«ҚОҒАМДЫ АҚПАРАТТАНДЫРУ» V ХАЛЫҚАРАЛЫҚ ҒЫЛЫМИ-ПРАКТИКАЛЫҚ КОНФЕРЕНЦИЯ
35
III. Hardware part of the elevator
The overall dimension of the elevator is
650x900x240mm. There are four stages in the
system. Each stage has a height of 170mm. In
the middle of each floor we mounted SIEMENS
SIMATIC HMI KTP-600 device.
Figure 5 - The overall view of elevator
IV. Conclusion
In this project design of laboratory sized
elevator prototype and PLC software description
have been introduced. The PLC based control of
the elevator system is determined to operate
properly and efficiently. The algorithm principle
performed can be applied in industries, hospitals
and can be used in educational purposes. The
PLC based elevator control is efficient to carry
more people in a given time, because the
program decides which cabin to use when the
given amount of people came to use the
elevator. For future works all conditions such as lightning in the cabin with efficient energy
management, safety precautions by installing ultrasonic sensors and others will be added.
References:
1. About Elevators Retrieved from http://www.otisworldwide.com/pdf/aboutelevators.pdf
2. S.Krishankant, ”Computer Based instrumentation Control’, New Delhi: PHI Learning Pvt. Ltd,
2009
3.Programmable
Logic
Controllers
Retrieved
from
http://www.coe.montana.edu/ee/courses/ee/ee367/pdffiles/aamunrud.pdf
4. Human Machine Interface: SIMATIC HMI - Efficient to a new level Retrieved from
http://w3.siemens.com/mcms/automation/en/human -machine-interface/pages/default.aspx
5. Xiaoling Yang; Qunxiong Zhu; Hong Xu, ”Design and Practice of an Elevator Control System
Based on PLC,” Power Electronics and Intelligent Transportation System, 2008. PEITS ’08. Workshop on ,
vol., no., pp.94,99, 2-3 Aug. 2008
UDC 616-008.22
KOZBAYEV B.M., SHAGDAROV A.E.
CREATION OF A MICROCONTROLLER CONTROL SYSTEM OF MOBILE OBJECTS
ON THE BASIS OF THE ANALYSIS OF THE ELECTROMAGNETIC ACTIVITIES OF A
BRAIN
(L.N. Gumilyov Eurasian National University, Astana)
Currently, one of the most informative methods of studying the human brain from the
standpoint of its integrated system activity is the method of electroencephalography. This method is
based on the registration of the total electrical activity of neurons in the brain, withdrawn from the
surface of the scalp - an electroencephalogram (EEG). Electroencephalography enables qualitative
«ҚОҒАМДЫ АҚПАРАТТАНДЫРУ» V ХАЛЫҚАРАЛЫҚ ҒЫЛЫМИ-ПРАКТИКАЛЫҚ КОНФЕРЕНЦИЯ
36
and quantitative analysis of the functional state of the brain and its reactions under the influence of
stimuli. EEG recording is widely used in diagnostic and therapeutic work (especially common in
epilepsy), in anesthesia, as well as in the study of brain activity associated with the implementation
of functions such as perception, memory, adaptation, etc ) [1-4].
In this paper, the technique of implementing the analysis of brain electromagnetic waves.
Reading the brain is proposed to implement using the microcontroller system "Arduino". "Arduino"
- is an open source platform designed for quick and easy development of a variety of electronic
devices (picture 1). [5,6]
Keywords: Brain, analysis, Arduino, control, wairless, helmet, mobile object.
Picture 1. Architecture computing complex brain wave reader.
Next, you need to collect a helmet or something like that, in my case is a set of shielded
electrodes, connected to the head. These are all connected to an amplifier, which in turn sends the
data to the device, which collects the database with which the mobile object management
performed.
For the analysis of waves used Fast Fourier Transform:
f(ω) =
√
∫
f(x)e
dx
(1)
Different sources may give definitions that are different from the coefficient of the given option to
the above integral, and the sign "-" in the exponent. However, all of the properties will be the same,
though the form of some formulas may change.
