SPECIAL SECTION ON DATA-DRIVEN MONITORING, FAULT DIAGNOSIS

AND CONTROL OF CYBER-PHYSICAL SYSTEMS

Received July 24, 2018, accepted August 9, 2018, date of publication August 23, 2018, date of current version September 21, 2018.
Digital Object Identifier 10.1109/ACCESS.2018.2866403

Data-Driven Monitoring and Safety Control of

Industrial Cyber-Physical Systems:
Basics and Beyond
YUCHEN JIANG 1 , (Student Member, IEEE), SHEN YIN 1, (Senior Member, IEEE),
AND OKYAY KAYNAK 1,2 , (Fellow, IEEE)
1 Harbin Institute of Technology, Harbin 150001, China
2 Boğaziçi University, 34342 Istanbul, Turkey
Corresponding author: Shen Yin ([email protected])
This work was supported by the National Natural Science Foundation of China under Grant 61873073.

ABSTRACT Industrial cyber-physical systems (ICPSs) are the backbones of Industry 4.0 and as such,
have become a core transdisciplinary area of research, both in industry and academia. New challenges
brought about by the growing scale and complexity of systems, insufficient information exchange, and the
exploitation of knowledge available have started threatening the overall system safety and stability. This work
is motivated by these challenges and the strategic and practical demands of developing ICPSs for safety-
critical systems such as the intelligent factory and the smart grid. It investigates the current status of research
in ICPS monitoring and control, and reviews the recent advances in monitoring, fault diagnosis, and control
approaches based on data-driven realization, which can take full advantage of the abundant data available
from past observations and those collected online in real time. The practical requirements in the typical ICPS
applications are summarized as the major issues to be addressed for the monitoring and the safety control
tasks. The key challenges and the research directions are proposed as references to the future work.

INDEX TERMS Cyber-physical system (CPS), data-driven, system monitoring, fault diagnosis, smart grid,
plug-and-play control.

I. INTRODUCTION As a result, feasibility analysis, robustness analysis, per-

In the era of Industry 4.0, the cyber-physical system (CPS) formance evaluation, and performance optimization of the
has become the central object of investigation in both indus- ICPS monitoring and control strategies become an impera-
trial and academic domains [1]. From the viewpoint of system tive requirement to be investigated in an overall systematic
architecture, a bridge is built between the virtual and the level.
physical dimensions. As a transdisciplinary topic, CPS is In addition to interpreting the abstract concept of ICPS,
receiving extensive research interests covering a wide spec- this review will discuss the recent advances based on the
trum of subjects such as industrial design, industrial technolo- practical challenges in the design of ICPS monitoring and
gies, computer science, electrical engineering and so forth. safety control systems and the future horizons.
While cyber-physical systems commonly exist in both the Figure 1 shows the trend of research focus in recent
industrial domain and in human daily life, this review studies years. Indexed by the Web of Science core collection,
the advances in industrial CPS (ICPS), with special focus on the statistics of the past five years illustrate the high qual-
novel approaches to system monitoring, fault diagnosis and ity research literature under the themes of cyber-physical
control. system (CPS), industrial CPS (ICPS), Industry 4.0, Internet
Typical examples of industrial CPS include the smart grid, of Things (IoT), as well as the application oriented ones,
intelligent factory and intelligent traffic systems, where the including smart grid, intelligent factory (intelligent manu-
physical entities and the virtual cyber world benefit from facturing) and autonomous vehicles. It can be seen that the
mutual association with each other [2]–[4]. A major new research output in most topics has increased continuously,
challenge is that the traditional problems regarding hard- with the number of journal publications in IoT surging over
ware, software and networked systems are correlated [5], [6]. 8, 000 level in 2017. As a subtopic of CPS, ICPS has emerged

2169-3536 2018 IEEE. Translations and content mining are permitted for academic research only.
47374 Personal use is also permitted, but republication/redistribution requires IEEE permission. VOLUME 6, 2018
See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
Y. Jiang et al.: Data-Driven Monitoring and Safety Control of ICPSs: Basics and Beyond

FIGURE 1. Trend of research focus on related topics.

since 2013, and has received increasing attention ever after.

Among the application topics, smart grid first received most
investigations and has over 3,000 SCI core collection pub-
lications per year since 2014. Researches in autonomous
vehicles show steady growth since 2014 and have output
over 2,600 papers in 2017. By contrast, it is surprising that
the intelligent factory and intelligent manufacturing related
studies have seen the least outcome among the investigated
research areas during the past five years.
Figure 2 depicts the indexing results of IEEE publica-
tions. The journal and the magazine papers reflect the recent
advances in different subjects, while the conference pub-
lications indicate the popularity of the topics in the IEEE
community, and the number of books published reveals how
systematically they have been studied. The plots have pro-
duced results essentially in agreement with the above anal-
ysis based on the Web of Science, except that the IEEE
publications on intelligent factory notably outweigh those
on ICPS. Smart grid and IoT are the most systematically
studied. By contrast, few books about Industry 4.0 and ICPS FIGURE 2. Comparison of publication numbers in IEEE.
are seen. It is also interesting to note that autonomous vehicle
related research is the hottest topic in terms of the volume of
conference papers. In Section IV, recent progress in data-driven control
Dedicated to producing a systematic and thorough review, strategies is reviewed. Based on the analysis of the
the contributions of this work are as follows. Firstly, the cur- requirements in ICPS applications given in Section V,
rent status of the theoretical research and the applications Section VI proposes the key challenges and significant
of ICPS monitoring and control are investigated. Secondly, topics of future research. The last section concludes the
the novel sensing techniques that provide the primary infor- paper.
mation used in the data-driven approaches are categorized
and compared. Thirdly, recent advances in data-driven moni- II. INTELLIGENT SENSING TECHNIQUES
toring, fault diagnosis and control approaches, as well as their A. SMART SENSORS
applications to the smart grid and the autonomous vehicles In most cyber-physical systems, sensing or perception acts
are reviewed in detail. Fourthly, key challenges and future as an important process of interacting with the external envi-
research directions are pointed out. ronment, through which the current status of the system is
The remainder of the paper is organized as follows. represented with quantified features. From the viewpoint of
The next section summarizes popular sensing devices and multi-agent systems, each intelligent device can be regarded
novel techniques. In Section III, the recent advances in as an agent that takes up some computing power and has some
the data-driven ICPS monitoring and fault diagnosis are autonomous capabilities, sharing data and information that
reviewed, with a special focus on smart grid applications. other agents cannot observe.

