An autonomous control framework for advanced reactors

Đăng ngày 7/3/2019 10:07:37 PM | Thể loại: | Lần tải: 0 | Lần xem: 2 | Page: 9 | FileSize: 0.94 M | File type: PDF
An autonomous control framework for advanced reactors. The long-term economic viability of these advanced reactor plants depends on significant reductions in plant operations and maintenance costs. To accomplish these goals, intelligent control and diagnostic capabilities are needed to provide nearly autonomous operations with anticipatory maintenance. A nearly autonomous control system should enable automatic operation of a nuclear power plant while adapting to equipment faults and other upsets. It needs to have many intelligent capabilities, such as diagnosis, simulation, analysis, planning, reconfigurability, self-validation, and decision.
Contents lists available at ScienceDirect
Nuclear Engineering and Technology
journal homepage: www.elsevier.com/locate/net
Invited Article
An autonomous control framework for advanced reactors
Richard T. Wood*, Belle R. Upadhyaya, Dan C. Floyd
Department of Nuclear Engineering, University of Tennessee, 309 Pasqua Engineering Building, Knoxville, TN 37996-2300, USA
a r t i c l e
i n f o
a b s t r a c t
Article history:
Several Generation IV nuclear reactor concepts have goals for optimizing investment recovery through
Received 1 July 2017
Accepted 1 July 2017
Available online 8 July 2017
phased introduction of multiple units on a common site with shared facilities and/or recongurable
energy conversion systems. Additionally, small modular reactors are suitable for remote deployment to
support highly localized microgrids in isolated, underdeveloped regions. The long-term economic
Keywords:
Autonomous Control
Instrumentation and Control System
Small Modular Reactor
viability of these advanced reactor plants depends on signicant reductions in plant operations and
maintenance costs. To accomplish these goals, intelligent control and diagnostic capabilities are needed
to provide nearly autonomous operations with anticipatory maintenance. A nearly autonomous control
system should enable automatic operation of a nuclear power plant while adapting to equipment faults
and other upsets. It needs to have many intelligent capabilities, such as diagnosis, simulation, analysis,
planning, recongurability, self-validation, and decision. These capabilities have been the subject of
research for many years, but an autonomous control system for nuclear power generation remains as-yet
an unrealized goal. This article describes a functional framework for intelligent, autonomous control that
can facilitate the integration of control, diagnostic, and decision-making capabilities to satisfy the
operational and performance goals of power plants based on multimodular advanced reactors.
© 2017 Korean Nuclear Society, Published by Elsevier Korea LLC. This is an open access article under the
1. Introduction
primary system that is entirely or substantially fabricated within a
factory, and (3) a primary system that can be transported by truck
Advanced reactors encompass the Generation IV nuclear reactor
or rail to the plant site. In addition to suitability for factory fabri-
concepts as well as small modular reactors (SMRs). Generation IV
cation, modularity of SMRs also refers implementation of multiple
reactor concepts include both thermal and fast spectrum reactors
modules (i.e., reactor units) at a plant site. These reactors can pre-
using coolants such as gas (helium, carbon dioxide), liquid metal
sent lower capital costs than large reactors, allow for incremental
(sodium, leadebismuth), molten salt (uoride salts with dissolved
additions to generation capacity at a centralized power park, and
fuel), and supercritical water. SMRs include water-cooled integral
support multiple energy applications (e.g., process heat, desalina-
primary system reactors as well as nonwater-cooled integral and
tion, hydrogenproduction, and electricitygeneration). Additionally,
loop reactor system designs. The former types of reactors are
SMRs can serve as a highly reliable foundation for smaller grids and
generally referred to as near-term SMR designs, whereas the latter
even be remotely deployed to support highly localized microgrids
types of reactors are identied as advanced SMR designs. The
in isolated, underdeveloped regions.
subsequent discussions in this article will focus on SMRs because
Two critical factors for the economic competitiveness of SMRs
many Generation IV nuclear reactor concepts adopt or are suitable
are (1) the up-front capital cost to construct the plant and (2) the
for the SMR approach to optimizing investment recovery through
day-to-day cost of plant management. The capital cost competi-
phased introduction of multiple small units on a common site with
tiveness factor is primarily dependent on the size and complexity of
shared facilities and/or recongurable energy conversion systems.
the components that must be fabricated and the methods of
Additionally, nonwater-cooled SMR designs are subsets of larger-
installation. In this area, SMRs have a clear advantage over large
scale Generation IV nuclear reactor concepts.
plants. Because of their small size and, in many cases, simplied
An SMR is generally characterized by: (1) an electrical gener-
nuclear island congurations, it is expected that capital costs will
ating capacity of less than 300 MWe (megawatt electric), (2) a
be much lower for SMRs compared to those of large, Generation
IIIþ light-water reactors. Advanced SMRs, which use coolants other
* Corresponding author.
