulphate electrolytes with various ratios of Ni2+
/Co2+
ions. The morphology, structure, and
chemical composition were examined by scanning electron microscope (SEM), X-ray diffraction (XRD) and atomic emission
spectrometry (AES), respectively. Magnetic properties of the produced Ni-Co alloy powders were also considered. The strong effect
of hydrogen evolution on the morphology of powder particles has been mentioned. The change of morphology of the particles from
cauliflower-like and dendritic to spongy-like ones was observed with the increasing Ni2+
/Co2+
ratio from 0.25 to 4.0. A novel type of
the particle, referred to as coral-like, was obtained with Ni2+
/Co2+
ratio of 1.5. These changes in morphologies of the particles with
increasing Ni2+
/Co2+
ratio are ascribed to intensification of hydrogen evolution reaction as a parallel reaction to this alloy powder
electrolysis. XRD analysis of the Ni-Co powders revealed that the decrease of Ni/Co ratios (the increase of Co content) caused a
change of structure from face-centered cubic (FCC) obtained for the ratios 4.0, 1.5 and 0.67 to a mixture of face-centered cubic (FCC)
and hexagonal closed-packed (HCP) phases for the ratio 0.25. The increasing content of nickel led to change of mechanism of
electrolysis from irregular (up to ≈ 40% wt. Ni in the electrolytes) to close to equilibrium (between ≈ 40 and 60% wt. Ni in the
electrolytes) and anomalous co-deposition (over 60% wt. Ni in the electrolytes) type. The change of mechanism of electrolysis was
correlated with morphology of the produced particles, indicating that formation of novel coral-like type of particles can be attributed
to close to equilibrium type. All of the obtained Ni-Co alloy samples behave as soft magnetic materials while their magnetic
parameters showed immediate composition dependence since both coercivity and saturation magnetization almost linearly increase
with the Co content.
Keywords: Ni-Co alloy powders; Morphology; Co-deposition; Hydrogen evolution
61
Accuracy analysis in advanced 3D photogrammetric reconstruction
Michele Calì 1,* and Rita Ambu
2
1 Department of Electric, Electronics and Computer Engineering, University of Catania, V.le A. Doria, 6, 95125 Catania, Italy 2 Department of Mechanical, Chemical and Materials Engineering, University of Cagliari, via Marengo 2, 09123 Cagliari, Italy; [email protected]
* Correspondence: [email protected]; Tel.: +39-095-738-2400
Abstract: The paper is a study of the photogrammetric reconstruction of large-scale objects using Unmanned Aerial Vehicle (UAV)
camera image acquisition aimed at developing a method which permits to enhance the accuracy of such reconstruction. The
proposed methodology was elaborated to optimize the UAV procedures so to assure the quality of the digital surface reconstruction.
With a high number of photos necessary in the three dimensional (3D) reconstruction of large-scale elements, accuracy is
increasingly influenced by the integration of images and the computational time of the algorithms. Differently from the problem of
the speed of reconstruction applications, the object of this study was the analysis of different acquisition grid shapes and the
acquisition grid geometric parameters (pitches, image rates, camera framing, flight heights) in terms of the improvement of results in
the 3D photogrammetric surface reconstruction accuracy. These relationships were studied specifically with reference to a case study
of the ancient arched brick Bridge of the Saracens in Adrano (Sicily, Italy). In particular, the study highlighted how much Ground
Sampling Distance (GSD), the necessary number of photos to assure the desired overlapping, and the surface reconstruction
accuracy were related to grid shapes, image rate, and camera framing at different flight heights (15 m, 20 m, 30 m, 40 m, and 50 m).
The reconstruction of the 3D surfaces of this structure, obtained by the efficient Structure-From-Motion (SfM) algorithms of the
commercial software Pix4Mapper, allowed to validate the methodology with experimental data. By comparing the surface
reconstruction with different acquisition grids at different flight heights and the data acquired by the 3D laser scanner and total
station-theodolites, accuracy was thus analyzed in terms of Euclidean distances.
