Upload
others
View
1
Download
0
Embed Size (px)
Citation preview
Lappeenranta University of Technology
School of Engineering Science
Degree Program in Computer Science
Roman Zhohov
EVALUATING QUALITY OF EXPERIENCE AND REAL-TIME
PERFORMANCE OF INDUSTRIAL INTERNET OF THINGS
Examiners: Professor Eric Rondeau, University of Lorraine
Professor Jari Porras, Lappeenranta University of Technology
Professor Karl Andersson, Lulea University of Technology
Supervisors: Professor Karl Andersson, Lulea University of Technology
Per Johansson, InfoVista AB, Skelleftea, Sweden
Dimitar Minovski, Lulea University of Technology
ii
ABSTRACT
Lappeenranta University of Technology
School of Engineering Science
Degree Program in Computer Science
Roman Zhohov
Evaluating Quality of Experience and real-time performance of Industrial Internet of
Things
Master’s Thesis
46 pages, 16 figures, 3 tables
Examiners: Professor Eric Rondeau, University of Lorraine
Professor Jari Porras, Lappeenranta University of Technology
Professor Karl Andersson, Lulea University of Technology
Keywords: QoE, Industry 4.0, QoS, real-time communications, IIoT, CPS.
The Industrial Internet of Things (IIoT) is one of the key technologies of Industry 4.0 that
will be an integral part of future smart and sustainable production. The current constituted
models for estimating Quality of Experience (QoE) are mainly targeting the multimedia
systems. Present models for evaluating QoE, specifically leveraged by the expensive
subjective tests, are not applicable for IIoT applications. This work triggers the discussion
on defining the QoE domain for IIoT services and applications. Industry-specific KPIs are
proposed to assure QoE by linking business and technology domains. Tele-remote mining
machines are considered as a case study for developing the QoE model by taking into
account key challenges in QoE domain. As a result, QoE layered model is proposed, which
as an outcome predicts the QoE of IIoT services and applications in a form of pre-defined
Industrial KPIs. Moreover, software tool and analytical model is proposed to be used as an
evaluation method for certain traffic types in the developed model.
iii
ACKNOWLEDGEMENTS
I would like to thank my thesis advisor prof. Karl Andersson at Lulea University of
Technology for providing a fantastic opportunity to work in such good environment and
made my second year of PERCCOM master program less stressful.
I must express gratitude to my supervisor Per Johansson, this work would be impossible
without him. Without doubt, it was his work that inspired me to do my research. I am glad
that he could always find the time for our endless discussions. Special thanks to Niklas
Ögren whose comments have shaped this work. His expertise in the fields of wireless
communications and protocols always made me to look at the problem from the different
angle. I thank Dimitar Minovski who combined roles of colleague, co-author, flat-mate and
advisor. It was a wonderful time that we shared, and I appreciate his help and support.
I would also like to acknowledge TEMS team (Ulf Marklund, Magnus Furstenborg, Anna
Lindberg and others) that provided me with great tools and vital help to create prototypes
and develop ideas. Thanks to all my colleagues from InfoVista, it was a pure pleasure to
work with them.
I also thank prof. Sergey Bunin who was my supervisor during bachelor studies. My
research work started under his supervision and I cannot overestimate how much I have
learnt during my work with him.
Finally, I thank my mother Olena Zhohova. Her endless love, support and care helped me
to achieve everything that I have now.
1
TABLE OF CONTENTS
1 INTRODUCTION ............................................................................................. 4
1.1 BACKGROUND......................................................................................................... 4
1.2 INDUSTRIAL CASE STUDY ........................................................................................ 5
1.3 MOTIVATION .......................................................................................................... 7
1.4 RESEARCH QUESTIONS ............................................................................................ 7
2 LITERATURE REVIEW ................................................................................... 8
3 METHODOLOGY .......................................................................................... 10
4 DEFINING QUALITY OF EXPERIENCE IN INDUSTRIAL INTERNET OF
THINGS ................................................................................................................ 13
5 EVALUATING QUALITY OF EXPERIENCE IN INDUSTRIAL INTERNET OF
THINGS ................................................................................................................ 18
5.1 QOE LAYERED-MODEL PROPOSAL ......................................................................... 18
5.1.1 Physical Layer ................................................................................................. 19
5.1.2 Network Layer .................................................................................................. 19
5.1.3 Service Layer ................................................................................................... 20
5.2 SYSTEM’S ARCHITECTURE AND TYPES OF TRAFFIC ................................................ 21
5.3 REAL-TIME SENSOR STREAM................................................................................ 21
5.4 REAL-TIME PERFORMANCE EVALUATION AND PREDICTION ................................... 23
5.4.1 Background and network architecture ............................................................ 23
5.4.2 Time synchronization ....................................................................................... 25
5.4.3 Experimental setup .......................................................................................... 28
5.4.4 Latency prediction ........................................................................................... 30
6 SUSTAINABILITY ASPECTS ....................................................................... 35
7 FUTURE WORK ............................................................................................ 39
8 CONCLUSION ............................................................................................... 40
9 REFERENCES .............................................................................................. 41
2
LIST OF SYMBOLS AND ABBREVIATIONS
ADB Android Debug Bridge
BER Bit error rate
BLER Block error rate
CPS Cyber-Physical System
D2D Device-to-device
EPC Evolved Packet Core
EPS Evolved Packet System
ICT Information and communication technology
IIoT Industrial IoT
IoT Internet of Things
KPI Key performance indicator
LTE Long Term Evolution
M2M Machine-to-machine
MIoT Multimedia IoT
ML Machine Learning
MOS Mean opinion score
NTP Network time protocol
OEE Overall equipment efficiency
PEVQ Perceptual Evaluation of Video Quality
POLQA Perceptual Objective Listening Quality Analysis
PTP Precision time protocol
QoC Quality of Context
QoD Quality of Data
QoE Quality of Experience
QoN Quality of Network
QoS Quality of Service
RAN Radio Access Network
RFID Radio-frequency identification
RSRP Reference signal received power
RSRQ Reference signal received quality
3
RSSI Received signal strength indicator
RTOOL Real-time Tool
RTT Round-trip time
SINR Signal to interference plus noise ratio
SLA Service-level Agreement
SNR Signal to noise ratio
UX User eXperience
4
1 INTRODUCTION
The fourth industrial revolution is predicted a-priori and manifested as Industry 4.0. Smart
Factories, Industrial Internet of Things (IIoT) and Cyber-Physical Systems (CPS) are main
enabling components of Industry 4.0 that have tremendous effect on industrial application
scenarios and automation [1]. Digitalizing of industrial processes will deliver an important
boost in productivity and trigger economic growth [2]. Industry 4.0 transformations have
attracted significant attention from academia and industry, it is reflected in the vast number
of global projects and initiatives that are addressing IIoT concept. For instance, Industrie
4.0 project was accepted in “Action Plan High-tech strategy 2020” by German Federal
Government in July 2010 [3]. Another example is the Industrial Internet Consortium (IIC)
which is enabling and accelerating adoption of the Industrial Internet as an essential step to
increase competitiveness in key industry sectors. According to the predictions
implementation of IIoT will have a tremendous effect on the global economy, PwC’s 2016
Global Industry 4.0 survey respondents expect to see US$421 billion in cost reductions and
US$493 billion in increased annual revenues p.a [4].
1.1 Background
The core of IIoT and CPS is essentially the robust exchange of information. Originally,
conventional telecom networks could not cope with the industry-specific requirements for
reliable, predictable and efficient communication. Industrial networks were mainly based
on diverse deterministic bus technologies (controller area network – CAN, PROFIBUS,
INTERBUS, etc.) to satisfy strict requirements of hard real-time automation systems.
Development of industrial communication systems and networks is shown in Fig. 1, as one
may notice industrial communication system move from various bus technologies to
mobile wireless solutions such as 5G that might cover all possible use cases. Recent
advances in communication technologies (especially, wireless solutions) made possible
interconnection of numerous IIoT and creating CPS. It is evident that mobile
communications will be a key enabler for IIoT [2]. However, each industrial scenario has
specific requirements in terms of communication technology. This leads to increasing the
5
number of metrics and technology specific parameters that should be considered while
evaluating system’s performance.
From the industrial and telecom perspective, quality evaluation is indispensable business
goal for harmonizing the users’ experiences, agreed and delivered through SLA. Recent
research discoveries in measuring QoE leverage the Internet Service Providers (ISPs) in
understanding the aberrant behavior of the communication link and service performance.