E-mail address: rwood11@utk.edu (R.T. Wood).
than water as the primary heat transport medium, introduce
several passive safety concepts and controllability features that
1738-5733/© 2017 Korean Nuclear Society, Published by Elsevier Korea LLC. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/
R.T. Wood et al. / Nuclear Engineering and Technology 49 (2017) 896e904
897
further reduce the complexity of primary system designs by elim-
automated control for an SMR is clearly feasible under optimum
inating redundant components and systems.
circumstances. Autonomous control is primarily intended to ac-
The latter competitiveness factor for SMRs (i.e., plant manage-
count for the nonoptimum circumstances when degradation, fail-
ment costs) is strongly affected by the loss of economy of scale. The
ure, and other off-normal events challenge the performance of the
most signicant controllable contributor to day-to-day costs arises
reactor, and the capability for immediate human intervention is
from operations and maintenance (O&M) activities, which heavily
constrained. There are clear gaps in the development and demon-
depend on stafng size and plant availability. The operation of a
stration of autonomous control capabilities for the specic domain
nuclear power plant is labor intensive. The O&M staff at a plant is
of nuclear power operations.
composed of operator teams for each shift at each unit, and on-site
maintenance personnel can involve a large number of technicians
2.1. Advanced control in nuclear power applications
and specialists. The current US nuclear industry average for O&M
staff is roughly one person per every 2 megawatts of generated
In the nuclear power industry, single-input, single-output clas-
power. Stafng size is affected by regulatory constraints, which
sical control has been the primary means of automating individual
establish minimum licensed operator and senior operator stafng
control loops. The use of multivariate control, such as three-
requirements for each reactor unit. These stafng requirements are
element controllers for steam generators, has been employed in
primarily driven by resource demands to respond to transients and
some cases. In a few cases, efforts were made to coordinate the
accidents and are based on traditional operational models with
action of individual control loops, based on an overall control goal,
limited automation. Without a signicantly higher degree of
and extend the range of automated control.
automation than is customary for current nuclear power plants,
Current Generation IIIþ reactor designs involve a substantial
high stafng levels relative to unit power production will pose the
increase in the use of digital I&C technology, but their control
threat of unsustainable O&M costs for SMRs.
systems maintain traditional control strategies. One of the most
The benets of SMRs can include reduced nancial risk, opera-
fully digital plants currently in operation in the United States is the
tional exibility, and modular allocation of power production ca-
Oconee Nuclear Station [14]. The three units at Oconee have digital
pacity. Achieving these benets can lead to a new paradigm for
reactor protection systems and a digital integrated control system
plant design, construction, and management to provide for multi-
(ICS). The digital ICS coordinates the main control actions of mul-
unit, multiproduct-stream generating stations while addressing the
tiple control loops through an integrated master controller that
need to compensate for reduced economy-of-scale savings. How-
establishes feedforward control demands based on desired overall
ever, there are technology needs that must be addressed to resolve
core thermal power. The ICS also has provisions for supplementary
challenges to establishing this new paradigm [1]. Automation to the
support
actions
among
control
loops
to
facilitate
optimized
point of near autonomy is the enabling technology that can support
performance.
achievement of the desired operational and stafng efciencies
The application of most advanced techniques for nuclear power
(i.e., the economy of automation).