Keywords: Accuracy; Digital Surfaces Models; Tolerances; Ground Sampling Distance; Structure-From-Motion.
1. Introduction
In the last years digital photogrammetry based on digital imagery and computer vision algorithms and computational techniques
has increased the use of UAV image acquisition thanks to enhanced performance [1–4] speeding up the processing time and the
quality of the surface reconstruction [5–7]. These techniques that have challenged traditional acquisition and processing methods have
been also used for shape detection [8, 9] and 3D surface reconstruction of large-scale elements, such as natural environments and
geographical configurations [10–12], buildings and urban textures [13–15], archaeological sites [16, 17], and industrial installations
[18,19]. The accuracy in the reconstruction of these 3D structures greatly depends on the integration of a higher number of photos
[20]. To solve the problem of the increased computational time of the algorithms due to the dimensions of the object to reconstruct,
researchers have proposed the use of improved algorithms for different situations based on early SfM algorithms [21] developing
incremental [22, 23], hierarchical [24], and global [25–27] approaches.
Rather than focusing on the speed of the implemented procedures and on the success in image orientation, to obtain precise
measurements of three-dimensional large-scale objects this study aimed at the improvement of results in the photogrammetric
reconstruction against different acquisition grid shapes and acquisition grid geometric parameters (pitches, image rates, camera
framing, flight heights). In particular, the aim of the research was to analyze the influence of acquisition grid shapes (rectangular,
elliptical, and cylindrical) and acquisition grid geometric parameters on the accuracy in the 3D reconstruction of large objects‘
surfaces. In the case study of the Bridge of the Saracens in Adrano (Sicily), a valuable example of Roman architecture, the obtained
results in terms of these relationships validated the proposed methodology with experimental data. Dense point clouds derived from
images obtained by a CMOS 12 MP sensor using a Pix4Dmapper version 3 commercial software. The degree of overlapping in the
image acquired was evaluated through Function Based Method (FBM) [27] and Area Based Matching (ABM) [28, 29] algorithms.
Keypoints were matched in a quick and accurate way using some cutting edge matching techniques with binary descriptors.
The evaluation of the quality of Digital Surfaces Models (DSM) produced by UAV image acquisitions was performed through
the comparison of the photogrammetric reconstructions to the data acquired by the 3D laser scanner and total station-theodolites.
62
2. Acquisition and reconstruction phases
In the case study of the bridge surface reconstruction, the aerial photogrammetric acquisition was performed using: - an amateur
UAV Hexacopter with Lipo 4S cells (4000 mha, 14.4 V, 576 Wh—over 20 min autonomy) (Figure 1a,b); - an Ardupilot APM 2.6
with Arducopter 3.1.5 flying software and a PC Mission Planner ground station as control board; - an action camera GoPro Hero 4
(Woodman Labs, San Francisco, CA, USA) Black Edition (Figure 1c) positioned beneath the drone on a ‗Gimbal‘ support with a
CMOS 12 MP 1/2.9″ solid-state sensor to sense electromagnetic data; - a video transmitter provided real time shots on a 7″ LCD
monitor incorporated into the radio control unit (Figure 1). The acquisition pitches p [m] in the grids forming the waypoints (Figure 2)
were calculated according to the following Equation (1) [30]:
p=((ImW_p×GSD)/100)×(1-overlap) (1)
where ImW_p was the image width [pixel]. The longitudinal pitch (pl) and (orthogonal to the first) a transversal pitch (pt) were
determined ensuring a constant overlap value of 66% (as shown in Figure 3 which highlights three areas, blue, orange and green,
corresponding to different acquired images and the zone of their overlap, which defines the length and width of overlapping). The
following acquisition pitch p values were so established:
p_l=((ImL_p×GSD)/100)×(1-overlap)= ((4000×GSD)/100)×33.3 (2)
p_t=((ImW_p×GSD)/100)×(1-overlap)= ((3000×GSD)/100)×33.3 (3)
where GSD [cm/pixel] was calculated according to the following Equation (4) as a function of h_v:
GSD= (h_v×S_w×100)/(F_l×ImW) (4)
In this equation, S_w was the sensor width and F_l was the focal length. The mesh data obtained by using Delaunay
triangulation derived from the 3D feature points calculated by the SfM so to create an outline of the object projected onto the plane of
the images.