However, the legacy performance evaluation techniques mainly define QoE as an
extension of network-centric Quality of Service (QoS), reckoning the social and context
parameters in the equations, which as a result produce a Mean Opinion Score (MOS). This
aligns the user’s perspective of perceived service quality throughout subjective and
objective tests. Evaluating QoE in IoT services, especially in industrial domain, deviates
from this conventional paradigm, as the scope goes beyond measuring the user’s
perception.
Fig. 1. Evolution of industrial communication [5]
1.2 Industrial case study
This work offers a real-world case study of IIoT that shows the potential of mine
digitalization that provides a number research challenges. The number of industrial and
research initiatives are targeting Industry 4.0 transformation. For instance, PIMM DMA
(Pilot for Industrial Mobile communication in Mining, Digitalized Mining Arena) project
6
which is continuation of the PIMM project. The PIMM project was a two-year long project
that was focusing on implementing the mobile network in the mine and testing a variety of
IIoT use cases. PIMM DMA is focused on implementing a state of the art mobile network
in a mine and test several applications that are enabled by mobile communication. The main
purpose of PIMM DMA project is innovations in the areas of [6]:
- Service operations for industrial mobile networks;
- Development of cellular communications;
- Industrial products and services enabled by mobile communications;
- Industrial automation and digitalization of the mining industry;
- Systems-of-systems;
As a result, SLA templates that considers technical and business interests will be created
for all stakeholders and business players. Underground mine is hazardous environment
with a risk of being injured. Moreover, work under these conditions can cause immediate
(acute) or long-term (latency) health effects. For instance, occupational diseases in mining
include: asbestosis, mesothelioma, silicosis, cancers, chronic obstructive lung disease,
hearing loss and others. Typical system’s architecture for tele-remote underground vehicles
is shown in Fig. 2 and consists of remote control station, different communication
networks and remotely operated mining vehicles. Each element will be further explained
later in this work.
Fig. 2. System’s architecure
7
1.3 Motivation
The motivation of this work is to examine implementation tele-remote operation of mining
vehicles that will lead reduce need of the human operator in the harsh environment of the
underground mine. From the industrial prospective, by introducing IIoT mining industry
may benefit from real-time monitoring, analytics and control. For example, the best
operation strategy can be find by measuring fuel efficiency, productivity and controlling
operator’s behavior.
From the research prospective, tele-remote system is extremely suitable for defining and
developing QoE concept in IIoT as it can be used as a sample to analyze the QoE domain
and locate the intrinsic challenges within IIoT. Firstly, it has attracted significant attention
both from academia and industry that is resulted in the number of research projects.
Secondly, remote operation of mining vehicles provides multimedia and multimodal
interaction between system and operator that will increase number of qualities and metrics.
Moreover, tele-remote operation of industrial vehicles can benefit from research on
robotics, industrial automation and real-time communications.
1.4 Research questions
Research questions formulation determines methodologies that will be used throughout the
work. This work is addressing two main research questions:
1.1 How QoE in IIoT is different from conventional definition? Why there is a need to
redefine QoE concept for IIoT? What is the definition of QoE in IIoT?
1.2 How to eval uate QoE in IIoT scenario? How to define a general model for QoE
evaluation in IIoT applications?
8
2 LITERATURE REVIEW
The emergence of the 5G networks creates new challenges in QoE evaluation, for instance,
author of [7] suggests adaptive estimation and self-optimization of the perceived quality in
the context of increasing number of M2M/D2D communication and IoT. According to the
Ericsson Mobility Report, the number of connected devices will exceed 30 billion by 2023,
among which 20b will be related to IoT, with short and wide range transmission
capabilities. In addition, more than 20 percent of the world’s population will be covered by
5G in 2023 [8].
Although, QoE, IIoT, and CPS are widely discussed in the research community and
industry [9, 10], but there are a few works on QoE in IoT [11, 12, 13], and absence of
studies regarding QoE in Industrial IoT.
Wu et al. in [11] proposed the concept of Cognitive IoT in which “things” act as agents to
build virtual environment. Authors develop the concept of layered-QoE framework which
consists of four main layers - Access, Communication, Computation, Application, with
corresponding metrics. However, authors do not provide any practical examples or
scenarios and do not attempt to redefine the QoE domain. Authors of [14] conducted an
experiment on correlation between QoS parameters (delay and packet loss ratio) and QoE
for the networked actuator in function of experimental parameters. In [12] authors made an
attempt to define QoE concept for Multimedia IoT (MIoT) by extending the previous
layered QoE framework. It consists of five main layers - Physical, Network, Combination,
Application and Context. Each of mentioned layers has corresponding qualities and
metrics, for instance, Quality of Data (QoD) for the physical layer. In contrast to other
works on QoE for IoT, authors conducted experimental evaluation for IoT vehicle
application by measuring QoS and QoD parameters and performed subjective assessment,
using the Mean Opinion Scores (MOS) scale. As a result, linear and non-linear regression
models for measured parameters and MOS have been computed. Overall QoE was defined
in terms of QoD, translated into data accuracy, and QoS, measuring throughput and
network delay. Authors of [13] refer to the layered-QoE model proposed in [12]. They
introduce physical and metaphysical metrics for IoT, by pointing out the complexity of
9
mapping the QoS parameters into QoE metrics. Metaphysical metrics are more scalable,
gathering context information, and considered as an intermediate layer between physical
metrics and QoE.
As we can see, the general trend is layered-QoE which considers several impact factors
which means that QoE in IoT is not only affected by network performance but also has
many other influence factors (for example, sensing quality, context quality, etc.)
According to [15, 16, 17], introducing Industrial IoT in the mining industry will result in
energy and cost benefits. Moreover, IIoT can improve safety by predicting the failures in
equipment/machines, moving from preventive to predictive maintenance strategy. Most of
the operations can be automated which leads to new business models and processes. Real-
time data collection and analytics will bring new insights and data-driven model for both
mine planners and business stakeholders.
Most of the mining companies identified IIoT and corresponding data analytics among
their top-three priorities. For instance, Rio Tinto [18] has already experimented with
autonomous mining vehicles since 2008. Radar guidance system, GPS receiver and more
that 200 sensors are installed on each mining vehicle. The mining site is managed from the
operation center that collects the data from the trucks and other equipment which results in
3D model of the work space and comprehensive analytics.
Zhou et al. in [19] discussed the advantages of open, highly connected and interoperable
IIoT-based systems compare to legacy monitoring solutions. Moreover, authors listed real-
world examples of the IIoT in mining and discussed feasibility of IIoT implementation in
coal mines. However, IIoT technology implementation leads to new challenges such as
security and privacy, equipment adoption for harsh environment (which is particularly
important for gassy environment of the coal mines with a constant risk of explosion),
network interoperability and industry-specific data analytics.
10
3 METHODOLOGY
Choosing right research methodology and methods is an essential step to plan and
performing the research work. Methodology defines organization, designing, conducting
and evaluating research. Moreover, appropriate selection of research strategy assures the
quality of the conducted research. This section gives an overview of the most commonly
used research methods and methodologies and justifies methodology selection for this
work. Research methodology or strategy is often referred as a “systematic process of
carrying out the research work and solving a problem including research methods” [20].
Research methods can be defined as “a part of methodology denoting its own category of
methods” [20].
Fig. 3 shows both qualitative and quantitative research methods and methodologies,
however, there are some methods that can work well for both parts, presented diagram
should be analyzed in top-to-bottom strategy in order to choose appropriate research
methods and methodology.
Fig. 3. The portal of qualitative and quantitative research methods and methodologies [20]
Quantitative and qualitative research methodologies are different from each other as a
result one of the first steps is to choose right approach. Quantitative methodologies are
typically used in case of proving a phenomenon by evaluating data sets (quantities) which
11
are obtained during the tests or experiments. On the other hand, qualitative strategies are
chosen in case of studying an artifact through creating new theories, hypothesis or
products. This work includes both activities: establishing a new theoretical foundation of
QoE in IIoT which is essentially a qualitative approach and experimental evaluation which
is quantitative task.
A philosophical assumption plays a significant role for the whole research since it defines
the stand point for the project. Main philosophical assumptions such as Positivism,
Realism, Interpretivism and Criticalism are listed in Fig. 3.
Positivism presumes that knowledge is gained through observation of the objectively given
reality that is independent of the researcher and its instruments. In positivism, researcher
adopts deductive approach to increase predictive understanding of a phenomenon [21, 22].
Realism approves that the entities hypothesized by scientific theories are real in the world,
with appropriately attributed attributes proposed by adequate scientific theories.
Interpretivism as an opposite to positivism involves researcher interest as a result of human
interaction and interest into the research process. Criticalism presumes that the research
process should be reflective and conducted as a critique of the given reality. [20]
Commonly used research methodologies for quantitative research are: Experimental
Research, ex post-facto Research, Surveys (Longitudinal and Cross-sectional) and Case
Study. For qualitative research, most frequently used methodologies are: Surveys, Case
Study, Action Research, Exploratory Research, Grounded theory, and Ethnography [20].