The Fourier transform is used in many areas of science - in physics, number theory, combinatory,
signal processing, probability theory, statistics, cryptography, acoustics, oceanography, optics,
geometry, and many others. The signal processing and related fields Fourier transformation is
generally regarded as the decomposition of the signal on the frequency and amplitude that is
reversible transition from a temporary area (time domain) to the frequency domain (frequency
domain). Rich application based on several useful properties of the transformation:
The discrete version of the Fourier transform can be quickly designed on computers using
the fast Fourier transform algorithm (FFT) [7] and for Arduino platform will look like:
int n=32 ,n1=0,m=7;
double p,w,z,r,k,r1,r2;
double c =0;
double s=0;
int y [31];
void setup() { Serial.begin(9600);
«ҚОҒАМДЫ АҚПАРАТТАНДЫРУ» V ХАЛЫҚАРАЛЫҚ ҒЫЛЫМИ-ПРАКТИКАЛЫҚ КОНФЕРЕНЦИЯ
37
for(int i = 0; i <=7; i++)
{ y[i]=1;}
double f= 250000;
double t = 0.125* pow(10,-6);
p = PI*f*t;
w= 2*p;
for (int i=0; i<=7;i++)
{ z=w*i;
c=c+(y[i]*cos(z));
s=s+(y[i]*sin(z)); }
r= sqrt((pow(c,2))+(pow(s,2)));
Serial.print("amplituda r=");
Serial.println(r,10);//
f= (-acos(c/r))-(PI*t*f);// угол фазы
Serial.print("ugol fazi f=");
Serial.println(f,10);
if (s<0) {f=-f;}
k = sin(p)/p;
r1 =k*r;
r2 =k*r1;
Serial.print(" reshet4atoi Y(t),S0(f)=");
Serial.println(r * t, 10);
Serial.print(" stupen4ataya Y(t),S1(f)=");
Serial.println(r1 * t, 10);
Serial.print("kuso4no lineinaya Y(t) S2(f)=");
Serial.println(r2*t,10);
Serial.print("fazovi ugol:");
Serial.println(f,10); }
void loop() { }
Thus, management is realized with a mobile object by analyzing brain waves.
References:
1. https://www.arduino.cc/
2. http://esp8266.ru/
3. http://cxem.net/arduino/arduino63.php
4. https://learn.adafruit.com
5. Cloud Technology - remote management of mobile objects, "Bulletin of ENU. LN Gumilyov",
special edition, ISSN 1028-9364, Astana, 2012, pp. 60-63Практические аспекты применения и
классификация микроконтроллеров, «Вестник ЕНУ им. Л.Н. Гумилева», специальный выпуск, ISSN
1028-9364, г.Астана, 2012 год, стр.361-363.
6. Prospects for analog signal processing in digital equipment, materials of the International
scientific-practical conference "Formation of modern science", Publishing House "Education and Science"
s.r.o. (Czech Republic, Prague), 2014.
7. Атанов С.К., Кази Д.Е. «Расчет эффективности работы микроконтроллера с аналоговым
вычислителем» №1503 от 11.11.13 МЮ РК
«ҚОҒАМДЫ АҚПАРАТТАНДЫРУ» V ХАЛЫҚАРАЛЫҚ ҒЫЛЫМИ-ПРАКТИКАЛЫҚ КОНФЕРЕНЦИЯ
38
UDC 004.2
KOZHIRBAYEV ZH., ISLAM SH.
A DISTRIBUTED PLATFORM FOR SPEECH RECOGNITION RESEARCH
(Nazarbayev University, National Laboratory, Astana, Kazakhstan)
Abstract
Distributed and parallel processing of big data has been applied in various applications for
the past few years. Moreover, huge advancements took place in usability, economic efficiency, and
multiplicity of parallel processing systems, with big data analysis and speech recognition research
supported by many researchers.
In this paper we examined and investigated which parts of speech recognition research may
be parallelized and computed using distributed computing platforms. Firstly, we address the case of
efficiently computing n-gram statistics on MapReduce platforms to build a language model (LM).
Secondly, we show how the Automated Speech Recognition (ASR) tool can work efficiently
regarding the speed and fault-tolerance in distributed environment such as Sun GridEngine (SGE).
Keywords: Distributed Computing, Sun GridEngine, Hadoop ecosystem, MapReduce
1. Introduction
The automatic speech recognition area went through several major progresses in the past
decade initiated by changes in algorithms, signal processing, system architectures and hardware.
The last two aspects play a significant role in Speech Recognition Research. The trigger for the
development of distributed computing is the affordability of the cost effective, powerful machines
as well as network tools. Several high-powered machines that are connected to one another make
the total achievable computing power significantly broad. This kind of system might perform
greater results rather than a single powerful machine. Distributed computing is the decentralized
way of dealing with the computing stages of the application which can be distributed among the
linked machines.
The remainder of this paper is organized as follows. Section 2 describes the parallelization
technologies which might be applied in the speech recognition research. To be precise, the Hadoop
ecosystem, which can be employed in building a language model when the size of the language
corpus is big; and the Sun GridEngine, which distributes the tasks such as data alignment and audio
decoding, will be presented in this section. Section 3 demonstrates the results obtained during the
experiments. Finally, the last section concludes the paper and suggests further investigations in this
area.