(c) (e)
(a) (b) (d) (f)
Figure 1. (a) UAV platform; (b) batteries and camera; (c) gimbal and camera; (d) LCD screen on the radio control; (e) laser scanner
Konica Minolta 9v-I; (f) Geodimeter 480 total station-theodolites.
Thus the estimated depth maps were obtained, then optimized and corrected using the pixel matching algorithm based on the
patch. From the fusion of the depth maps, dense point cloud data were obtained so to get a 3D polygonal mesh (point 4 of previous
reconstruction steps).
The polygonal mesh can be easily transformed, through open source algorithms, into a nurbs surface for different applications.
The flowchart relative to the reconstruction algorithm is summarized in Figure 3. Within the blocks highlighted in yellow, the
equivalent open source algorithm is shown.
63
Figure 2. Waypoint and overlap in RGVC at ℎ𝑣 = 40 m.
Figure 3. Flowchart of the main steps of the surface reconstruction algorithm.
3. Reconstructions Accuracy Evaluation
The case study of the bridge has allowed an accurate validation of the method thanks to the favorable spatial distribution of two
reference shapes measured (the pounding upper surface (pus) and the south side of the north-east surveyed arch (arc) of the bridge).
Accuracy (acc) was measured by evaluating the distances between the mesh obtained with Pix4Dmapper and the point cloud
obtained with 3D laser scanning. Accuracy was evaluated by introducing the standard deviation (σ) of such differences for a typical
length of the shape. The surface deviation was estimated by the CloudCompare while the alignment of the partial scans with the 3D
model produced by Pix4Dmapper was obtained using Meshlab. For the alignment the ICP algorithm, implemented in Meshlab was
used. For the comparison of the two data types, the cloud-to-mesh distance function offered by CloudCompare was selected so to
compute the distances between each vertex of the point cloud to the nearest triangle of the mesh surface.
Accuracy was then measured in the 20 reconstructions by considering the values of the mean error distance (𝐴𝐵 𝑚𝑒𝑎𝑛 ; 𝑅𝑚𝑒𝑎𝑛 ) and
standard deviation distances (𝜎𝐴𝐵 ; 𝜎𝑅) expressed in cm in the 20 reconstructions. Figure 4 shows the mean distance [cm] and standard
deviation distances (𝜎𝐴𝐵 ; 𝜎𝑅) calculated in cm in the case of surface reconstruction with CG at ℎ𝑣 = 15 m.
64
(a) (b)
Figure 4. Error in mean distances [cm] and its standard deviation distances [cm] distribution graphs in CG at 15 m: (a) pounding upper
surface; (b) south side of the north-east surveyed arch.
We concluded that the most accurate reconstructions were those characterized by smaller values of mean error for distance and
smaller values of standard deviation error.
When considering a comparison of the 20 different types of surface reconstruction related to the number of photos used (n.
GRC) and the mean value of GSD offered by each of them, the following parameters were calculated ξ [cm2] by multiplying the
accuracy acc by GSD and the number of photos (n. GRC):
𝜉𝑝𝑢𝑠 = 𝑎𝑐𝑐𝑝𝑢𝑠 × 𝐺𝑆𝐷 × n. GRC (5)
𝜉𝑎𝑟𝑐 = 𝑎𝑐𝑐𝑎𝑟𝑐 × 𝐺𝑆𝐷 × n. GRC (6)
Thanks to the inferior values of these parameters, the accuracy and the speed of reconstruction were improved at the same time.