This work is logically divided into two parts: defining QoE in IIoT and evaluating QoE;
each part should be tackled using different research methodologies and methods. It is
important to notice that industrial case study is complex and involves many stakeholders
and players as a result evaluation can be performed partially for certain traffic types or
services. Case study can be applied for the first part since this empirical study investigates
a phenomenon in the particular context using the mix of quantitative and qualitative
methods [23]. Regarding our scenario, the aim is to analyze QoE domain within a specific
industrial scenario (tele-remote operation of mining vehicles). The results that are obtained
by investigating use-case can be later generalized to the entire range of industrial systems.
12
However, hypotheses that were formulated after case study need to be evaluated through
the number of experiments. Experimental research strategy can be applied to verify
hypotheses and provide cause-effect relationships, specially correlation between QoE in
IIoT and quality parameters that are measured in ICT and industrial systems.
Data collection methods play a significant role for the overall research project thus method
selection should have a proper reasoning. There are several main methods collect data for
various scenarios: experiments, questionnaire, case study, observations, interviews,
language and text methods, etc. In our work, we rely on case study data collection that
corresponds to our research methodology and experiments that perfectly fits our approach
to evaluate our assumptions.
13
4 DEFINING QUALITY OF EXPERIENCE IN INDUSTRIAL
INTERNET OF THINGS
The vast number of concepts were proposed to get better understanding of quality in the
complex ICT systems. QoS, UX and QoE were introduced to analyze, measure or estimate
overall quality of the delivered service.
In the networking domain, QoS was proposed to gather insights and improve management
of communication network, the main goal is to define clear requirements for the offered
service in terms of network metrics [24]. QoS is well-established industrial and research
domain with the vast number of research works and standards [24, 25, 26]. Conventional
QoS mainly answers ‘what’-questions, for example, “what is the state of the network?” by
measuring typical network metrics (delay, jitter, bandwidth, etc.) and non-network
performance parameters (provision time, repair time, etc.) as shown in Fig. 4 [24].
Fig. 4. QoS components [24]
Conventional QoS models are not able to point at the root-cause of the quality degradation
and answer to ‘why’-questions, for instance, ‘why is the user unsatisfied with certain
service’. The user does not distinguish each network element but perceives the overall
service performance [27]. To understand the user perception, the concept of QoE was
proposed. ITU-T [27] defines QoE as: “The overall acceptability of an application or
service, as perceived subjectively by the end-user.” However, ITU-T also acknowledges
the following definition: “Quality of Experience includes the complete end-to-end system
effects (client, terminal, network, services infrastructure, etc.)”. QoE in Qualinet White
Paper [28] is defined as: “the degree of delight or annoyance of the user of an application
or service. It results from the fulfillment of his or her expectations with respect to the utility
and / or enjoyment of the application or service in the light of the user’s personality and
current state.”
14
However, classic QoE definition and model cannot be directly applied to IIoT domain due
to the number of challenges. The following list summarizes the presented studies by
identifying problems, research gaps and proposing future directions in the domain of
defining, modeling and evaluating QoE in the IIoT domain:
1. The vast number of applications in IIoT leads to the changes in quality
requirements. Unlike conventional multimedia services, where the network QoS
parameters (delay, jitter, bandwidth, etc.) are the major metrics that affect QoE,
IIoT involve a range of factors that can degrade QoE. For example, to fulfill safety,
productivity and efficiency requirements with IIoT, its architecture should provide
not only reliable network with low latency and jitter but also accuracy in sensing,
such as presence of various sensors (proximity, vibrations, etc.), microphones and
cameras, to maintain the context-awareness of the mining site. On the other hand,
industrial efficiency and productivity are incorporated into QoE definition and
become quality metrics as well. Therefore, the first goal is a definition of QoE
domain and application specific quality metrics for IIoT.
2. Deployment of Industry 4.0 and IoT solutions will lead to changes in requirements
for telecommunication providers, ISPs and ICT companies. Conventional approach
to ensure quality provided by communication network without considering user’s
equipment or industry-specific service requirements will no longer be acceptable by
stakeholders. For example, conventional QoE models typically map QoS metrics to
users’ experience while ignoring a performance evaluation of users’ hardware.
These gaps in performance evaluation will become more noticeable in the context
of remote control and e-health services, where the domain of the quality evaluation
is more complex. Moreover, some services like autonomous driving might require
complex data aggregation and processing services on-IoT device (from devices like
cameras, microphones, proximity sensors, etc.) while the quality of the network can
vary. As a result, the future IIoT services and applications will require the
telecommunication and ICT providers to change business model and extend their
business domains to cover entire product. In the same time, it will transform
understanding of QoE for all stakeholders.
15
3. Due to considerable number of IIoT applications, performing subjective
experiments every time will become expensive and time consuming. Since IIoT
redefines the role of the operator as an end user. Compared to traditional
multimedia applications complexity of subjective experiment for IIoT increases due
to the amount of metrics and complicated industrial environment (for example,
conducting such experiments at the mining site can be inconvenient or even
dangerous). From the business point of view, performing subjective experiments
results in financial loss due to the expenses spent on renting real industrial site and
equipment.
Considering the presented matters, one may conclude that QoE in IIoT is intended not only
to reflect the end-user, such as operators’ satisfaction with the tele-remote mining machine,
but also satisfy several industry-specific metrics and business goals. For instance, in the
tele-remotely controlled mining vehicles, the overall live-streamed video quality could be
degraded during the service run-time, but the productivity metric may still be high as long
as the end-user is able to complete the task effectively, without annoyance or discomfort.
Considering factors discussed above, we can introduce refined QoE domain for IIoT.
According to [29], QoE interaction model consists of Technological & Business domain
and QoE domain (Fig. 5).
16
Fig.5. Conventional QoE domain [29]
Traditional QoE model relies on human subjective aspects (evaluated using qualitative
metrics such as MOS), while authors of [29] proposed to consider objective human
cognitive factors. Objective human factors are quantitative and intended to predict human
performance.
Considering challenges that were mentioned, we can refine QoE domain for IIoT which is
shown in Fig. 6. It includes various human factors that are incorporated in QoE model in
combination with objective industrial factors that were mentioned previously. Objective
industrial factors consist of safety, efficiency, productivity requirements. They can be
rather general such as OEE or industry-specific such as ton/hour, ton/l (for mining
industry). By incorporating objective industrial factors into QoE domain, we can see the
tradeoff between industrial factors and Subjective and Objective human factors which
means that QoE should be evaluated by assessing all impact factors that are important for
the system in the given industrial context. The role of human entity changes from the
customer to employees prospective. Operator or driver as an employee has well-defined
task and corresponding skills that makes it different from conventional multimedia
systems. One the key factors that is seamlessly embedded into proposed domain is that
stakeholders earn revenue from providing services to customer but in industrial case
17
productivity and efficiency determine income for the stakeholders. That’s why satisfying
those requirements is more important for all players.
Fig. 6. QoE domain for IIoT
In our case, objective human factors are well-established research domain. Works on
remote manipulative control strategies started from 60s. Sheridan in [30] shows how
operators strategy changes with time delay. In [31], authors outline fundamental limitations
on remote operation. For example, multiple camera views can cause change blindness and
attention switching which can lead to mental workload and degraded performance. Time
delays essentially have negative effects on telepresence which is resulted in degradation on
accuracy and motion sickness. All factors that were mentioned above are due to operator’s
motor skills, mental models and physical limitations. As a result, QoE in IIoT can be
defined as objective satisfaction of main industry-specific efficiency metrics
and subjective acceptability of the end-user (operator, driver, etc.).
18
5 EVALUATING QUALITY OF EXPERIENCE IN INDUSTRIAL
INTERNET OF THINGS
The focus of this section is to explore various ways of QoE evaluation in IIoT by
examining the three pillars – Model, Measure and Predict, as described in [29].
Considering QoE domain that was defined in previouse section it is important to link
various stackeholders and connect technology and business entities. This will allow to take
into account all factors that might affect QoE in complex industrial scenarios.
5.1 QoE layered-model proposal
In view of the identified research gaps and directions for future development, discussed in
previous section, a layered-QoE model (Fig. 7) is proposed. Layered models were found
extremely suitable in the networking domain in order to describe interaction of various
protocols and procedures, such as the TCP/IP and OSI networking models [32], and later
applied in other areas, such as software engineering [33]. Layered QoE-model showed in
Fig. 7 is designed based on the proposed use-case, however, it is intended to be applicable
in other IIoT systems. This model enables an evaluation of QoE by separate assessment of
various qualities that are present in the system. Layered structure makes evaluation and
prediction simpler by considering different entities separately and applying different
evaluation methods.