2. Parallelization technologies
During the research the parts of the speech recognition processes are examined in order to
identify the tasks which may be parallelized. The two major tasks were distinguished which take a
while when they are running on one single machine. Therefore, the distributed computing was
applied to these processes and they will be described in this section in more details.
2.1 MapReduce in building LM
In this work, we address the problem of efficiently computing n-gram statistics on
MapReduce platform. This is needed to build a language model which will be later converted to the
ARPA format.
MapReduce [2, 3] is used widespread since the past few years as a programming model as
well as its open-source realization Hadoop. A platform for parallel data computing is supplied by
MapReduce. It enforces a harsh programming model; however, it provides its users with technical
options such as dealing with machine errors as well as an automatic spread of the processing. In
order to utilize it effectively, issues have to be cast into its programming model, considering its
characteristic features.
«ҚОҒАМДЫ АҚПАРАТТАНДЫРУ» V ХАЛЫҚАРАЛЫҚ ҒЫЛЫМИ-ПРАКТИКАЛЫҚ КОНФЕРЕНЦИЯ
39
In previous experiments only one single machine was used to build a language model. It
takes a while to process such a big corpus to generate unigrams, bigrams and trigrams. Moreover,
the corpus might be increased in the future and the necessity to build a language model will rise
again. Therefore, the Hadoop cluster was built on seven nodes where each node has 8 Gb RAM and
4 cores. The results that obtained using MapReduce platform were significant which can be seen in
Section 3.
Table 1
Comparison of the computing environments
Single machine
Hadoop cluster (7 nodes)
# of cores
RAM
# of cores
RAM
4
16 Gb
28
56 Gb
2.2 Parallelization in Kaldi
A toolkit named Kaldi [1] was used for speech recognition research. The perfect condition
for processing is a cluster of Linux nodes using SGE, with the admission to shared folders through
either NFS or similar network filesystem [4, 5]. The perfect form of processing environment as well
as required limits to perform Kaldi will be explained below in this section.
The speech recognition research using Kaldi toolkit was conducted in one single machine
since it can be easily configured to run on a single machine if it is a supercomputer. However, the
machine used in the research has 16 cores and only 32 Gb RAM. Kaldi toolkit performs some tasks
sequentially and some tasks parallel. For example, the data alignment and audio decoding jobs are
run parallel. Also, the aspect which should be considered during the research is the size of the
language model. The decoding task depends on the size of the LM and requires approximately 6 Gb
RAM. Therefore, only 5 cores of the single machine are useful for the recognition process. This
approach was inefficient. Therefore, the new approach using SGE was conducted because of the
availability of the cost effective, much powerful nodes and network tools.
Sun GridEngine is the open-source grid control instrument which is used widespread.
Recently Oracle started supporting SGE and renamed it Oracle GridEngine [4]. The currently used
version in the mentioned system is 6.2u5; SGE is time proven and earlier made versions are still
stable and widely used. Furthermore, different open-source possible tools to SGE do exist, however,
built platform in the mentioned system refer to the version that is nowadays supported by Oracle.
Table 2
Comparison of the computing environments
Single machine
SGE cluster (17 nodes)
Total
Used for recognition
Total
Used for recognition
16 cores
5 cores
144 cores
46 cores
32 Gb RAM
30 Gb RAM
336 Gb RAM
276 Gb RAM
The grid cluster was build using 13 machines where each node has 8 cores and 16 Gb RAM,
and 4 virtual machines, that are deployed on Openstack, with 8 cores and 32 Gb RAM. The results
which obtained using SGE platform were significant which can be seen in Section 3.
3. Experiments and Results
This section provides the results of both building the language model and audio decoding. It
can be seen from the below tables that results given by Hadoop and SGE significantly overcomes
results by single machine.
«ҚОҒАМДЫ АҚПАРАТТАНДЫРУ» V ХАЛЫҚАРАЛЫҚ ҒЫЛЫМИ-ПРАКТИКАЛЫҚ КОНФЕРЕНЦИЯ
40
Table 3
Results of building LM
# of runs Corpus
and
its size
NGrams
Elapsed time
Single machine
Hadoop cluster
1
Kazcorpus
1.6G
Unigram
>> 4 hours
19mins, 55sec
2
Bigram
>> 4 hours
20mins, 28sec
3
Trigram
>> 4 hours
20mins, 19sec
Table 4
Results of audio decoding
# of runs
Data Set
Elapsed time
Single machine
SGE cluster
1
10 hours audio set
≈ 3184mins
≈ 385mins
4. Conclusion
In this paper, we have applied MapReduce to compute n-gram statistics for the language
model. Moreover, the ideal environment which has a SGE installed in it may provide a significant
improvement for Kaldi toolkit. Both of these applied platforms decrease the processing time in a
sufficiently great way. Further investigations will be conducted to explore new features of
distributed computing.