4. The Comparative analysis
The comparative analysis of the data obtained from the twenty reconstructions was a useful tool for the photogrammetric surface
reconstruction of large-scale objects. In all the reconstructions, the accuracy resulted in Gaussian-like distributions and was always
proportional to GSD. However, the number of the images necessary to obtain the desired accuracy varied according to the grid shapes
and the acquisition parameters utilized. The quality of the reconstruction was highly superior at all the flight heights by normalizing at
one the sum of the two factors (𝜉𝑝𝑢𝑠 + 𝜉𝑎𝑟𝑐 = 𝜉). On average, the CG grid shape improved the accuracy by more than a factor of
six/seven. The image overlaps were proportional to flight height and inversely proportional to grid pitch studied. The synergic effects
of grid shapes, grid pitch, and camera framing, instead, could not always be predicted and their right combination might provide an
advanced accuracy in photogrammetric surface reconstruction. The correlation between GSD, flight altitudes, and overlap with one
another was obtained by the equations from (1) to (6) used to locate the flight heights, which assured the desired values of GSD and
overlap.
Accuracy was not sufficiently reached by the acquisition with a single RGVC. Moving to the acquisition with RGVO, the
transversal overlap was kept steady (66%) when increasing the transversal pitch. This occurred according to Equation (1), at the
expense of GSD. Moreover, the longitudinal overlap increased by rotating the camera of the angle ϑr. Similar to image overlap, the
surface deviation values and the distances between feature points appeared to be inversely proportional to flight height and highly
dependent on grid acquisition type. Overlap in the acquisitions with constant grid pitch seemed to be extremely variable, especially in
the elliptical grid. In the acquisition around the bridge with elliptical grids, some efficiency in terms of number of images against the
surface area being covered was lost. The analysis permitted to check that all reconstructions contained errors proportional to flight
height, but this was true, especially for RGVC. This acquisition, being one of the most used types, provided the worst results both in
terms of accuracy (acc) and parameter ξ. The best results were obtained instead by the acquisition with CG (Figure 5). This kind of
65
grid, thanks to the modern technologies and to a good GPS system, was easily implementable with a high accuracy in a system of
acquisition by UAV.
Figure 5. 𝜉 factor normalized.
6. Conclusions
The study focused on the improvement of the quality of the 3D digital surface reconstruction of large-scale objects by the
photogrammetric technique using the proposed replicable and generalizable methodology. The analysis indicated the existing
relationships between the grid shapes, the acquisition grid parameters, the image overlap, and the accuracy of reconstruction by using
commercial software (Pix4Dmapper) based on the Structure-from-Motion algorithms. The proposed relationships allowed to
appropriately select flight heights, acquisition grid shape, and camera framing in correspondence to a pre-established overlap value
and required GSD.
From the experimental results it is clear that in large-scale objects characterized by an elongated longitudinal shape, geometric
singularities, and multiple and variously inclined features, the accuracy can be improved by this method increasing the speed of
reconstruction at the same time by more than a factor of six/seven. In further analyses the study could be applied to more complex
acquisition grid shapes (i.e., a zig-zag grid).
References
1. Chen, S.; Laefer, D.F.; Mangina, E. State of technology review of civilian UAVs. Recent Pat. Eng. 2016, 10, 160–174,
doi:10.2174/1872212110666160712230039.
2. Cano, E.; Horton, R.; Liljegren, C.; Bulanon, D.M. Comparison of small unmanned aerial vehicles performance using image processing. J. Imaging 2017, 3, 4,
doi:10.3390/jimaging3010004.
3. Colomina, I.; Molina, P. Unmanned aerial systems for photogrammetry and remote sensing: A review. ISPRS J. Photogramm. Remote Sens. 2014, 92, 79–97,
doi:10.1016/j.isprsjprs.2014.02.013.
4. Byrne, J.; Laefer, D.F.; O‘Keeffe, E. Maximizing feature detection in aerial unmanned aerial vehicle datasets. J. Appl. Remote Sens. 2017, 11, 025015,
doi:10.1117/1.JRS.11.02501.