Fig. 7. Proposed QoE model
19
5.1.1 Physical Layer
Physical layer is responsible for discovering and interacting with the physical objects and
entities in the environment. The IIoT as well as CPS consist of large number of sensing
devices (such as various sensors, cameras, microphones, etc.), actuators and control
devices. These physical objects often generate data with heterogeneous data types,
sampling frequencies and traffic types as a result it affects upper layers (network, service
layers). Due to the number of devices and their properties it is evident that they might
affect overall QoE of the system thus this layer outputs huge number of metrics and KPIs
for each device separately, such as types of produced data, sampling frequencies, sensing
features, etc. These metrics can be aggregated to output Quality of Data (QoD) or Quality
of Sensing. QoD is defined as high if it “fit for [its] intended uses in operations, decision
making and planning” [34]. Applicable to our model QoD shows how accurate is sensing
for particular industrial scenario which is usually derived from SLA and business
requirements.
5.1.2 Network Layer
Communication networks by nature exhibit performance and reliability limitations, caused
by the conveying data, various network failures, capacity, environmental conditions, etc.
[43]. Network Layer is responsible for complex evaluation of the underlying
communication networks by assessing the network performance in terms of conventional
parameters such as delay, jitter, bandwidth, etc. as suggested by ITU-T Y.1540 [46] and
IETF IP PMWG [47]. ITU-T G.1011 [48] is intended to provide more comprehensive
analysis of the encoded bitstream and audio/video streams. Typically, quality
measurements are not performed continuously on one-way or round-trip data transmissions
but executed during a well-defined time periods, that can be defined as observation
windows [49]. The duration of the observation window itself is a trade-off between
intrusiveness and granularity of obtained results. A larger observation windows increases
the probability of detecting the problems in the network, while a shorter observation
windows results in more detailed and accurate report on network state. For example, larger
20
observation windows have higher chances in detecting of unforeseen spikes in the end-to-
end delay that might be a result of route changes [50].
Different traffic types with varying quality requirements might depend upon various
techniques for evaluating network performance in IP-based services. ITU-T G.1030 [49]
proposes end-to-end performance estimations of IP applications by troubleshooting a
classical client-server connection.
5.1.3 Service Layer
Service Layer is intended to serve as a bridge between end-user entity and industrial
service by measuring objective and subjective factors showed in Fig. 6. A service is a
collection of independent applications and subsystems that interact by synchronizing on
time, events or by sharing resources and variables [35], which is often referenced in the
literature as System-of-systems (SoS) [36].
Evaluation of subjective end-user acceptability is performed with respect to various traffic
types that are present in the system. Such evaluation can be done by combining output
metrics from Physical and Network Layers. One of the most important evaluation metrics
is the service perception from the human-entity point of view. For instance, ITU-T
standard, P.863 [35] Perceptual Objective Listening Quality Assessment (POLQA), that is
used to evaluate QoE in VoIP applications or Perceptual Evaluation of Video Quality
(PEVQ), ITU-T Rec. J. 247 [36] are part of this evaluation. As described earlier,
evaluating QoE in a form of a MOS represents, or targets the user subjective perception
and satisfaction, however, it is not enough to evaluate IIoT service as a whole. Another
aspect of the service layer is evaluation of industrial scenario in terms of efficiency,
productivity and safety. This evaluation can be performed according to specific procedures
that are industry and domain specific. However, it is important to note that subjective
evaluation and objective industrial evaluation should be combined in order to get overall
QoE that can be used as a base for SLA between different stakeholders and industrial
players.
21
5.2 System’s architecture and types of traffic
Industrial use-case of remotely controlled vehicles offers a variety of traffic types and
patterns including critical real-time data. Moreover, tele-remote operation and control have
attracted significant attention of researchers in recent years [19, 31, 37]. The architecture of
the system is shown on the Fig. 2 and composes a remote-control station, remotely
operated mining vehicle and communication network (LTE and public Internet). Typical
control station for teleoperated vehicles includes:
• Displays showing the surrounding environment, typically includes multiple
cameras that provide certain field of view (FOV);
• Speakers that provide context awareness and give audio feedback;
• Sensor view that is responsible for displaying data regarding vehicle operation and
space around it;
• Control devices (joysticks, wheels, pedals, etc.)
• Health status of the vehicle and output of monitoring systems;
• Map that shows machine’s position to provide situation awareness and to facilitate
navigation [31];
As a result, typical traffic in the system includes:
• Video stream from multiple cameras;
• Audio stream from multiple microphones;
• Sensor stream from a monitoring system and various sensors;
• Control stream from remote control station back to the vehicle;
Each data stream can be evaluated separately according to the proposed model considering
impact on the end-user (operator, driver, engineer) and system’s performance.
5.3 Real-Time Sensor Stream
Sensor streams can be further split into two types: critical real-time control and non-critical
monitoring stream that should be differentiated in the requirements on reliability and end-
to-end latency. The purpose of non-critical monitoring stream is to periodically send
information regarding machine operation and surroundings. Typically, this data stream
does not carry critical information for the machine operation or safety. On the other hand,
22
critical real-time sensor stream is sensitive to delays and packet loss and carries
information that is necessary for safe tele-operation. In ideal case, delay and jitter for this
type of streaming should be upper-bounded following the principles of hard real-time
system. Typical examples of critical real-time sensor data in remote control of mining
vehicles are time-to-collision, speed and vehicle-specific parameters. For instance,
maximum working speed of the underground mining vehicles, such as load-haul dumper,
can vary up to 15-20 km/h, simple calculation shows that for 20 ms end-to-end delay and
relatively low speed of 11 km/h equals to approximately 6 cm of the displacement which
can be critical for this industrial scenario.
Real-time sensor streaming plays a significant role in industrial applications:
• It defines real-time strictness of the scenario and fundamental limitations on
communication technology, protocols and network architecture;
• It defines resolution and accuracy;
• It defines user-interaction models and impacts overall QoE;
• It enriches the user experience by complementing the video and audio streams. For
instance, reduced FOV, degraded depth perception and image quality result in
inability to estimate speed, time-to-collision, perception of objects, locations and
distance to obstacles, and the start of a sharp curve [38].
Typically, real-time sensor stream does not require significant throughput due to size of
data chunks. Time lag essentially has negative effects on telepresence which is resulted in
degradation on accuracy and motion sickness. For instance, latency that occurs in the
network can cause delays of the measurements from a speed sensor which might result in a
collision due to inability of the operator to estimate speed from video.
Studies on remote manipulative control strategies started in the 60s. Sheridan in [30] shows
how operators’ strategy changes with time delay. Normally, when communication latency
is about 1s, the driver’s strategy changes to “move and wait” one, instead of continuous
control. MacKenzie and Ware [39] demonstrated that movement times increased by 64%
and error rates increased by 214% when latency was increased from 8.3 to 225ms. Other
studies [37, 31, 40, 41] show that variable delay degrades the driving performance more
compare to constant delay even with higher magnitude. Unpredictability of time lag can
cause over-actuation (e.g. repeating control commands and over-steering) [31].
23
Performance of LTE network for remote driving was evaluated in [37] by testing the
possibility of tele-remote operation of a vehicle under the state-of-the-art commercial LTE
network conditions. However, authors have not discussed evaluation of critical sensor
communication.
5.4 Real-time performance evaluation and prediction
Real-time performance of the underlaying communication infrastructure is an integral part
of QoS delivered by a network, affecting the entire service QoE regarding subjective
perception, productivity and safety. Evaluation of real-time sensor stream is challenging
task due to the absence of standards. The absence of IIoT as a domain is evident in the QoS
classes recommended by ITU-T in Y-1541 [42]. Moreover, the ongoing work items within
ITU addressing data transmission quality techniques, such as G.OM_HEVC, P.NATS, and
G.vidmos [43, 44, 45] are not intended to assess critical real-time IIoT service. The cited
recommendations are targeting multimedia systems, which poses different quality
requirements in comparison to the IIoT. Also, utilizing round-trip time (RTT)
measurements might not be the most suitable techniques for IIoT due to the abundance of
installed sensors and actuators.
5.4.1 Background and network architecture
To tackle challenge of real-time performance evaluation, as a part of this study a real-time
tool (RTOOL) was designed and developed. The main goal of the designed setup is to
precisely estimate end-to-end and one-trip time (OTT) latency per transmission. The idea
is to collect radio measurements from the source-node for each transmission and further
use these metrics to predict the performance of the network using machine learning (ML)
model. Prediction can be used to calculate latency budgets for critical IoT, introduce
possibilities to reduce latency and perform root cause analysis. Fig. 8 gives a high-level
view of the communication network in our case. IIoT system consists of remotely operated
mining vehicle connected to the terminal (UE) in LTE RAN. The packet data network
gateway (P-GW) provides connectivity to the public IP network. Evolved packet system
(EPS), which is composed of LTE RAN and EPC, forms IIoT access network.