VOLUME 6, 2018 47375

 Y. Jiang et al.: Data-Driven Monitoring and Safety Control of ICPSs: Basics and Beyond

Smart sensors are sensing devices that have digitaliza- extensively studied [16]. It has been revealed that there is
tion ability and digital information processing functionalities a one-to-one mapping from an arbitrary parity vector to a
[7]–[9]. Smart sensors generally consist of micro-processor, normalized diagnostic observer if the observer dimension
electronic circuits and I/O interfaces. Compared with tradi- is fixed. Using historical data rather than the demanding
tional sensors, smart sensors have greater capabilities for data knowledge about the systems’ mechanism models, the ‘‘par-
processing, storage and information transformation. Useful ity space design–observer realization’’ scheme can greatly
functions such as automatic calibration, zero correction and reduce the design effort of the diagnostic observer, which is
scaling of the measured signals are achieved by the micro- the core of constructing a data-driven state observer. It can
processors, and thus do not require as much meticulous elec- also be guaranteed that the observer achieves unbiased state
tronic design, debugging and testing as the traditional analog tracking and unbiased output estimation for LTI systems.
signal based sensors do [10]–[12]. In addition, efficient com- For nonlinear systems and large-scale complex systems,
munication and flexible networking are also irreplaceable the observer based approaches will lose their advantages.
features of smart sensors in order to implement standardized In these circumstances, learning based schemes show bet-
I/O interfaces [13]. ter performance and robustness against noises. Just-in-time
learning (JITL) and deep neural networks are some exam-
B. FREQUENCY DISTURBANCE RECORDER (FDR)
ples of the current research focus of soft sensing for non-
Frequency disturbance recorders have been developed as a linear systems with robustness design against missing data,
plug-and-play (PnP) device which has easy access to the infrequently measured key-performance-indicator (KPI) and
power grid by simply plugging into the 110 Volts or 220 Volts stochastic noise [17]–[20].
power outlets. They collect the frequency data in real- In contrast to real-time sensing, which corresponds to soft
time and send them to the information management sys- measurement, predictive soft sensing can ‘‘measure’’ vari-
tem through long-distance communication (Ethernet) for the ables that traditional sensors cannot measure. This is espe-
dynamic monitoring purpose. The first generation of fre- cially meaningful for KPI oriented process monitoring and
quency disturbance recorder is based on the GPS timing high-level production scheduling. For instance, in the steel
mechanism, while the second generation is based on the production industry, the quality of steel plate rolled out of
Internet timing synchronization technique. the rollers cannot be instantly measured due to the extremely
C. SCADA high temperature. It usually takes a period of time before the
SCADA is the abbreviation of ‘‘supervisory control and data product is finally evaluated. Predictive soft sensing provides
acquisition’’. The SCADA system is a computer based pro- the approximation of significant evaluation index before real
duction process control and dispatching automation system. experimental tests are completed. To this end, various KPI
It can monitor and control the field devices. Applications of prognosis approaches have been proposed and applied to
SCADA systems include real-time process monitoring such wine production, chemical reaction process and battery health
as oil refining and water treatment processes, as well as monitoring [21]–[26]. Among these approaches, the mod-
critical urban infrastructure such as the power grid and the ified partial least squares (PLS) based ones are effective in
transportation systems [14]. As for the third generation of terms of robust KPI prognosis and KPI oriented fault detec-
SCADA systems, they can be deployed over long physical tion, since they can properly address the issues of high dimen-
distances and the design targets are for the distributed con- sionality and collinearity. These algorithms were realized in
trol tasks and large-scale networked monitoring systems. For a recently published open source Matlab toolbox called the
the fourth generation of SCADA systems, the open network Data based KPI oriented fault detection toolbox (DB-KIT)1 .
protocol plays a central role in the design of more powerful
SCADA systems, enabling cloud computing and Internet- III. ADVANCES ON DATA-DRIVEN MONITORING
of-things (IoT) technologies to be employed to increase AND FAULT DIAGNOSIS
the robustness and flexibility of the objective systems [15]. A. THREATS IN INDUSTRIAL CPS
However, these open configurations bring about increased Industrial CPSs are exposed to more severe threats from
security vulnerability and expose the monitoring systems to multiple sources than traditional industrial systems. This
various potential cyber-attacks. part briefly summarizes different categories of threats with
supportive real cases.
D. SOFT SENSORS There are external threats and internal threats endanger-
Soft sensors are the software libraries or algorithms that ing the industrial CPSs. Typical external threats include
achieve state estimation and key-performance-indicator pre- hacking activities, APT (Advanced Persistent Threat) attacks
diction. In the data-driven process monitoring and fault and hardware targeted attacks, while the internal ones can
diagnosis framework, soft sensing approaches have been be triggered intentionally by an employee or an external
developed based on data-driven observers, multivariate contractor.
analysis (MVA) and machine learning techniques.
Over the past decade, the relationship between the par- 1 https://cn.mathworks.com/matlabcentral/fileexchange/65348-db-kit
ity vectors and the Luenberger type observer has been Or access via http://homepage.hit.edu.cn/yinshen (English version)

47376 VOLUME 6, 2018

Y. Jiang et al.: Data-Driven Monitoring and Safety Control of ICPSs: Basics and Beyond

While external attacks are mostly making use of the sys- variability to the energy consumption end. Considering these
tems’ vulnerabilities, internal attacks access the central con- facts, the least square based state estimation approach has
trol units much more easily and can cause major breakdown the ‘‘local optimum problem’’, where the current state is not
to the operations. A famous case tells about a disaffected suitable for the initial guess of the next state [33]. Likewise,
employee attacking the waste water treatment control sys- the gradient descent based approaches become invalid in
tem [27], [28]. The employee worked for a company that these cases. To deal with the above concerned problem, recur-
provides SCADA system installation service for Queens- sive tracking techniques of the statistical properties based
land Maroochy Shire Council. By hacking into the control on first order perturbation are used for continuous varying
system and using unauthorized instructions, 800 thousand process [34]. For processes with abrupt changes and strong
litres sewage influxed into local parks, rivers and even hotel nonlinearity, locally weighted projection regression (LWPR)
lobbies. Serious damage to the environment was caused by based approaches show better performance and robustness.
this shocking event.
Human error is also one of the biggest threats to indus- C. UNOBSERVABLE ATTACKS
trial control systems, despite being unintentional in most While some researchers are dedicated to designing reliable
cases. For example, investigations showed that the Three Mile monitoring systems against disturbances and uncertainties,
Island nuclear accident started from the negligence of an some others are trying to reveal the defects of the exist-
initial cleaning equipment [29]. The reactor was completely ing fault diagnosis systems by proposing potentially fac-
destroyed, and the whole process took only 120 seconds. titious attacking schemes, among which the low-sparsity
Reports show that this event was graded as a five level nuclear unobservable attack is one of the most challenging types
accident, caused by a series of erroneous human operations in the anti-hacker-attack practice. Low-sparsity unobservable
and failing to deploy an effective fault diagnosis system. attack refers to the adversarial false data injection methods
The unsafe operation of complex industrial systems has that tamper with minor measurement variables. A successful
caused huge losses to people’s lives and property. In 1993, attack will not trigger alarms when the traditional monitoring
an explosion that occurred at the Beilun power plant in and fault detection approaches are used, such as the bad data
Zhejiang, China led to 22 deaths and 8 serious injuries. detection (BDD) approach [35]. Specifically, the monitoring
In 2005, an explosion occurred at the Petro China Jilin and control of the smart grid requires a timely update of
Petrochemical Corporation, resulting in serious environmen- the system status, which is determined based on real-time
tal pollution, and the water supply to 9 million people in the collected meter data. There is a potential security hazard for
downstream city was cut off for five days. The large-scale and the smart grid when the polluted measurements from the
networking of modern industry also magnifies local and tiny unprotected and easily attacked meters are used to monitor
failures through the control loops [30]. Based on the above and manage the smart grid, because the false data could
facts, reliable process monitoring and fault diagnosis system mislead the SCADA system to make wrong decisions. It is
design schemes that can be applied to large-scale processes revealed that using partial (low-sparsity) meter leakage data
are urgently in need. could launch an unobservable attack to the smart grid. It is of
vital importance to prevent this from happening.
B. DATA-DRIVEN STATE ESTIMATION There has been literature reporting potential unobserv-
Traditionally, state estimation approaches rely on a system able attack schemes, which provides references for sys-
model with known structures and parameters, and are well tem safety design before an actual attack makes use of
developed in the framework of state observers [31]. For these defects of the monitoring and fault diagnosis systems.
instance, Kalman filter is recognized as a powerful estima- Reference [36] proposed a subspace aided unobservable
tor for linear time varying systems with stochastic noise. attack approach using partial measurement data. It was also
However, the a priori knowledge these approaches required is shown that although such an attack triggers an alarm with the
not usually available, in which circumstances the data-driven existing fault diagnosis systems, it ‘‘launders’’ the injected
state estimation methods are necessary. data and incriminates the normal data.
Multivariate statistical analysis (MVA) based and subspace
aided approaches are two typical alternatives. For smart D. DATA-DRIVEN FAULT DIAGNOSIS
grid state estimation, a notable property is the frequent and Traditionally, industrial systems operating in the steady
significant changes in power generation and power con- state are monitored with univariate statistical analysis meth-
sumption [32]. Compared with traditional power grid, power ods or signal processing methods. However, these methods
generation and consumption in smart grid is uncertain in fail to explore the patterns and correlation relationship among
nature because (i) New energy, such as wind energy and solar the variables, knowledge of which is characteristic in reveal-
energy, is hooked up to the grid. These new energy sources ing the malfunctioning in the systems. This fact is even more
are seriously influenced by natural conditions. (ii) Newly noticeable for the modern large-scale complex systems with
introduced scheduling of the power in the large-scale grid multi-loop, multi-level coupling and correlated features [34].
re-distributes the power generation requirement. (iii) Plug- Considerable efforts have to be made to monitor multiple
and-play devices such as the hybrid electric vehicles bring monitoring screens if these methods are applied in such