5. Aguilar, W.G.; Angulo, C. Real-time video stabilization without phantom movements for micro aerial vehicles. EURASIP J. Image Video Process. 2014, 2014,
46.
6. Heng, L.; Honegger, D.; Lee, G.H.; Meier, L.; Tanskanen, P.; Fraundorfer, F.; Pollefeys, M. Autonomous visual mapping and exploration with a micro aerial
vehicle. J. Field Robot. 2014, 31, 654–675.
7. Byrne, J.; O‘Keeffe, E.; Lennon, D.; Laefer, D.F. 3D reconstructions using unstabilized video footage from an unmanned aerial vehicle. J. Imaging 2017, 3, 15,
doi:10.3390/jimaging3020015.
66
8. Gaszczak, A.; Brekon, T.P.; Han, J. Real-time people and vehicle detection from UAV imagery. In Intelligent Robots and Computer Vision XXVIII: Algorithms
and Techniques, Proceedings SPIE 7878, San Francisco Airport, CA, USA, 23–27 January 2011; Röning, J., Casasent, D.P., Hall, E.L., Eds.; SPIE: Bellingham,
WA, USA, 2011; Article 78780B, doi:10.1117/12.876663.
9. Karim, S.; Zhang, Y.; Laghari, A.A.; Asif, M.R. Image processing based proposed drone for detecting and controlling street crimes. In Proceedings of the 17th
IEEE International Conference on Communication Technology, Chengdu, China, 27–30 October 2017; pp. 1725–1730.
10. Mancini, F.; Dubbini, M.; Gattelli, M.; Stecchi, F.; Fabbri, S.; Gabbianelli, G. Using unmanned aerial vehicles (UAV) for high-resolution reconstruction of
topography: The structure from motion approach on coastal environments. Remote Sens. 2013, 5, 6880–6898, doi:10 3390/rs5126880.
11. Liénard, J.; Vogs, A.; Gatziolis, D.; Strigul, N. Embedded, real-time UAV control for improved, image-based 3D scene reconstruction. Measurement 2016, 81,
264–269, doi:10.1016/j.measurement.2015.12.014.
12. Liang, Z.; Guo, Q.; Xu, J.; Hu, J. A UAV Image and Geographic Data Integration Processing Method and Its Applications. In Proceedings of the 2017 2nd
International Conference on Communication and Information Systems, Wuhuan, China, 7–9 November 2017; Association for Computing Machinery: New
York, NY, USA, 2017; pp. 268–272.
13. Ham, Y.; Han, K.K.; Lin, J.J.; Golparvar-Fard, M. Visual monitoring of civil infrastructure systems via camera-equipped Unmanned Aerial Vehicles (UAVs): A
review of related works. Vis. Eng. 2016, 4, 1, doi:10.1186/s40327-015-0029-z.
14. Koutsoudis, A.; Vidmar, B.; Ioannakis, G.; Arnaoutoglou, F.; Pavlidis, G.; Chamzas, C. Multi-image 3D reconstruction data evaluation. J. Cult. Herit. 2014, 15,
73–79, doi:10.1016/j.culher.2012.12.003.
15. Zhang, W.; Li, M.; Guo, B.; Li, D.; Guo, G. Rapid texture optimization of three-dimensional urban model based on oblique images. Sensors 2017, 17, 911,
doi:10.3390/s17040911.
16. Campana, S. Drones in Archaeology. State-of-the-art and Future Perspectives. Archaeol. Prospect. 2017, 24, 275–296, doi:10.1002/arp.1569.
17. Sauerbier, M.; Eisenbeiss, H. UAVs for the documentation of archaeological excavations. Int. Arch. Photogramm. Remote Sens. Spat. Inf. Sci. 2010, XXXVIII,
526–531.
18. Li, Q.; Li, D.C.; Wu, Q.F.; Tang, L.W.; Huo, Y.; Zhang, Y.X.; Cheng, N. Autonomous navigation and environment modeling for MAVs in 3-D enclosed
industrial environments. Comput. Ind. 2013, 64, 1161–1177, doi:10.1016/j.compind.2013.06.010.