24
Fig. 8. Communication network architecture of an IIoT service
Each element of the communication system introduces variable delay due to the
connectivity procedures, scheduling and network fluctuations. Authors of [46] provide a
detailed overview of the latencies that can occur in LTE access domain. Connection
establishment on the control and user plane is the most significant part of the latency in
LTE access domain and can take up to 106 ms. Moreover, additional delays can be
introduced because of scheduling, retransmission and processing on User plane (up to 28
ms). However, these figures assume that the radio coverage is ideal and quality of the
signal is not degraded. In this paper, we are aiming to analyze the delay that can occur due
to the radio connectivity problems in LTE RAN. Evaluation of the control plane requires
access to the core networks and analysis on metrics such as routing diagnostics, queuing
length, load, bandwidth utilization, etc. Such quality metrics give insights to the
performance of the network, but licensed RANs are typically complex and hardly
accessible [47]. This means that performance evaluation of the control plane requires post-
processing, offline analysis on the gathered metrics. Therefore, the latency induced by the
control plane is out of scope for this study and the focus of the evaluation is on the user
data plane. The main reason is the real-time accessibility to the radio measurements at the
source-node, such as Received-signal-strength-indicator (RSSI), throughput, Signal-to-
interference+Noise-ratio (SINR), etc. The hypothesis under test is a real-time analysis on
the radio metrics in predicting network performance parameters, such as the absolute
delay, jitter and packet losses.
Delay figures of the IP backbone will vary depending on the region, network load and
number of hopes between P-GW and application server. For instance, delay can vary from
15 ms up to 150 ms in Europe [46]. In this work, we assume that application server (e.g.
remote control station) is placed close to the P-GW and this delay is negligible.
25
Evaluating LTE RAN from an end-node perspective is a challenging task that requires
special tools and devices to capture network events and measure radio parameters. One of
the commercially available products is TEMS Pocket [48]. It is described as a state-of-the-
art phone-based test tool developed for measuring the performance and quality parameters
of wireless networks. Main functionality of TEMS Pocket are following:
• A real-time radio measurements and event data collection;
• Indoor and outdoor testing of wireless networks;
• Drive testing capabilities with positioning;
• Capturing network data for post processing using other TEMS-ecosystem tools
such as TEMS Discovery;
Using TEMS Pocket limits the scope of implementation options which means that RTOOL
should be implemented on commercial mobile phone or tablet under Android OS. Devices
that are supported by TEMS Pocket are: Sony, HTC, LG, Samsung. TEMS Pocket
supports following mobile technologies: LTE, WCDMA/HSDPA/HSUPA,
GSM/GPRS/EDGE, CDMA/EV-DO. Moreover, TEMS Pocket has several control
functions to modify device’s behavior in LTE network. Control functions work in real-time
and allow to perform quick and non-intrusive tests. The following control functions are
available in TEMS Pocket:
• Radio Access Technology (RAT) lock (LTE/WCDMA/GSM; CDMA/EV-DO);
• Band lock (LTE/WCDMA/GSM);
• LTE EARFCN lock; EARFCN/PCI lock;
• WCDMA cell lock (UARFCN, UARFCN + SC);
• GSM cell lock/prevent (ARFCNs);
• Access class lock;
5.4.2 Time synchronization
Evaluation of real-time performance requires precise estimation of end-to-end delay
between control center and tele-remote vehicle. This task can be achieved by having two
nodes perfectly synchronized with each other. Synchronization of the devices in the
network is a complex task that can be done in two main ways:
26
• Synchronization over the network (using various protocols and services);
• Synchronization using external clock references (using signals from Global
Navigation Satellite System (GNSS), atomic clocks, etc.);
Convenient approaches to synchronize two nodes utilize external references, such as GNSS
signals from Global Positioning System (GPS), GLONASS, GALILEO or COMPASS. It
is possible to provide accurate time synchronization typically better than 100 nanoseconds
to UTC [49]. However, due to the inability to receive GNSS signals in underground
environments, such technique is not suitable for the mining industry, thus synchronization
for our case-study may be only achieved using existing LTE network. Main protocols
which were developed to keep nodes over the network synchronized are Network Time
Protocol (NTP) and Precision Time Protocol (PTP) known as IEEE Standard 1588-2008
[50].
NTP is de-facto time-keeping standard across the Internet [51]. NTP organizes clocks in
layered hierarchal way in terms of a stratums. The stratum level specifies the distance
between reference clock and time server which is used for synchronization. As a result,
accuracy of the synchronization that can be achieved using NTP is less than 1ms in LAN
and in order of 10ms over WAN [51]. Typical NTP clock hierarchy is shown in Fig. 9.
Stratum 0 consists of high-precision atomic clocks, GPS receivers or other similar sources.
Stratum 1 represents a number of time servers that are connected to stratum 0 devices.
Stratum 2 servers are typically attached to several servers from stratum 1 and correct their
clocks using NTP algorithm. Stratum 3+ devices are connected to several stratum 2 servers
and with each other. NTP algorithm is based on estimation round-trip time (RTT) between
two nodes by sending 64-bits timestamps using UDP as a Transport layer protocol. The 64-
bits timestamp is composed of two parts: 32 bits for seconds and 32 bits for fractional
seconds. Compensation of the offset on the client’s side is performed by measuring RTT to
NTP server. The crucial assumption that NTP makes at this step is that the link is
symmetrical and in ideal case uplink and downlink delays are equal.
Another example of clock synchronization protocols is Precision Time Protocol (PTP),
which was initially developed by IEEE to provide more accurate synchronization compare
to NTP. Better accuracy is achieved by using PTP aware switches, also known as PTP
Transparent Switches or Boundary Clocks. PTP takes into account switching and queueing
27
delays by using PTP aware switches and routers. Master-Slave topology is utilized in PTP.
However, PTP makes the same crucial assumption as NTP that link is symmetrical, and
that switches and routers have PTP enabled which is not usually the case.
Fig. 9. NTP clock hierarchy [51]
For the purpose of the described case-study, a modified version of NTP is developed. A
slightly changed topology is used since absolute synchronization to UTC time is not
needed as long as NTP server and RTOOL are synchronized to each other. In conventional
NTP topology, synchronized nodes to the NTP servers will have different clock errors due
to the network fluctuations and clock offsets on reference NTP servers. The proposed
solution is to implement stand-alone NTP server and have its clock as a reference time for
the entire system, in this case we mitigate error on one side completely since
synchronization error on our server is equal to zero. Fig. 10 shows proposed topology for
experiment using NTP-based synchronization. RTOOL synchronizes phone’s hardware
clock to NTP server adaptively which means that synchronization can be done only under
excellent/good radio conditions (Table I).
Fig. 10. Experimental synchronization setup
28
Table I. LTE signal quality
Signal Quality RSRP, dBm RSRQ, dB CQI
Excellent > -90 > -9 > 10
Good -90 … -105 -9 … -12 9 … 7
Fair -106 … -120 < -13
6 … 1
Poor < -120 0
It is important to outline that LTE wireless link is not symmetrical due to the differences in
uplink and downlink radio technologies, scheduling mechanisms and bandwidth. However,
measurements showed that clock error was acceptable for our scenario and provides best
effort that can be achieved in this specific use-case. Clock error was measured on the live
network of two mobile operators by connecting phone directly to the NTP server using
Android Debug Bridge (adb). The results are shown in Fig. 11.
a) b)
Fig. 11. Synchronization accuracy for a) TELE2, b) TELIA
5.4.3 Experimental setup
RTOOL is intended to mimic the sensor stream sent from real tele-operated machine.
Logical components of designed tool are illustrated in Fig. 12. The software tool consists
of following components: RTOOL Core, Adaptive NTP client for synchronization, logging
system for post-processing and simple User Interface (UI). RTOOL Core has many
configurable parameters for each logical element:
29
• Data Generator mimics a real sensor and generates sensor data with specific size,
type and format;
• Sampler is a crucial element that specifies the sampling period which determines
how often sensor data is sent based on application requirements;
• Encoder encrypts sensor data and formats it according to the requirements;
• Multiplexer combines the data from several sensors into one stream;
• Socket is used to send sensor data from the phone to the server using specified
protocol (UDP);
Fig. 12. RTOOL architecture
End-to-end or OTT latency measurements refer to the time it takes to send a packet from
the source-node until it is received at the end-node. These measurements are done by
periodically sending UDP packets with a sensor’s payload from RTOOL to the server. The
experiment is performed using different time periods (sampling rate) to send the data. Each
UDP packet is time-stamped at RTOOL and the server using local clocks that are
synchronized to each other. Time-stamps and network measurements from TEMS Pocket
are stored at log files for offline analysis.