To sum up, we will continue to improve the Speech Recognition Research in terms of
parallelization.
Acknowledgments
The authors would like to thank the National Laboratory Astana for the resources used to
perform these investigations and Karabalayeva M. and Yessenbayev Zh. for providing results of
audio decoding that added a value in better analysis for this paper.
References:
1. Povey, D, Ghoshal, A, Boulianne, G, Burget, L, Glembek, O, Goel, N, Hannemann, M, Motlicek,
P, Qian, Y, Schwarz, P, Silovsky, J, Stemmer, G & Vesely, K 2011, “The Kaldi Speech Recognition
Toolkit”, IEEE 2011 Workshop on Automatic Speech Recognition and Understanding
2. Zaharia, M 2014, Introduction to MapReduce and Hadoop, UC Berkeley RAD Lab
3. MapReduce, viewed 15 April 2016, URL https://hadoop.apache.org
4. Sun microsystems 2009, Sun N1 Grid Engine 6.1 User's Guide, Santa Clara, CA, USA
5. Open Grid Engine, viewed 15 April 2016, URL http://gridscheduler.sourceforge.net
«ҚОҒАМДЫ АҚПАРАТТАНДЫРУ» V ХАЛЫҚАРАЛЫҚ ҒЫЛЫМИ-ПРАКТИКАЛЫҚ КОНФЕРЕНЦИЯ
41
UDC 004.5
NURGALIYEV K. S.
1
, KAPSALYAMOV A. B.
1
, SERIMBETOV B. A.
2
STUDY AND DEVELOPMENT OF A DESKTOP HAPTIC INTERFACE FOR
TELEOPERATION
(
1
Master's Degree student of the Information Technologies Department, Kazakh University of
Technology and Business, Astana, Kazakhstan,
2
Associate Professor, Candidate of Technical sciences, Head of the Information Technologies
department, Kazakh University of Technology and Business, Astana, Kazakhstan)
Abstract
Teleoperation systems provide the users with a possibility to perform sophisticated tasks in
distant environments. A haptic interface can allow the operator to use not only visual and audial
senses but also the tactile senses to perceive his/her environment and provide more accurate and
precise control over the task that is being performed by the robot remotely. This overview is a
description of a haptic interface used for teleoperation of the complex anthropomorphic robotic
systems in an intuitive way. The system allows six degrees of freedom when connected with two
points of the human limb. Force feedback is provided at the users fingertips when the robot is in
contact with an object recognized as an obstacle.
Keywords: degree of freedom, haptic, tactile, interface, simulation, design
I. Haptic interface
Haptic interface is a synergy between human tactile senses and a robotic system, which
consists of three main components: (1) the electromechanical system with a certain number of
degrees of freedom (DOF); (2) the teleoperated robotic device; (3) and the algorithm from the
computer or any other programmable machine that provides the transmission of the commands. [1]
It is one of the growing areas in human computer interaction, which deals with sensory
communication with robotic systems by applying motions, vibrations or forces to the user.
Teleoperation of the robots is a term used to name the remote control of a robotic system to make it
perform a certain task given by the operator. [2] Haptic interfaces enable manual interactions with
virtual environments or teleoperated remote systems using force feedback.
A haptic interface can be mainly of two kinds: a grounded system, which has its base, fixed
with the wall or the floor[3]; and a wearable system where the main fixing point is represented by
the human body [4], [5]. The advantages of the first type are the capability to deliver a higher force
feedback and to spare the user sustaining the weight of the whole system. While the second type
generally has a bigger workspace and it results more natural because it is normally wearable and
repeats human anatomy and ergonomics. In comparison with classical joysticks the main
advantages of using an interface that connects with the forearm is that the operator can give the
orientation commands to the robot not using the wrist but the entire arm. However, manipulating
the end-effector of the robot using only the fingertips provides larger amount of precision. By
applying less force, the operator is then able to maintain more complex and more wide-scale
performance from the teleoperated robot. The main objective of the present work is to describe this
particular approach, as well as to show a simulation methodology using the programming platform.
The Interface Design section includes the technical description of the hardware and all the electrical
components used during the building process. The chosen methodology allows to manipulate the
robotic system and to visualize the robots behavior from different perspectives. The discussion of
the results with appropriate outflow of Conclusion part will summarize the paper and give a clearer
understanding of what purposes this robot might serve and in what fields it might be implemented.
The suggestions on the future work regarding the improvement of this robots design will be
provided prior to the end of this paper.
|