VOLUME 6, 2018 47377

 Y. Jiang et al.: Data-Driven Monitoring and Safety Control of ICPSs: Basics and Beyond

systems, which introduces potential threats to the safety and using the relationship with parity vectors (as mentioned in
reliability of the system. Section II-D), which can be identified using the historical
In recent years, multivariate statistical analysis (MVA) input and output data.
based process monitoring and fault diagnosis approaches From the sampled data point of view, the observed vari-
have been extensively investigated. The corresponding sys- ables represented in time series forms can be transformed into
tem design and realization procedure follows an ‘‘offline a compact form where only one sample of the historical state
training—online implementation’’ procedure [37]. Principal variables is of concern [47]. By stacking the state variables
component analysis (PCA) and PLS techniques are two into a compact data form, the so-called parity relation is
representative basis of the MVA monitoring and diagnosis established. In order to achieve precise estimation of the
schemes [38]. For PCA, it is assumed that there exists a lower outputs and force the residual signals to converge to zero
dimensional principal component space, which represents under fault-free condition, the term with the state variables
significant variance information of the whole dataset. This should be eliminated by a parity vector orthogonal to the
assumption enables the monitoring of a lower dimension coefficient matrix. All the qualified parity vectors constitute
principal component space rather than the high dimensional the parity space, which is the null space of the coefficient
mutually coupled variables. In PLS based schemes, optimal matrix of the state variables. Therefore, the selection of parity
decomposition of the measurement space is achieved in the vector used for residual generation becomes an important
sense of maximizing the correlation between the input and research focus. The goal is to choose such favorable parity
output variables [39]. vectors that lead to fault detectors sensitive to the faults
In practice, for those systems where process dynamics can- and at the same time ensures robustness to uncertainties and
not be omitted, the performance of the basic MVA based pro- disturbances. Other research efforts in this area are dedi-
cess monitoring and fault diagnosis schemes will be seriously cated to reducing the computational load by dimensional-
affected [40], [41]. Specifically, the normalization procedures ity reduction [48] and dealing with nonuniform sampling
are no longer suitable for these processes due to the time- issues [49].
varying nature of the variables’ statistics such as the mean and
variance values. As a result, the estimations are biased, and IV. ADVANCES IN DATA-DRIVEN CONTROLLER DESIGN
the interpretability of these statistics is weakened. To handle Industrial cyber-physical systems that have hierarchical
the problem, the state space representation of the dynam- control architecture generally consist of four levels: the com-
ical processes is used to characterize necessary dynamics. ponent level, the control loop level, the functional subsys-
In this context, data-driven analytical redundancy is used to tem level, and the plant-wide decision-making level. The
describe the process and generate virtual estimation signals, component and control loop levels have been the subjects
which are used for subsequent fault diagnosis procedures, of traditional control theory and the system identification
as well as performance analysis and evaluation [42]. It can area [30], [50]. The upper decision-making levels have access
also be regarded as a reformulation of the fault diagnosis to the cyber resources so they are more ‘‘intelligent’’. By con-
system design objective: variation detection by comparing trast, the lower levels have to complete their tasks under the
and evaluating of the difference between the real plant and specifications from the upper levels. This part reviews two
its analytical redundancy. Furthermore, by means of the categories of fault-safe approaches applicable to the lower
so-called stable kernel representation (SKR), it is possible levels and to the higher levels, respectively.
to bridge the observer based methods with their data-driven
realizations [43]. A. PLUG-AND-PLAY CONTROL
Approaches based on fault detection filter (FDF), diagnos- In terms of the existing control systems, especially those with
tic observer (DO) and Kalman filter have been designed for encapsulated modules in complex control loops, equipping
systems corrupted with deterministic disturbances and pro- them with fault-tolerant capability requires embedding addi-
cess faults [44], [45]. The observer based residual generation tional monitors and controllers without modifying the exist-
schemes are in closed-loop configuration with a feedback ing ones. This requires the monitoring systems to have strong
gain to be designed. Through a proper choice of this gain, scalability and modularity. Plug-and-play (PnP) control aims
the residual dynamics can be designed to possess arbitrary at designing control strategies that add new controller mod-
convergence rate given any initial value while suppressing ules while reserving the existing ones in use (such as the
the effects of model uncertainties. Robust approaches against widely deployed PID controllers) [51], [52]. In practice, full-
unknown input signals such as disturbances have been pro- time operating processes and live systems are frequently
posed based on Kalman filter. It should be noted that for data- maintained. Redesign of the whole system in response to
driven process monitoring and fault diagnosis, the parameters small changes is usually not feasible due to cost and com-
of the mechanism models are not known a priori, which missioning concerns. In this sense, PnP strategies are one of
indicates that the system matrices in the Luenberger equations the most practically applicable alternatives to improve the
are unavailable in the design of the feedback gain L for state systems’ long-term performance and optimize the set-point
tracking [46]. To handle this issue, it was reported in [16] that configuration according to the instructions from the upper
these observer based residual generators can be constructed decision-making levels.

47378 VOLUME 6, 2018

Y. Jiang et al.: Data-Driven Monitoring and Safety Control of ICPSs: Basics and Beyond

Another challenging problem is the online configuration of

the controllers to accommodate faults and retain the system
stability and performance. For data-driven fault-tolerant con-
trol, some passive control schemes have been developed with
the aid of predesigned fault scenarios [57]. However, passive
control improves the system robustness while introducing
more conservativeness, which limits the online performance.
In recent years, active fault tolerant control schemes have
been attracting extensive attention. Badihi et al. [58] have
extended attention to adaptive control strategies of actuator
faults. The work [59] is one of the first that proposed an active
fault tolerant scheme for uncertain strict-feedback nonlinear
systems. Kamal et al. [60] proposed a multi-observer switch-
FIGURE 3. Schematic of the idea and realization of PnP control. ing control strategy for robust active fault tolerant fuzzy
(a) Existing control system (b) Controller cannot be modified due to
encapsulation (c) PnP architecture (d) PnP realization. control of variable-speed wind energy conversion systems.