30
5.4.4 Latency prediction
As was mentioned earlier, RTOOL is intended to predict delay figures from the network
measurements provided by TEMS Pocket. In recent years, machine learning (ML) models
were successfully used in various applications from bioinformatics to speech and image
recognition [52]. ML tries to construct data-driven models that can capture complex and
sometimes hidden dependencies. The recent developments of hardware (e.g.,
computational devices like GPU and TPU) and software (ML libraries like Tensorflow and
Scikit-Learn) and distributed data processing frameworks (e.g., Hadoop and Spark) enable
opportunities to unleash true power of machine learning for solving complex problems in
networking domain. The task of real-time performance prediction from the network
measurements perfectly fits into the ML approach. Authors of [53] provide a general
workflow (Fig. 13) that can be used to build ML model for predicting network
performance.
Fig. 13. Typical workflow of ML for networking [53]
1) Problem formulation. The first step of the workflow is a problem formulation. As
was stated earlier the goal is to predict network delay caused by the radio
environment and RAN in LTE. For the second step, RTOOL and TEMS Pocket
will be used to collect timestamps, network measurements and IP streams.
2) Data processing and Feature extraction. As described earlier, the latency is
generated by several fac-tors, but only few of them (i.e., features) have the most
effect on the target metric. The goal of the third step is to preprocess the data by
cleaning, formatting and performing feature engineering. Feature engineering is the
major step of the entire process of ML model creation. Better features utilization
31
enables simpler models and produce improved results. Measured network
parameters and end-to-end delay are presented as time-series data. There are
several ways to create new features from time-series data:
- Lag features that represent measured values of network metrics at prior time
samples;
- Window features that represent a value obtained from the values over a fixed
window of prior time;
TEMS Pocket may collect more than 1500 parameters (i.e., features), but for
latency prediction basic L1 radio measurements were utilized. These measurements
are extensively described in 3GPP standards and detailed description of LTE L1
measurements is beyond the scope of this work. Parameters such as RSRP, RSRQ,
RSSI, SINR and physical throughput were considered as features for the model
construction.
Collected data have various features with values in different ranges. Most of the
ML algorithms are sensitive to features’ scaling. All features and label (i.e. delay)
were scaled using Standard scaler [54].
3) Model construction. The goal of the model construction step is to select
appropriate ML algorithm to get reliable model with the best prediction. In this
paper, we evaluated different ML-regressors: Artificial Neural Networks (MLP)
and Decision Tree Regressor. Another powerful ML technique to get better
prediction is model ensembling. In this work, we propose to ensemble models using
bagging technique. Decision Tree Regressor was used as a base for bagging
ensembling [55]. All regressors were trained using the training set and evaluated
against the testing set. Models were assessed by evaluating the accuracy of the
delay prediction. Results for each sampling period are shown in Table II.
Performance varies with a sampling rate due to the amount of radio measurements
and events collected within a timeseries. Lower sampling periods allow to monitor
network with higher resolution and capture all fluctuations of the radio
environment.
32
Table II. Prediction performance [56]
R2 (coefficient of determination) Mean Absolute Error (MAE)
Sampling
period
NN
(MLP)
Decision
Tree
Bagging
Decision
Tree
NN
(MLP)
Decision
Tree
Bagging
Decision
Tree
20 ms 82 % 82.2 % 90.7% 0.23 0.11 0.091
50 ms 75.5 % 77.7 % 85.1 % 0.28 0.16 0.15
100 ms 73.7 % 67.3 % 81.8 % 0.29 0.19 0.13
200 ms 60% 50 % 66.8 % 0.35 0.31 0.22
a)
33
b)
c)
Fig. 14. Comparison of true delay and predicted delay figures by Bagging Decision Tree
regressor, a) sampling rate 20 ms; b) sampling rate 50 ms; c) sampling rate 100 ms
34
Obtained results show the possibility to predict latency on the user plane by collecting
basic RAN measurements and utilizing ML techniques. In ideal case, such prediction
should be part of the benchmarking process in order to verify real-time performance of the
underlaying communication network. Moreover, output (delay figures) of the proposed
model can be used as an input for further analysis in the model that was proposed in
Section VI. It is important to note that proposed solution has low intrusiveness considering
size of data packets.
35
6 SUSTAINABILITY ASPECTS
As was stated in motivation section, the goal of this work is to examine implementation
tele-remote operation of mining vehicles that will lead reduce need of the human operator
in the harsh environment of the underground mine. This will make a dramatical change on
perception of mining industry from sustainability point of view. In addition, proposed QoE
domain and evaluation model might speed up implementation of IIoT services in mining
industries. Moreover, presented evaluation methods will make operation of such services
more sustainable. The notion of sustainability is broadly acknowledged by researchers,
official institutions and governments. The term ‘sustainability’ was coined by Gro Harlem
Brundtland from the World Commission on Environment and Development in 1983. The
report titled “Our common future” defined sustainability as economic-development that
“meets the needs of current generations without compromising the ability of future
generations to meet their own needs” [57]. This fairly abstract definition was widely
explained and extended. For instance, Erek et al. further explain sustainability as “a
survival assurance meaning that an economical, ecological or social system should be
preserved for future generations and, thus, necessary resources should only be exploited to
a degree where it is possible to restore them within a regeneration cycle” [58]. Some have
argued that ideas of sustainability were proposed by Thomas Malthus in “An Essay on the
Principle of Population” late in eighteenth century. He noticed that exponential population
growth would exceed Earth’s capability to support human beings. Sustainability from this
point of view is not only conservation or preservation of natural resources but attempt to
find some sort of balanced state so the Earth and human population can support economic
and social transformation without catastrophic effects on next generations, animals, plants,
etc. As a result, every domain of human activity can have diverse ways of implementing
sustainability principles and methods [59].
Sustainability includes three principal components often referred as the three E’s or pillars
of sustainability [57]:
• Economy;
• Environment;
• Equity;
36
The main concern is that sustainability can be accomplished only by protecting the
environment while providing economic growth and development and encouraging equity.
Fig. 15 shows that main elements of sustainability are overlapping and should be treated
simultaneously without sacrificing any components.
Fig. 15 - Three elements of sustainability
Historically, mining industry was considered as a polluting and environmentally hostile.
The mining and mineral industry deals with some of the most difficult sustainability
challenges of almost any industry. However, recent technological advances and
digitalization incorporate sustainability principles into business activities. For example,
vast number of indicators are used to ensure compliance with sustainability principles.
These monitoring indicators are shown in Table III [60, 61]. Also, main sustainability
issues have been identified by the European Commission [62] and the MMSD project [63].
Table III. Sustainability indicators in the mining industry.
Economic Contribution Social Aspects and Equity Environmental protection
Performance of shares;
Net earnings;
Percent return on capital
investment;
Amount of materials
produced;
Taxes and royalties
created;
Capital expenditures;
Lost time injury frequency;
Injuries requiring medical
treatment;
Regular medical surveillance
of employees;
Occupational diseases;
Ranking of company as
desirable place to work;
Diversity of workers;
Control harmful emissions
from equipment;
Minimize deforestation for
new mines;
Control use of ozone creating
gases and sprays;
Use low sulfur diesel and
fuels;
Economy
EquityEnvironment
37
Wages;
Workforce and training;
Treat emissions to minimize
nitrogen oxides;
minimize use of benzene and
volatiles;
reduce flaring of natural gas;
minimize methane releases
into the air;
eliminating spills of
hazardous substances;
minimizing water used;
reduce power usage;
recycle products and waste;
The case study of this work investigates digitalization in the mine by introducing IIoT use
case. It is important to understand how modern ICT systems can influence sustainability
performance in the mining sector. Remotely controlled mining vehicles can drastically
change mining sector and affect all aspects of sustainability.
It was shown that the mining productivity declined over past decade (Fig. 16).
38
Fig. 16. Productivity of the mining industry [64].
Adoption of digital technologies and modern ICTs has the potential to enhance
productivity and efficiency. Matthias Breunig et al. showed main clusters of technologies
that will shift industrial paradigm within Industry 4.0 transformations [64]:
• Data, computational power, and connectivity.
• Analytics and intelligence.
• Human–machine interaction.
• Digital-to-physical conversion.
From the social point of view, removing human operators from the mine will reduce or
completely negate injuries and occupational diseases. Moreover, it will change ranking of
the mining companies as a desirable place to work and encourage diversity of workers.