19. Rabbani, T. Automatic Reconstruction of Industrial Installations Using Point Clouds and Images; Publications on Geodesy, 62; NCG Nederlandse Commissie
voor Geodesie Netherlands Geodetic Commission: Delft, The Netherlands, 2006; pp. 1–175.
20. Wu, C. Towards linear-time incremental structure from motion. In Proceedings of the International Conference on 3D Vision-3DV2013, Seattle, WA, USA, 29
June–1 July 2013; pp. 127–134.
21. Beardsley, P.A.; Torr, P.H.S.; Zisserman, A. 3D model acquisition from extended image sequences. In Lecture Notes in Computer Science, Computer Vision—
ECCV96; Buxton, B., Cipolla, R., Ed.; Springer: Berlin, Germany, 1996; pp. 683–695.
22. Mohr, R.; Veillon, F.; Quan, L. Relative 3-D reconstruction using multiple uncalibrated images. In Proceedings of the IEEE Conference on Computer Vision and
Pattern Recognition, New York, NY, USA, 15–17 June 1993; pp. 543–548.
23. Dellaert, F.; Seitz, S.M.; Thorpe, C.E.; Thrun, S. Structure from motion without correspondence. In Proceedings of the IEEE Conference on Computer Vision
and Pattern Recognition-CVPR 2000, Hilton Head Island, SC, USA, 15 June 2000; pp. 557–564.
24. Gherardi, R.; Farenzena, M.; Fusiello, A. Improving the Efficiency of Hierarchical Structure-and-Motion. In Proceedings of the 2010 IEEE Computer Society
Conference on Computer Vision and Pattern Recognition, San Francisco, CA, USA, 13–18 June 2010; pp. 1594–1600.
25. Moulon, P.; Monasse, P.; Marlet, R. Global fusion of relative motions for robust, accurate and scalable structure. In Proceedings of the 2013 IEEE International
Conference on Computer Vision, Sydney, NSW, Australia, 1–8 December 2013; pp. 3248–3255.
26. Crandall, D.J.; Owens, A.; Snavely, N.; Huttenlocher, D.P. SfM with MRFs: Discrete-continuous optimization. IEEE Trans. Pattern Anal. Mach. Intell. 2013,
35, 2841–2853, doi:10.1109/TPAMI.2012.218.
27. Sweeney, C.; Sattler, T.; Höllerer, T.; Turk, M. Optimizing the viewing graph for structure-from-motion. In Proceedings of the 2015 IEEE International
Conference on Computer Vision (ICCV), Santiago, Chile, 7–13 Dec. 2015; pp. 801–809.
28. Cheng, G.; Han, J. A survey on object detection in optical remote sensing images. ISPRS J. Photogramm. Remote Sens. 2016, 117, 11–28.
29. Zhang, Q.; Li, Y.; Blum, R.S.; Xiang, P. Matching of images with projective distortion using transform invariant low-rank textures. J. Vis. Commun. Image
Represent. 2016, 38, 602–613.
30. Pix4d Manual. Available online: http://pix4d.com (accessed on 10 August 2018).
31. Calì, M., & Ambu, R. Advanced 3D Photogrammetric Surface Reconstruction of Extensive Objects by UAV Camera Image Acquisition. Sensors, 18(9), 2815,
2018.
67
Microstructures design of precursor derived ceramics towards structural and functional
applications
Ravi Kumar
Laboratory for High Performance Ceramics, Department of Metallurgical and Materials Engineering, Indian
Institute of Technology Madras, Chennai – 600 036 (India).
Polymer derived ceramics (PDC) as structural ceramics for high temperature applications have been intensely
studied in the recent few years. The transformation from a polymeric precursor to the corresponding ceramic is
achieved via. solid state thermolysis (SST). The as-thermolysed ceramic is often amorphous and is found to
possess exceptional thermo-mechanical properties such as high temperature stability, oxidation and creep
resistance. At higher temperatures, these ceramics undergoes phase separation and crystallization, leading to the
formation of ceramic nanocomposites. Silicon oxycarbide (Si-O-C) and silicon carbonitride (Si-C-N) based
ceramics, obtained by the SST of polysiloxanes and polysilazanes have been extensively investigated.