B. REINFORCEMENT LEARNING AIDED

CONTROL STRATEGIES
Reinforcement learning (RL) is an experience based approach
to achieve (near) optimal control policy design using only
data obtained from actual observations [61]. Without an
explicit model of the system dynamics, RL keeps maintaining
updates of the value function to achieve better approximation,
based on which to select actions that approach to the long
term goal formulated by the accumulated future rewards.
For vehicle control problems, one of the major difficul-
ties to apply RL approaches in learning a controller from
scratch lies in the oversized state space. Traditional ‘‘tabular
approaches’’ estimate the value function v(s) (or the value-
action function q(s, a)) by keeping updates of values corre-
FIGURE 4. Complete control diagram of PnP-PMCA [53]. sponding to the individual states (or state-action pairs). These
approaches are no longer suitable because (1) the memory
In [53], an advanced PnP process monitoring and con- requirement for storing the values are exponentially explod-
trol architecture (PnP-PMCA) was proposed, which is an ing (2) values in some states could never be learned or updated
integrated implementation of process monitoring and con- during the value function estimation phase (3) from the
trol with scalable structure and modularized components. incremental learning point of view, the online computational
By introducing Youla parameterization, all stabilizing con- burden is too heavy to meet the real-time requirement.
trollers are represented in a uniform formula [54]. This The research towards solving the above mentioned chal-
achieves a smooth transformation from the original system lenges fall into two directions. The first direction focuses
to the modified system with guaranteed internal stability. on the Q-learning and actor-critic learning where conver-
Based on these studies, some fundamental schemes have gence condition should be carefully discussed and the balance
been proposed for observer-based residual generator design between exploration and exploitation be addressed.
and online configuration. In [53], the plug-in module of In [61], a Takagi-Sugeno fuzzy inference system was con-
process monitoring system is developed. In order to achieve structed to generate control signals, and a Q-estimator net-
self-configuration, adaptive and iterative online configuration work formulated as a time-delay neural network was used to
approaches are proposed in [55]. The adaptive approach has approximate the state-action value function. Then the param-
a high convergence speed, however, it carries huge online eters of the controller were adjusted based on the Q-estimator
computational load at each sampling instant. Compared with network so as to maximize Q(x, a). To ensure exploration of
the adaptive approach, the iterative scheme avoids the numer- actions in all states, a stochastic action modifier was added to
ical sensitivity problem and significantly reduces the online change the actual control command. This scheme was suc-
computational load. On the other hand, the scheme has a cessfully applied to a vehicle longitudinal-control problem
lower convergence speed. Considering the fact that the pro- which aims to maintain a safe distance of the object from
cess industries are generally corrupted with unknown distur- its preceding vehicle, and maintained a spacing difference
bances, a reliable process monitoring scheme is developed within 6 meters at all time after only 68 learning episodes.
in [56] for stationary processes to ensure the monitoring Reference [62] proposed a batch-learning control approach
performance. developed under the actor-critic framework where batch

VOLUME 6, 2018 47379

 Y. Jiang et al.: Data-Driven Monitoring and Safety Control of ICPSs: Basics and Beyond

processing was designed for both value function estimation 2) Interconnection is more complex. To achieve effi-
and value function update. The value function approximator cient transmission, more city-wide connections will be
was parameterized as a linear combination of the kernel basis. established and as a consequence the topological struc-
The work suggested to share the same kernel based features ture will be greatly changed. The number and location
in the actor and the critic to obtain better generalization of agents will change frequently.
capability. Experimental results also showed that the learned 3) Grid-wide scheduling with closed-loop control is inte-
policy (actor) processes better adaptability to multiple work- grated. Based on the real-time information provided by
ing conditions including road sloping and slippery pavement. the smart monitoring devices, it is easier to redistribute
The second direction is closer to trivial solutions since the available power resource.
it treats RL as an optimization tool to optimize existing 4) More potential safety hazards. The smart sensors have
stabilizing controllers with undetermined parameters to be a decentralized physical distribution that is more vul-
tuned. In this scenario, parameterization of the stabilizing nerable to hostile attack. Electrical substations at key
controllers and the definition of optimization goals play the intersections are responsible for multiple domains.
central role. The advantage is that the systems’ stability can Closed-loop configuration introduces stability issues.
be guaranteed. The only concern would be the controller’s New monitoring and fault diagnosis schemes are
performance. One of the possible means of Youla parameteri- required against hardware failure and cyber malfunc-
zation has gained extensive research attention and is expected tioning (such as false data injection).
to be integrated with the fault-tolerant controllers.
B. CONSTRAINED CONTROL
V. PRACTICAL REQUIREMENTS AND CURRENT Another challenge to the controller design task lies in the
STATUS OF ICPS APPLICATIONS constrained control problem under system dynamics varia-
A. SMART GRID tion due to external environment changes [64]. Regarding
Power grid is the electric network that provides the electrical autonomous driving, the newly introduced difficulty of such
energy to the users. In the broad sense, power grid consists of a problem mainly results from the contradiction between the
all the connections between the power generation end and the demanding high level commands and the limited ability of
customer end. The major functions of the traditional power low level execution. In particular, the control objectives are
grid include power transmission, power transformation and set in real-time by the top decision making levels of the
power distribution. Real-time monitoring of such large-scale autonomous system, which are oriented to the transactions
complex systems requires robust and efficient approaches, from external environment condition perception to internal
as well as reliable measurement devices and communication responses to adapt to various working conditions. By contrast,
networks. the control loops and actuation components at the bottom
To manage the huge system of power grid, measurement levels must guarantee the systems’ stability while trying to
devices are installed at a great number of terminals to measure achieve the tracking performance required by the upper lev-
and inspect the real-time voltage, current and frequency. This els, which may be unachievable.
open loop configuration is incapable of achieving grid-wide A reasonable balance and coordination among the hier-
fault diagnosis and energy scheduling. It should be noted achical cyber-physical system should be achieved. To this
that unlike water supply systems, electrical energy is difficult end, researches in output constraint control were carried out
and expensive to store. Large-scale electricity storage plan is to deal with path following problem in [65], and trade-off
not a economical option to compensate the excessive electric strategies for over-actuated vehicles using control allocation
supply at peak hours for that at low hours, and relies heav- laws were proposed to improve the operational efficiency and
ily on the power infrastructure. How to balance the power motion control performance of the vehicle in [66] and [67].
generation capacity and guarantee the supply for heavy loads
at peak hours is essential for environmental and economic C. INTELLIGENT FACTORY
concerns. Conventionally, the production lines in factories are yield
Smart grid development not only aims at including more oriented. Given the desired specifications of final product
supervisory equipment to enhance the observability and con- quality, production quantity and efficiency are the top pri-
trollability of the network system, but also extends various orities. By contrast, the new generation of factories based
forms of energy sources [63]. Compared with traditional on ICPS aim to achieve plant-wide intelligent manufacturing
power grids, smart grids have the following features: and mass customization, and the ultimate goal is oriented
1) Network scales are enlarged with better expendabil- to the marketing. This needs increased flexibility among the
ity. In the CPS framework, the local power plants departments and units so that data can be shared across the
are treated as the input terminals that can serve glob- levels of the multi-hierarchical architectures.
ally. In other words, their energy can be delivered to Qingdao Haier Mould Corporation is the largest mould and
remote customers with less border limitation. In addi- fixture manufacturer in China. Haier Mould Corporation is
tion, a variety of new energy sources and types such as equipped with world’s leading products, mold design, analy-
wind power and solar energy are hooked up to the grid. sis and processing software, as well as various kinds of high