Also, removing operators from the mine can also decrease fuel consumption and efficiency
by enabling assisted mining vehicle operation and consequently decrease carbon footprint
from the industry. According to [65], the operator is the main parameter, that affects the
fuel efficiency and productivity the most.
39
7 FUTURE WORK
Evaluating QoE in IIoT is a complex task that involves many stakeholders and players.
The contribution of this work is a proof of the concept that end-to-end latency in mobile
communications can be predicted with a certain degree of accuracy by utilizing very basic
RAN measurements. Next steps of this work can be complex evaluation of IIoT system by
combining all traffic types and application scenarios. Another aspect that was not
mentioned in this work is control stream that has similar characteristics as discussed
critical sensor stream. However, it operates on DL and uses different radio technologies,
scheduler and RAN procedures.
Delay figures that are produced by RTOOL can be further transformed to QoE of IIoT
service that will allow stakeholder to view application in the domain of efficiency,
productivity and safety.
As was mentioned earlier, video/audio quality evaluation methods are extremely important
in IIoT scenarios which means that video/audio quality assessment should be simple, non-
intrusive and precise. Developed software solution allows incorporating other traffic types
and complex quality assessment of IIoT service.
40
8 CONCLUSION
The research community and the industry acceptance of IIoT suggests rapid digitalization
of industrial processes. Being applied in various domains, each IIoT service requires
prioritization of different KPIs and service requirements. This work dives into the
complexity of assessing the QoE in IIoT domain, by identifying the gaps throughout a
thorough research on the subject. By reflecting on the identified research challenges for
future development, this paper triggers the discussion on defining the QoE domain for IIoT
services and application, covering the business aspects. As a result, a QoE layered-model is
proposed, which as an outcome predicts the QoE of IIoT services and applications in a
form of pre-defined Industrial KPIs, such as productivity, efficiency, reliability and safety.
Moreover, the network performance evaluation becomes linearly more complex as each
IIoT requires different QoS assurances. In this work, real-world industrial scenario was
analyzed to evaluate importance of critical real-time sensor streaming. For this purpose, a
software tool was developed to capture the absolute, one-way delay for each transmission.
The latency metrics and further analyzed with various LTE RAN and radio measurements.
A machine learning technique is used to grasp the relation between the latency metrics and
the captured radio measurements. The contribution of this study is a delay prediction for
each transmission in real-time based on the correlation and learning processes. The initial
results prove the possibility to estimate delay figures caused by the LTE RAN events and
radio disturbances from the environment. The highest accuracy of the prediction is
estimated at 90%. The approach taken in this study is the first step in assessing the
performance of IIoT service. The achieved results enable further calculation of latency
budgets for a given critical IoT service, as well as opens the possibilities to reduce latency
and perform root cause analysis [56].
41
9 REFERENCES
[1] H. Kagermann, W. Wahlster ja J. Helbig, ”Recommendations for implementing the strategic
initiative Industrie 4.0: Final report of Industrie 4.0 Working Group,” Franfurt, 2013.
[2] Ericsson, ”Manufacturing reengineered: robots, 5G and Industrial IoT,” Ericsson Business
Review, Stockholm, 2015.
[3] European Comission, ”Implementation of an Industry 4.0 Strategy - The German Plattform
Industrie 4.0,” 25 January 2017. [Online]. Available: https://ec.europa.eu/digital-single-
market/en/blog/implementation-industry-40-strategy-german-plattform-industrie-40. [Haettu 5
May 2018].
[4] PWC, ”Industry 4.0: Building digital enterprise,” 2016. [Online]. Available:
www.pwc.com/industry40. [Haettu 5 May 2018].
[5] M. Wollschlaeger, T. Sauter ja J. Jasperneite, ”The Future of Industrial Communication:
Automation Networks in the Era of the Internet of Things and Industry 4.0,” IEEE Industrial
Electronics Magazine, osa/vuosik. 11, nro 1, pp. 17-27, 2017.
[6] RISE, ”PIMM Digitalised Mining Arena,” Software and Systems Engineering Laboratory (SSE),
RISE SICS Västerås, 2017. [Online]. Available: https://www.sics.se/projects/pimm. [Haettu 7
May 2018].
[7] L. Pierucci, ”The quality of experience perspective toward 5G technology,” IEEE Wireless
Communications, osa/vuosik. 22, nro 4, pp. 10-16, 2015.
[8] Ericsson, ”Ericsson Mobility Report,” Ericsson, Stockholm, 2017.
[9] L. D. Marez ja K. D. Moor, ”The Challenge of User-And QoE-Centric Research and Product
Development in Today's ICT-Environment,” Observatorio, osa/vuosik. 1, nro 3, 2007.
[10] Y. Wang ja e. al, ”A Data-Driven Architecture for Personalized QoE Management in 5G
Wireless Networks,” IEEE Wireless Communications, osa/vuosik. 24, nro 1, pp. 102-110, 2017.
[11] Q. Wu ja e. al, ”Cognitive Internet of Things: A New Paradigm Beyond Connection,” IEEE
Internet of Things Journal, osa/vuosik. 1, nro 2, pp. 129-143, 2014.
42
[12] A. Floris ja L. Atzori, ”Quality of Experience in the Multimedia Internet of Things: Definition
and practical use-cases,” tekijä: IEEE International Conference on Communication Workshop
(ICCW), London, 2015.
[13] Y. Ikeda, S. Kouno, A. Shiozu ja K. Noritake, ”A framework of scalable QoE modeling for
application explosion in the Internet of Things,” tekijä: IEEE 3rd World Forum on Internet of
Things (WF-IoT), Reston, VA, 2016.
[14] J. Aráuz ja T. Fynn-Cudjoe, ”Actuator quality in the Internet of Things,” tekijä: IEEE
International Workshop of Internet-of-Things Networking and Control (IoT-NC), New Orleans,
2013.
[15] Accenture, ”Industrial Internet Insights Report,” 2014. [Online]. Available:
https://www.accenture.com/t20150523T023646Z__w__/us-en/_acnmedia/Accenture/Conversion-
Assets/DotCom/Documents/Global/PDF/Dualpub_11/Accenture-Industrial-Internet-Insights-
Report-2015.pdf. [Haettu 7 May 2018].
[16] H. Merry, ”5 benefits IoT is having on the mining industry,” 8 March 2017. [Online]. Available:
https://www.ibm.com/blogs/internet-of-things/mining-industry-benefits/. [Haettu 7 May 2018].
[17] I. Som, ”The Internet Of Things And Innovation In Mining,” 4 October 2017. [Online].
Available: http://www.digitalistmag.com/iot/2017/10/04/internet-of-things-innovation-in-mining-
05384218. [Haettu 7 May 2018].
[18] Rio Tinto, ”Mine of the Future™,” Rio Tinto, [Online]. Available:
http://www.riotinto.com/australia/pilbara/mine-of-the-future-9603.aspx. [Haettu 7 May 2018].
[19] C. Zhou, N. Damiano, B. Whisner ja M. Reyes, ”Industrial Internet of Things: (IIoT) applications
in underground coal mines,” Mining Engineering, osa/vuosik. 12, pp. 50-56, 2017.
[20] A. Håkansson, ”Portal of Research Methods and Methodologies for,” tekijä: International
Conference on Frontiers in Education: Computer Science and Computer Engineering, Las
Vegas, 2013.
[21] M. D. Myers, Qualitative Research in Business & Management, Thousand Oaks: SAGE
Publications Ltd , 2008.
[22] N. J. Salkind, Exploring Research, Upper Saddle River, N.J.: Pearson Prentice Hall, 2006.
[23] R. K. Yin, Case Study Research, Design and Methods, SAGE Publications Ltd, 2008.
43
[24] ITU-T, ”Recommendation E.800,” 30 April 2009. [Online]. Available: https://www.itu.int/rec/T-
REC-E.800-200809-I/en. [Haettu 8 May 2018].
[25] C. Xiuzhong, W. Chunfeng , X. Dong, L. Zhongcheng, M. Yinghua ja Z. Wei, ”Survey on QoS
management of VoIP,” tekijä: International Conference on Computer Networks and Mobile
Computing, Shanghai, 2003.
[26] H. J. Kim ja S. G. Choi, ”A study on a QoS/QoE correlation model for QoE evaluation on IPTV
service,” tekijä: The 12th International Conference on Advanced Communication Technology
(ICACT), Phoenix Park, 2010.
[27] ITU-T, ”G.100 : Definitions used in Recommendations on general characteristics of international
telephone connections and circuits,” November 2017. [Online]. Available:
https://www.itu.int/rec/T-REC-G.100/en. [Haettu 8 May 2018].