Moreover, incorporation of additional elements into these ceramic systems have been claimed to enhance the
overall thermal stability. In this context, the effect of addition of Zr and Hf into the Si-C-O ceramic matrix is
discussed.
The synthesis involves the chemical modification of polysiloxane and polysilazane by zirconium tetra(n-
propoxide) and hafnium tetra(n-butoxide) for the production of SiZrCN(O) and SiHfCN(O), respectively [1-6]
.
The information regarding the polymer to ceramic conversion is derived from the thermogravimetry data.
Furthermore, the presence of Si-O-Zr and Si-O-Hf band is confirmed from the Fourier transform infrared
(FTIR) spectroscopy. The X-ray diffraction data revealed that the as-thermolysed ceramics are essentially
amorphous up to 800 ˚C. Further, heat-treatment has resulted in the phase separation and crystallization of ZrO2
and HfO2, from the SiZrCN(O) and SiHfCN(O) system, respectively. This can also be inferred from the
transmission electron microscopy (TEM) image (Fig.1).
68
Fig. 1 HR-TEM image indicating the presence of zirconia[3]
and hafnia crystallites in the metal oxide modified
ceramics[4]
Moreover, in both the cases, it is observed that the tetragonal phases are the first to crystallize and found to be
stable at room-temperature, though the monoclinic is the room temperature stable phase. This stability can be
attributed to the crystallite size effect, since it is observed that the coarsening of the zirconia and hafnia have
resulted in the respective phase transformation from tetragonal to monoclinic phase. The XRD and the TEM
studies have provided the evidence of this coarsening to occur upon heat-treatment and increased holding time
(Fig.2). In case of hafnia, the critical crystallite radius essential for the stability of tetragonal phase is
determined to be ~ 4 nm. Further, heat-treatment at 1600 ˚C also has indicated the presence of hafnon,
produced by the solid state reaction between SiO2 and HfO2.
Fig. 2 TEM image indicating coarsening to occur due to heat-treatment and increased holding time[5]
.
HfO2
69
Besides, significant influence of the chemistry and polymer architecture are observed on the microstructural
evolution of these metal oxide modified PDCs [6]
. The results indicated that the hafnium alkoxide modification
of commercial polysilazane (HTT-1800) and cyclotrisilazane (CTS) have resulted in entirely different
microstructures [Fig.4]. Significant amount of hafnium enrichment is observed in the case of the latter, and also
the CTS modified system has shown lower thermal stability. This emphasizes the importance of proper
selection of polymeric precursor as the reaction path ways will be entirely different leading to diverse
microstructures.
Fig. 4. TEM indicating the presence of enriched hafnium in the case of CTS precursor [6]
.
References
1. E. Ionescu, B. Papendorf, H. J. Kleebe, F. Poli, K. Muller and R. Riedel, J. Am. Ceram. Soc., 93, 1774-82
(2010).
2. E. Ionescu, B. Papendorf, H. J. Kleebe and R. Riedel, J. Am. Ceram. Soc., 93, 1783-89 (2010).
3. E. Ionescu, C. Linck, C. Fasel, M. Muller, H.J. Kleebe and R. Riedel, J. Am. Ceram. Soc., 93[1] 241-50
(2010).
4. R. Sujith, A. B. Kousaalya and R. Kumar, J. Am. Ceram. Soc., 94[9] 2788-91 (2011).
5. R. Sujith, A. B. Kousaalya and R. Kumar, Ceram. Intnl. 38 1227-33 (2012).
6. B. Papendorf, K. Nonnenmacher, E. Ionescu, H. J. Kleebe and R. Riedel, Small 7 970-8 (2011).