47380 VOLUME 6, 2018

Y. Jiang et al.: Data-Driven Monitoring and Safety Control of ICPSs: Basics and Beyond

speed machining center, spark machine, wire cutting machine Over the past decade, the market has promoted the appli-
and many other types of professional equipment. To achieve cations of a series of smart terminals and devices which serve
a deep fusion of informatization and industrialization, Haier for information digitalization, data acquisition, open-loop
Mould Corporation started building smart factory based on monitoring, etc. One of the most successful applications is
CPS management system since 2013. the vehicle license plate recognition system, using the reliable
pattern recognition based image processing techniques [69].
With the recognized digital plate numbers, the access infor-
mation of vehicles into a certain area can be automatically
collected. They are also used in the automatic snapping sys-
tem to record infringing vehicles, whose data are sent to the
public security website for self-serve fine payment.
For logistics companies, freight volume prediction and
route planning algorithms are also developed to maximize
the efficiency–cost ratio [70], [71]. They predict how busy
the highways and lanes are, so most traffic congestions
can be avoided, and optimal scheduling can be achieved.
However, the effectiveness of these schemes is subject to ideal
assumptions, and their performance will be greatly degraded
if the randomly occurring natural and social factors violate
FIGURE 5. An industrial CPS collaboration and management system. the applicable conditions.

The core of the intelligent factory is the integrated ICPS VI. KEY CHALLENGES AND FUTURE
collaboration and management system, as shown in Figure 5. RESEARCH DIRECTIONS
At the first stage, the production plans are input to the A. KEY CHALLENGES TO ICPS MONITORING
collaborative platform based on higher level decisions. When AND SAFETY CONTROL
the processing tasks are updated, program preparation, tool In this paper, we have looked into ICPSs including the smart
preparation and electrode preparation are carried out in par- grid, autonomous vehicle, as well as intelligent factory and
allel. In the meantime, status information of different subsys- intelligent traffic systems. It can be seen that the transition
tems is transmitted to the corresponding sectors in charge. from the theoretical breakthroughs to applicable techniques
Then, automatic check and automatic calibration will be car- of the monitoring and control systems is urgently in need.
ried out before the raw material is pushed to the machine The existing monitoring and control schemes mostly focus
tools. Throughout the process, the digital machine tools’ on lower level performance indices. For instance, at the con-
status is fed to the control and management center. The core trol loop level, the closed-loop stability, the response time and
features of the intelligent mould production factory are listed the tracking error are generally of concern. However, such
as below. control loop level performance indices are insufficient in the
• Deep fusion of information software and manufacturing context of Industry 4.0. The conventional system monitoring
equipment. and control techniques are applied mostly to a few control
• Multi-sectoral collaboration and management. loops rather than the whole process. However, for large-
• Decision support based on big data analysis. scale complex systems, it is more important to achieve global
• Improved OEE (Overall Equipment Efficiency). stabilization and performance optimization at the plant-wide
decision making level [72]. From the overall system stability
D. INTELLIGENT TRAFFIC and performance guarantee point of view, the plant-wide
The traditional research topics of traffic engineering have oriented decisions such as a more reasonable resource allo-
been extremely complicated due to the simultaneous require- cation strategy may outweigh the performance improvement
ments from various subjects including systematics, engineer- of a local maintenance action that is less urgent. Moreover,
ing, law and regulations, etc. There are too many constraints the online learning and optimization policies are beneficial
the system should fulfill. Therefore, in the general case, some for transparent global system management. The macroscopic
factors have to be compromised [68]. Intelligent traffic is guidelines can avoid catastrophic breakdowns to the best
the central development direction of future traffic engineer- extent, and reduce the chance of a massive shutdown. In addi-
ing. Oriented to the goal of high transportation capacity, tion, the global performance indicators are more intuitive
low traffic accident rate, low energy consumption and more and goal-oriented. It is therefore more feasible to alter the
economical transport, intelligent traffic systems should make overall economic strategy based on instructions given by the
full use of the newly built information infrastructure and plant-wide performance supervised monitoring and control
vehicles (from the hardware aspect), as well as advanced architecture. Such features are of great value to increase the
prediction approaches and scheduling strategies (from the competitiveness in the current rapidly changing economic
software aspect) [68]. environment and boost the economic benefits [73].

VOLUME 6, 2018 47381

 Y. Jiang et al.: Data-Driven Monitoring and Safety Control of ICPSs: Basics and Beyond

TABLE 1. Traditional tasks and new challenges of ICPS monitoring and safety control.

Industrial standards and theoretical frameworks are being VII. CONCLUSIONS

established from the leading research institutes and scholars Oriented to ICPS monitoring, fault diagnosis and control
around the world. Since the topic of ICPS monitoring and tasks, this paper investigates the recent status of research
control covers too many domains, we only summarize the and reviews the advances in data-driven approaches over the
key challenges to the existing problems from a macroscopic past two decades. Data sources and preliminary processing
perspective. Table 1 compares the traditional tasks and the and transmission are summarized in relation with the intel-
corresponding new challenges in different stages of system ligent sensing techniques. The new challenges in the smart
monitoring and control. grid systems that the conventional monitoring approaches
cannot handle are pointed out. Furthermore, the data-driven
B. FUTURE RESEARCH DIRECTIONS state estimation problems, unobservable attacks and recent
To achieve the target of plant-wide monitoring and safety advances in the data-driven fault diagnosis schemes are dis-
control of the large-scale complex ICPSs in a data-driven cussed with a focus on their limitations in real systems.
manner, and in the face of the aforementioned challenges, Advances in the control strategies including the plug-and-
the following research topics are proposed. play control, reinforcement learning aided control and con-
strained control are reviewed. Finally, new challenges and
• Design of data-driven fault-tolerant controllers against
future research topics are proposed based on the practical
unknown faults, especially for nonlinear systems, time
requirements of ICPS applications.
delay systems, and distributed systems.
• Online fault localization and isolation based on the fault ACKNOWLEDGEMENT
propagation analysis, with little or no a priori knowledge The authors would like to sincerely thank the anonymous
of the system structure. reviewers and the editors who provided appropriate sugges-
• Plant-wide key performance indicator supervised sys- tions and made detailed corrections, and have made the paper
tem monitoring that reduces unnecessary shutdown much more readable.
maintenance.
• Deep learning based data analysis against false data REFERENCES
injection and low sparsity unobservable attack. [1] K. Sampigethaya and R. Poovendran, ‘‘Aviation cyber–physical systems:
Foundations for future aircraft and air transport,’’ Proc. IEEE, vol. 101,
• Reinforcement learning aided online optimization and
no. 8, pp. 1834–1855, Aug. 2013.
adaptive control of highly dynamic systems. [2] S. Sridhar, A. Hahn, and M. Govindarasu, ‘‘Cyber-physical system security
• Plug-and-play controller design (PnP-PMCA) that for the electric power grid,’’ Proc. IEEE, vol. 100, no. 1, pp. 210–224,
Jan. 2012.
reduces subsystem maintenance time. [3] A. Carcano, A. Coletta, M. Guglielmi, M. Masera, I. N. Fovino, and
• OEE oriented life cycle management based on key per- A. Trombetta, ‘‘A multidimensional critical state analysis for detecting
formance indicator prediction and fault prognosis. intrusions in SCADA systems,’’ IEEE Trans. Ind. Informat., vol. 7, no. 2,
pp. 179–186, May 2011.
• Sensor data fusion based knowledge exploration, such
[4] J. M. Hendrickx, K. H. Johansson, R. M. Jungers, H. Sandberg, and
as feature selection, pattern recognition and statistical K. C. Sou, ‘‘Efficient computations of a security index for false data
modelling. attacks in power networks,’’ IEEE Trans. Autom. Control, vol. 59, no. 12,
• Prototype system development based on microproces- pp. 3194–3208, Dec. 2014.
[5] I. N. Fovino, A. Coletta, A. Carcano, and M. Masera, ‘‘Critical state-based
sors, FPGA based design for multi tasks, as well as GPU filtering system for securing SCADA network protocols,’’ IEEE Trans. Ind.
acceleration for deep learning based solutions. Electron., vol. 59, no. 10, pp. 3943–3950, Oct. 2012.