[28] K. Brunnström, S. A. Beker, K. D. Moor, A. Dooms, S. Egger ja e. al., ”Qualinet White Paper on
Definitions of Quality of Experience,” 2013. [Online]. Available: https://hal.archives-
ouvertes.fr/hal-00977812/document. [Haettu 8 May 2018].
[29] K. u. R. Laghari, N. Crespi, B. Molina ja C. E. Palau, ”QoE Aware Service Delivery in
Distributed Environment,” tekijä: IEEE Workshops of International Conference on Advanced
Information Networking and Applications, Biopolis, 2011.
[30] T. B. Sheridan ja W. R. Ferrell, ”Remote Manipulative Control with Transmission Delay,” IEEE
Transactions on Human Factors in Electronics, Osa %1/%2HFE-4, nro 1, pp. 25-29, 1963.
[31] J. Y. C. Chen, E. C. Haas ja M. J. Barnes, ”Human Performance Issues and User Interface Design
for Teleoperated Robots,” IEEE Transactions on Systems, Man, and Cybernetics, Part C
(Applications and Reviews), osa/vuosik. 37, nro 6, pp. 1231-1245, 2007.
[32] ISO/TC97/SC16/DIS7498, ”Data Processing - Open Systems Interconnection - Basic Reference
Model”.
[33] J. A. Rolia ja K. C. Sevcik, ”The Method of Layers,” IEEE Transactions on Software
Engineering, osa/vuosik. 21, nro 8, pp. 689-700, 1995.
[34] T. C. Redman, Data Driven: Profiting from Your Most Important Business Asset, Harvard
Business Press, 2013.
[35] S. Jeschke, C. Brecher, H. Song ja D. B. Rawat, Industrial Internet of Things:
Cybermanufacturing Systems, Publisher Name, 2017.
44
[36] M. Jamshidi ja A. P. Sage, System of Systems Engineering: Innovations for the 21st Century,
Wiley, 2008.
[37] R. Liu, D. Kwak, S. Devarakonda, K. Bekris ja L. Iftode, ”Investigating Remote Driving over the
LTE Network,” tekijä: The 9th International Conference on Automotive User Interfaces and
Interactive Vehicular Applications (AutomotiveUI '17) ACM, New York, 2017.
[38] J. Bao, Z. C. Wu, J. J. Wu, P. P. Pei ja C. Liu, ”The Impact of Image Parameters of Indirect
Viewing Systems on Driver Fatigue and Task Performance,” Advanced Materials Research, Osa
%1/%2765-767, pp. 2046-2050, 2013.
[39] S. MacKenzie ja C. Ware, ”Lag as a determinant of human performance in interactive systems,”
tekijä: Conference on Human factors in computing systems (INTERCHI '93), Amsterdam, 1993.
[40] J. C. Lane, C. R. Carignan, B. R. Sullivan, D. L. Akin, T. Hunt ja R. Cohen, ”Effects of time
delay on telerobotic control of neutral buoyancy vehicles,” tekijä: IEEE International Conference
on Robotics and Automation, Washington DC, 2002.
[41] B. Watson, N. Walker, W. Ribarsky ja V. Spaulding, ”Effects of Variation in System
Responsiveness on User Performance in Virtual Environments,” Human Factors, osa/vuosik. 40,
nro 3, p. 403 – 414, 1998.
[42] ITU-T, ”Recommendation ITU-T Y-1541,” [Online]. Available: https://www.itu.int/rec/T-REC-
G.1030/en. [Haettu 8 May 2018].
[43] ITU-T, ”Recommendation P.1203.1,” October 2017. [Online]. Available:
https://www.itu.int/rec/T-REC-P.1203/en. [Haettu 8 May 2018].
[44] ITU-T, ”Recommendation G.1071,” November 2017. [Online]. Available:
https://www.itu.int/rec/T-REC-G.1071/en. [Haettu 8 May 2018].
[45] ITU-T, ”Recommendation P.913,” March 2016. [Online]. Available: https://www.itu.int/rec/T-
REC-P.913/en. [Haettu 8 May 2018].
[46] N. Nikaein ja S. Krea, ”Latency for Real-Time Machine-to-Machine Communication in LTE-
Based System Architecture,” tekijä: 17th European Wireless 2011 - Sustainable Wireless
Technologies, Vienna, 2011.
[47] ETSI, ”ETSI TR 136 912: Feasibility study for Further Advancements for E‐UTRA (LTE‐
Advanced),” September 2014. [Online]. Available:
45
http://www.etsi.org/deliver/etsi_tr/136900_136999/136912/09.01.00_60/tr_136912v090100p.pdf.
[Haettu 8 May 2018].
[48] InfoVista, ”TEMS Pocket,” [Online]. Available: https://www.infovista.com/tems/products-for-
radio-and-core-networks/radio-network-engineering/portable-testing-for-wireless-networks.
[Haettu 8 May 2018].
[49] Endace, ”IEEE 1588 PTP clock synchronization over a WAN backbone,” Endace Technology
Limited.
[50] IEEE, ”IEEE Standard for a Precision Clock Synchronization Protocol for Networked
Measurement and Control Systems,” IEEE Std 1588-2008 (Revision of IEEE Std 1588-2002), pp.
1-300, 2008.
[51] P. Corcoran, ”A Matter of Timing: Consumer Electronics and Network Time,” IEEE Consumer
Electronics Magazine, osa/vuosik. 2, nro 4, pp. 20-25, 2013.
[52] I. Goodfellow, Y. Bengio, A. Courville ja Y. Bengio, Deep learning (Vol. 1), Camridge: MIT
press, 2016.
[53] M. Wang, Y. Cui, X. Wang, S. Xiao ja J. Jiang, ”Machine Learning for Networking: Workflow,
Advances and Opportunities,” IEEE Network, osa/vuosik. 32, nro 2, pp. 92-99, 2018.
[54] scikit-learn developers, ”StandardScaler,” [Online]. Available: http://scikit-
learn.org/stable/modules/generated/sklearn.preprocessing.StandardScaler.html. [Haettu 10 May
2018].
[55] D. S. Clifton, ”Classification and Regression Trees, Bagging, and Boosting,” Handbook of
Statistics, osa/vuosik. 24, pp. 303-329, 2005.
[56] R. Zhohov, D. Minovski, P. Johansson ja K. Andersson, ”Real-time performance evaluation of
LTE for IIoT,” tekijä: IEEE Conference on Local Computer Networks (LCN) (accepted),
Chicago, 2018.
[57] World Commission on Environment and Development, Our common future, Oxford, New York:
Oxford University Press, 1987.
[58] K. Erek, N.-H. Schmidt, R. Zarnekow ja L. Kolbe, ”Sustainability in Information Systems:
Assortment,” tekijä: Americas Conference on Information Systems, 2009.
46
[59] A. Klimova, E. Rondeau, K. Andersson, J. Porras, A. Rybin ja A. Zaslavsky, ”An international
Master's program in green ICT as a contribution to sustainable development,” Journal of Cleaner
Production, osa/vuosik. 135, pp. 223-239, 2016.
[60] K. Fytas, ”Sustainable development in the Canadian mineral industry,” tekijä: Sustainable
Development Indicators in the Mineral Industries, Milos, 2003.
[61] A. Azapagic, ”Developing a framework for sustainable development indicators for the mining
and minerals industry,” Journal of Cleaner Production, osa/vuosik. 12, nro 6, p. 639–662, 2004.
[62] ”Communication from the Commission - Promoting sustainable development in the EU non-
energy extractive industry,” European Union, Brussels, 2000.
[63] International Institute for Environment and Development, ”Mining, Minerals and Sustainable
Development (MMSD),” [Online]. Available: https://www.iied.org/mining-minerals-sustainable-
development-mmsd. [Haettu 7 May 2018].
[64] McKinsey & Company, ”Industry 4.0 How to navigate digitization of the manufacturing sector,”
McKinsey Digital, 2015. [Online]. Available: http://worldmobilityleadershipforum.com/wp-
content/uploads/2016/06/Industry-4.0-McKinsey-report.pdf. [Haettu 7 May 2018].
[65] B. Frank, L. Skogh, R. Filla ja M. Alaküla, ”On increasing fuel efficiency by operator assistant
systems in a wheel loader,” tekijä: International Conference on Advanced Vehicle Technologies
and Integration (VTI 2012), Changchun, 2012.
[66] A. Bryman ja E. Bell, Business Research Methods, Oxford: Oxford University Press, 2007.
[67] L. D. Xu, W. He ja S. Li, ”Internet of Things in Industries: A Survey,” IEEE Transactions on
Industrial Informatics, osa/vuosik. 10, nro 4, pp. 2233-2243, 2014.