47382 VOLUME 6, 2018

Y. Jiang et al.: Data-Driven Monitoring and Safety Control of ICPSs: Basics and Beyond

[6] P. Nader, P. Honeine, and P. Beauseroy, ‘‘lp -norms in one-class classifica- [29] G. E. Cummings, ‘‘Operator/instrumentation interactions during the three
tion for intrusion detection in SCADA systems,’’ IEEE Trans. Ind. Inform., mile island incident,’’ IEEE Trans. Nucl. Sci., vol. 27, no. 1, pp. 931–934,
vol. 10, no. 4, pp. 2308–2317, Nov. 2014. Feb. 1980.
[7] T. A. Tameh, M. Sawan, and R. Kashyap, ‘‘Smart integrated optical rotation [30] Z. Gao, ‘‘Fault estimation and fault-tolerant control for discrete-
sensor incorporating a fly-by-wire control system,’’ IEEE Trans. Ind. time dynamic systems,’’ IEEE Trans. Ind. Electron., vol. 62, no. 6,
Electron., vol. 65, no. 8, pp. 6505–6514, Aug. 2018. pp. 3874–3884, Jun. 2015.
[8] G. D. J. Martinez-Figueroa, D. Morinigo-Sotelo, A. L. Zorita-Lamadrid, [31] Z. Gao, S. X. Ding, and C. Cecati, ‘‘Real-time fault diagnosis and fault-
L. Morales-Velazquez, and R. D. J. Romero-Troncoso, ‘‘Fpga-based smart tolerant control,’’ IEEE Trans. Ind. Electron., vol. 62, no. 6, pp. 3752–3756,
sensor for detection and classification of power quality disturbances using Jun. 2015.
higher order statistics,’’ IEEE Access, vol. 5, pp. 14259–14274, 2017. [32] H. Fawzi, P. Tabuada, and S. Diggavi, ‘‘Secure estimation and control for
[9] N. A. Cloete, R. Malekian, and L. Nair, ‘‘Design of smart sensors for real- cyber-physical systems under adversarial attacks,’’ IEEE Trans. Autom.
time water quality monitoring,’’ IEEE Access, vol. 4, pp. 3975–3990, 2016. Control, vol. 59, no. 6, pp. 1454–1467, Jun. 2014.
[10] F. Maciá-Pérez, F. Mora-Gimeno, D. Marcos-Jorquera, [33] Y. Weng, R. Negi, C. Faloutsos, and M. Ilic, ‘‘Robust data-driven state
J. A. Gil-Martínez-Abarca, H. Ramos-Morillo, and I. Lorenzo-Fonseca, estimation for smart grid,’’ IEEE Trans. Smart Grid, vol. 8, no. 4,
‘‘Network intrusion detection system embedded on a smart sensor,’’ IEEE pp. 1956–1967, 2017.
Trans. Ind. Electron., vol. 58, no. 3, pp. 722–732, Mar. 2011. [34] Y. Jiang and S. Yin, ‘‘Recursive total principle component regression based
[11] Y. Hu, S. Zhang, Y. Yan, L. Wang, X. Qian, and L. Yang, ‘‘A smart electro- fault detection and its application to vehicular cyber-physical systems,’’
static sensor for online condition monitoring of power transmission belts,’’ IEEE Trans. Ind. Inform., vol. 14, no. 4, pp. 1415–1423, Apr. 2018.
IEEE Trans. Ind. Electron., vol. 64, no. 9, pp. 7313–7322, Sep. 2017. [35] S. Xie, J. Yang, K. Xie, Y. Liu, and Z. He, ‘‘Low-sparsity unobservable
[12] M. E. E. Alahi, X. Li, S. Mukhopadhyay, and L. Burkitt, ‘‘A temperature attacks against smart grid: Attack exposure analysis and a data-driven
compensated smart nitrate-sensor for agricultural industry,’’ IEEE Trans. attack scheme,’’ IEEE Access, vol. 5, pp. 8183–8193, 2017.
Ind. Electron., vol. 64, no. 9, pp. 7333–7341, Sep. 2017. [36] J. Kim, L. Tong, and R. J. Thomas, ‘‘Subspace methods for data attack on
[13] F. Lin, A. Wang, Y. Zhuang, M. R. Tomita, and W. Xu, ‘‘Smart insole: state estimation: A data driven approach,’’ IEEE Trans. Signal Process.,
A wearable sensor device for unobtrusive gait monitoring in daily life,’’ vol. 63, no. 5, pp. 1102–1114, Mar. 2015.
IEEE Trans. Ind. Inform., vol. 12, no. 6, pp. 2281–2291, Dec. 2016. [37] S. X. Ding, S. Yin, K. Peng, H. Hao, and B. Shen, ‘‘A novel scheme
[14] G. Falco, C. Caldera, and H. Shrobe, ‘‘IIOT cybersecurity risk modeling for key performance indicator prediction and diagnosis with application
for scada systems,’’ IEEE Internet Things J., to be published. [Online]. to an industrial hot strip mill,’’ IEEE Trans Ind. Informat., vol. 9, no. 4,
Available:https://www.researchgate.net/publication/324264634_IIoT_ pp. 2239–2247, Nov. 2013.
Cybersecurity_Risk_Modeling_for_SCADA_Systems, doi: 10.1109/JIOT. [38] G. A. Cherry and S. J. Qin, ‘‘Multiblock principal component analy-
2018.2822842. sis based on a combined index for semiconductor fault detection and
[15] A. Sajid, H. Abbas, and K. Saleem, ‘‘Cloud-assisted IoT-based SCADA diagnosis,’’ IEEE Trans. Semicond. Manuf., vol. 19, no. 2, pp. 159–172,
systems security: A review of the state of the art and future challenges,’’ May 2006.
IEEE Access, vol. 4, pp. 1375–1384, 2016. [39] R. Muradore and P. Fiorini, ‘‘A PLS-based statistical approach for fault
detection and isolation of robotic manipulators,’’ IEEE Trans. Ind. Elec-
[16] S. X. Ding, Data-Driven Design of Fault Diagnosis and Fault-tolerant
tron., vol. 59, no. 8, pp. 3167–3175, Aug. 2012.
Control Systems. London, U.K.: Springer, 2014.
[40] H. Chen, B. Jiang, N. Lu, and Z. Mao, ‘‘Deep PCA based real-
[17] W. Yan, D. Tang, and Y. Lin, ‘‘A data-driven soft sensor modeling method
time incipient fault detection and diagnosis methodology for electrical
based on deep learning and its application,’’ IEEE Trans. Ind. Electron.,
drive in high-speed trains,’’ IEEE Trans. Veh. Technol., vol. 67, no. 6,
vol. 64, no. 5, pp. 4237–4245, May 2017.
pp. 4819–4830, Jun. 2018.
[18] X. Yuan, Z. Ge, B. Huang, and Z. Song, ‘‘A probabilistic just-in-time
[41] H. Chen, B. Jiang, and N. Lu, ‘‘An improved incipient fault detec-
learning framework for soft sensor development with missing data,’’ IEEE
tion method based on kullback-leibler divergence,’’ ISA Trans., vol. 79,
Trans. Control Syst. Technol., vol. 25, no. 3, pp. 1124–1132, May 2017.
pp. 127–136, Aug. 2018. [Online]. Available: https://doi.org/10.1016/j.
[19] X. Yuan, Z. Ge, B. Huang, Z. Song, and Y. Wang, ‘‘Semisupervised isatra.2018.05.007
JITL framework for nonlinear industrial soft sensing based on locally [42] S. Anwar and L. Chen, ‘‘An analytical redundancy-based fault detection
semisupervised weighted PCR,’’ IEEE Trans. Ind. Informat., vol. 13, no. 2, and isolation algorithm for a road-wheel control subsystem in a steer-by-
pp. 532–541, Apr. 2017. wire system,’’ IEEE Trans. Veh. Technol., vol. 56, no. 5, pp. 2859–2869,
[20] M. Jarvisalo, T. Ahonen, J. Ahola, A. Kosonen, and M. Niemela, Sep. 2007.
‘‘Soft-sensor-based flow rate and specific energy estimation of industrial [43] S. Ding, Y. Yang, and Y. Zhang, ‘‘Data-driven realizations of kernel and
variable-speed-driven twin rotary screw compressor,’’ IEEE Trans. Ind. image representations and their application to fault detection and control
Electron., vol. 63, no. 5, pp. 3282–3289, May 2016. system design,’’ Automatica, vol. 50, no. 10, pp. 2615–2623, 2014.
[21] Y. A. W. Shardt, H. Hao, and S. X. Ding, ‘‘A new soft-sensor-based process [44] C. Keliris, M. M. Polycarpou, and T. Parisini, ‘‘A distributed fault detection
monitoring scheme incorporating infrequent KPI measurements,’’ IEEE filtering approach for a class of interconnected continuous-time nonlinear
Trans. Ind. Electron., vol. 62, no. 6, pp. 3843–3851, Jun. 2015. systems,’’ IEEE Trans. Autom. Control, vol. 58, no. 8, pp. 2032–2047,
[22] S. Yin and G. Wang, ‘‘Robust pls approach for kpi related prediction and Aug. 2013.
diagnosis against outliers and missing data,’’ Int. J. Syst. Sci., vol. 45, no. 7, [45] M. S. Mahmoud and P. Shi, ‘‘Robust Kalman filtering for continuous
pp. 1375–1382, 2014. time-lag systems with Markovian jump parameters,’’ IEEE Trans. Circuits
[23] Z. Ge, ‘‘Supervised latent factor analysis for process data regression mod- Syst. I, Fundam. Theory Appl., vol. 50, no. 1, pp. 98–105, Jan. 2003.
eling and soft sensor application,’’ IEEE Trans. Control Syst. Tech., vol. 24, [46] C. Choi, K. Lee, and W. Lee, ‘‘Observer-based phase-shift fault detection
no. 3, pp. 1004–1011, May 2015. using adaptive threshold for rotor position sensor of permanent-magnet
[24] Y. Jiang, S. Yin, and Y. Yang, ‘‘Comparison of KPI related fault detection synchronous machine drives in electromechanical brake,’’ IEEE Trans. Ind.
algorithms using a newly developed MATLAB toolbox: DB-KIT,’’ in Proc. Electron., vol. 62, no. 3, pp. 1964–1974, Mar. 2015.
IECON 42nd Annu. Conf. IEEE, Oct. 2016, pp. 7149–7154. [47] Z. Gao, X. Liu, and M. Z. Q. Chen, ‘‘Unknown input observer-based
[25] S. Yin, L. Liu, X. Gao, and H. R. Karimi, ‘‘Multivariate methods based soft robust fault estimation for systems corrupted by partially decoupled dis-
measurement for wine quality evaluation,’’ Abstract Appl. Anal., vol. 1, turbances,’’ IEEE Trans. Ind. Electron., vol. 63, no. 4, pp. 2537–2547,
pp. 1–7, Jun. 2014. Apr. 2016.
[26] S. Yin, X. Zhu, and O. Kaynak, ‘‘Improved PLS focused on key- [48] S. X. Ding, P. Zhang, A. Naik, E. L. Ding, and B. Huang, ‘‘Subspace
performance-indicator-related fault diagnosis,’’ IEEE Trans. Ind. method aided data-driven design of fault detection and isolation systems,’’
Electron., vol. 62, no. 3, pp. 1651–1658, Mar. 2015. J. Process Control, vol. 19, no. 9, pp. 1496–1510, 2009.
[27] M. Pajic, J. Weimer, and N. Bezzo, ‘‘Design and implementation of attack- [49] M. Suo, B. Zhu, and D. Zhou, ‘‘Neighborhood grid clustering and its
resilient cyberphysical systems: With a focus on attack-resilient state application in fault diagnosis of satellite power system,’’ Proc. Inst.
estimators,’’ IEEE Control Syst., vol. 37, no. 2, pp. 66–81, Apr. 2017. Mech. Eng., to be published. [Online]. Available: https://journals.sagepub.
[28] M. Abrams and J. Weiss, ‘‘Malicious control system cyber security attack com/doi/10.1177/0954410017751991, doi: 10.1177/0954410017751991.
case study: Maroochy water services,’’ MITRE Corp., McLean, VA, USA, [50] X. Xie and J. Lam, ‘‘Guaranteed cost control of periodic piecewise linear
Tech. Rep. 08-1145, Aug. 2008. time-delay systems,’’ Automatica, vol. 94, pp. 274–282, Aug. 2018.

VOLUME 6, 2018 47383

 Y. Jiang et al.: Data-Driven Monitoring and Safety Control of ICPSs: Basics and Beyond

[51] J. Bendtsen, K. Trangbaek, and J. Stoustrup, ‘‘Plug-and-play control— [71] W. Zhang, J. Tang, H. Kristian, Y. Zou, and Y. Wang, ‘‘Hybrid short-term
Modifying control systems online,’’ IEEE Trans. Control Syst. Technol., prediction of traffic volume at ferry terminal based on data fusion,’’ IET
vol. 21, no. 1, pp. 79–93, Jan. 2013. Intell. Transport Syst., vol. 10, no. 8, pp. 524–534, 2016.
[52] S. Riverso, F. Boem, G. Ferrari-Trecate, and T. Parisini, ‘‘Plug-and-play [72] W. Kohn, Z. B. Zabinsky, and A. Nerode, ‘‘A micro-grid distributed intel-
fault detection and control-reconfiguration for a class of nonlinear large- ligent control and management system,’’ IEEE Trans. Smart Grid, vol. 6,
scale constrained systems,’’ IEEE Trans. Autom. Control, vol. 61, no. 12, no. 6, pp. 2964–2974, Nov. 2015.
pp. 3963–3978, Dec. 2016. [73] A. Merabet, K. T. Ahmed, H. Ibrahim, R. Beguenane, and
[53] H. Luo, X. Yang, M. Kruger, S. X. Ding, and K. Peng, ‘‘A plug-and- A. Y. M. Ghias, ‘‘Energy management and control system for laboratory
play monitoring and control architecture for disturbance compensation in scale microgrid based wind-PV-battery,’’ IEEE Trans. Sustain. Energy,
rolling mills,’’ IEEE/ASME Trans. Mechatron., vol. 23, no. 1, pp. 200–210, vol. 8, no. 1, pp. 145–154, Jan. 2017.
Feb. 2018, doi: 10.1109/TMECH.2016.2636337.
[54] H. Khatibi and A. Karimi, ‘‘H∞ controller design using an alternative
to Youla parameterization,’’ IEEE Trans. Autom. Control, vol. 55, no. 9,
pp. 2119–2123, Sep. 2010.
YUCHEN JIANG (S’16) received the B.E.
[55] S. S. Ge, Z. Li, and H. Yang, ‘‘Data driven adaptive predictive control for
holonomic constrained under-actuated biped robots,’’ IEEE Trans. Control
degree (Hons.) in automation from the Harbin
Syst. Technol., vol. 20, no. 3, pp. 787–795, May 2012. Institute of Technology, Harbin, China, in 2016,
[56] S. Riverso, F. Boem, G. Ferrari-Trecate, and T. Parisini, ‘‘Plug-and-play where he is currently pursuing the Ph.D. degree
fault detection and control-reconfiguration for a class of nonlinear large- with the Department of Control Science and
scale constrained systems,’’ IEEE Trans. Autom. Control, vol. 61, no. 12, Engineering.
pp. 3963–3978, Dec. 2016. He participated in an Exchange Program at the
[57] M. Benosman and K. Y. Lum, ‘‘Passive actuators’ fault-tolerant control University of New South Wales, Sydney, NSW,
for affine nonlinear systems,’’ IEEE Trans. Control Syst. Technol., vol. 18, Australia, in 2014. His research interests include
no. 1, pp. 152–163, Jan. 2010. data-driven fault diagnosis, process monitoring,
[58] H. Badihi, Y. Zhang, and H. Hong, ‘‘Wind turbine fault diagnosis and fault- cyber-physical systems, and artificial intelligence.
tolerant torque load control against actuator faults,’’ IEEE Trans. Control
Syst. Technol., vol. 23, no. 4, pp. 1351–1372, Jul. 2015.
[59] M. Shahriari-kahkeshi and F. Sheikholeslam, ‘‘Active fault tolerant control
design for uncertain strict-feedback non-linear systems,’’ IET Control SHEN YIN (M’12–SM’15) received the B.E.
Theory Appl., vol. 10, no. 11, pp. 1205–1214, 2016.
degree in automation from the Harbin Institute
[60] E. Kamal, A. Aitouche, R. Ghorbani, and M. Bayart, ‘‘Robust fuzzy
fault-tolerant control of wind energy conversion systems subject to sensor
of Technology, Harbin, China, in 2004, and the
faults,’’ IEEE Trans. Sustain. Energy, vol. 3, no. 2, pp. 231–241, Apr. 2012. M.Sc. degree in control and information systems
[61] X. Dai, C.-K. Li, and A. B. Rad, ‘‘An approach to tune fuzzy controllers and the Ph.D. degree in electrical engineering and
based on reinforcement learning for autonomous vehicle control,’’ IEEE information technology from the University of
Trans. Intell. Transp. Syst., vol. 6, no. 3, pp. 285–293, Sep. 2005. Duisburg-Essen, Duisburg, Germany.
[62] Z. Huang, X. Yu, H. He, J. Tan, and Z. Sun, ‘‘Parameterized batch He is currently a Professor with the Department
reinforcement learning for longitudinal control of autonomous land of Control Science and Engineering, Harbin Insti-
vehicles,’’ IEEE Trans Syst., Man, Cybern. Syst., to be published. tute of Technology. His research interests include
[Online]. Available: https://ieeexplore.ieee.org/document/8002677/, doi: model-based and data-driven fault diagnosis and prognosis in process control
10.1109/TSMC.2017.2712561. and their applications, fault-tolerant control, and big data focused on indus-
[63] H. Jiang, J. J. Zhang, W. Gao, and Z. Wu, ‘‘Fault detection, identification, trial electronics applications.
and location in smart grid based on data-driven computational methods,’’
IEEE Trans. Smart Grid, vol. 5, no. 6, pp. 2947–2956, Nov. 2014.
[64] H. Yang, Y. Jiang, and S. Yin, ‘‘Fault-tolerant control of time-delay
Markov jump systems with itô stochastic process and output disturbance
based on sliding mode observer,’’ IEEE Trans. Ind. Informat., to be OKYAY KAYNAK (F’03) received the B.Sc.
published. [Online]. Available: https://ieeexplore.ieee.org/document/ (Hons.) and Ph.D. degrees in electronic and elec-
8307165, doi: 10.1109//TII.2018.2812754. trical engineering from the University of Birm-
[65] C. Hu, R. Wang, F. Yan, and N. Chen, ‘‘Output constraint control on ingham, Birmingham, U.K., in 1969 and 1972,
path following of four-wheel independently actuated autonomous ground respectively.
vehicles,’’ IEEE Trans. Veh. Technol., vol. 65, no. 6, pp. 4033–4043, From 1972 to 1979, he held various positions
Jun. 2016. within the industry. In 1979, he joined the Depart-
[66] H. Jing, F. Jia, and Z. Liu, ‘‘Multi-objective optimal control allocation for ment of Electrical and Electronics Engineering,
an over-actuated electric vehicle,’’ IEEE Access, vol. 6, pp. 4824–4833, Bo?aziçi University, Istanbul, Turkey, where he
2018. is currently a Professor Emeritus, holding the
[67] J. Tjønnås and T. A. Johansen, ‘‘Stabilization of automotive vehicles using UNESCO Chair on Mechatronics and also a ‘‘1000 Talents Program’’ Profes-
active steering and adaptive brake control allocation,’’ IEEE Trans. Control sor at the University of Science and Technology Beijing, China. He has hold
Syst. Technol., vol. 18, no. 3, pp. 545–558, May 2010. long-term (near to or more than a year) Visiting Professor/Scholar positions
[68] M. B. Younes and A. Boukerche, ‘‘Intelligent traffic light controlling
at various institutions in Japan, Germany, USA, Singapore, and China. He has
algorithms using vehicular networks,’’ IEEE Trans. Veh. Technol., vol. 65,
no. 8, pp. 5887–5899, Aug. 2016.
authored three books and edited five and authored or co-authored almost
[69] S. Du, M. Ibrahim, M. Shehata, and W. Badawy, ‘‘Automatic license plate 400 papers that have appeared in various journals, books, and conference
recognition (ALPR): A state-of-the-art review,’’ IEEE Trans. Circuits Syst. proceedings. His current research interests include the fields of intelligent
Video Technol., vol. 23, no. 2, pp. 311–325, Feb. 2013. control and CPS.
[70] S. Yin, Y. Jiang, Y. Tian, and O. Kaynak, ‘‘A data-driven fuzzy information Dr. Kaynak has been on many committees of the IEEE and was the
granulation approach for freight volume forecasting,’’ IEEE Trans. Ind. President of the IEEE Industrial Electronics Society from 2002 to 2003.
Electron., vol. 64, no. 2, pp. 1447–1456, Feb. 2017.

47384 VOLUME 6, 2018