control has primarily been the domain of universities and national
laboratories. Some of the techniques employed in controls research
2. State of the technology
for both power and research reactors include adaptive robust
control for the Experimental Breeder Reactor II, fuzzy logic control
To support a technology assessment, the authors conducted an
for power transition, H-innity control and genetic algorithm-
investigation of autonomous control. Control systems with varying
based control for steam generators, and neural network control
levels of autonomy have been employed in robotics, transportation,
for power distribution in a reactor core, and model predictive
spacecraft, and manufacturing applications. For manufacturing and
control to enable fault tolerance and reconguration features for
robotics [2], much of the work involves augmenting automation of
primary power control of advanced reactors. Proceedings of past
routine tasks with the capability to diagnose and adapt to varying
International Topical Meetings on Nuclear Plant Instrumentation,
conditions,
often based
on
a
constrained,
predened
set
of
Control and HumaneMachine Interface Technologies provide a
responsive actions. Robotic applications can also employ un-
useful compendium of ndings from such research activities
manned maneuverable platforms to enable transit within harsh or
[15e23].
remote environments. The basis for this autonomy is equivalent to
Aspartof the AdvancedLiquid MetalReactor(ALMR)Programfor
that of unmanned vehicles (both aerial and ground) [3], which
the US Department of Energy, the Oak Ridge National Laboratory
involve autonomous capabilities as part of guidance, navigation,
developed the concept of supervisory control for multimodular
and control systems. In transportation, recent developments have
advanced reactors [24,25]. Recent activity on the DOE (U.S. Depart-
focused on self-driving automobiles [4,5]. Deep-space robotic
ment of Energy) Advanced Reactor Technologies Program has
missions have been the primary focus of autonomous ight control
extended that concept for advanced multimodular SMR plants [26].
for space exploration [6].
Although the level of autonomy and the specic control algo-
2.2. Autonomy in space exploration
rithms differ, each case illustrates key characteristics and a high-
level functional framework to enable autonomy. Overviews of
National Aeronautics and Space Administration has pursued
autonomous control characteristics, capabilities, and applications
autonomy for spacecraft and surface exploration vehicles (e.g., ro-
were found that establish the existing experience and current
vers) to reduce mission costs, increase efciency for communica-
technology
readiness
[7e13].
The
desirable
characteristics
of
tions
between
ground
control
and
the
vehicle,
and
enable
autonomous control include intelligence, robustness, optimization,
independent operation of the vehicle during times of communi-
exibility, adaptability, and reliability.
cations blackout. For rovers, functional autonomy addresses navi-
Although various degrees of autonomy have been demonstrated
gation, target identication, and science package manipulation. For
in the cited application domains, autonomous control has not been
spacecraft, functional autonomy has focused on automated guid-
implemented for an operating nuclear power plant nor have it been
ance, navigation, and control.
extensively
developed
for
any
emerging
advanced
emerging
Autonomy for rovers has progressed during the last two decades
advanced reactor concept. Current automated control technologies
with prominent examples from efforts to explore the surface of
for nuclear power plants are reasonably mature, and highly
Mars. The Mars Pathnder rover, Sojourner, explored the Martian
898
R.T. Wood et al. / Nuclear Engineering and Technology 49 (2017) 896e904
terrain beginning in July 1997 [27]. The Sojourner had very limited
life-limited components (such as batteries and actuators), adapta-
autonomy to enable navigation and provide for resource manage-
tion to changing or degrading conditions, and validation and
ment and contingency response. Because it only provided super-
maintenance of control system performance.
vised autonomy, repetitive ground monitoring was required. In
Key characteristics of autonomy include intelligence, robust-
January 2004, Spirit and Opportunity, the twin Mars Exploration
ness, optimization, exibility, and adaptability. Intelligence facili-
Rovers (MERs), began a surface exploration mission that has
tates minimal or no reliance on human intervention and can
continued into 2006. These rovers employ expanded autonomy
accommodate an integrated, whole system approach to control. It
over what was feasible for Sojourner and provide model-based re-
implies embedded decision-making and management/planning
covery, resource management, and autonomous planning capabil-
authority. Intelligence in control provides for anticipatory action
ities in addition to autonomous obstacle detection and navigation.
based on system knowledge and event prediction. To support
The integration software architecture used to facilitate MER au-
control and decision, real-time diagnostic/prognostic capabilities
tonomy is the Coupled Layer Architecture for Robotic Autonomy
are important for state identication and health/condition moni-
(CLARAty) [28]. CLARAty provides a dual-layer architecture con-
toring. Additionally, self-validation is an aspect of intelligence that
sisting of a decision layer for articial intelligence (AI) software and
addresses data, command, and system performance assessment
a functional layer forcontrols implementations. Implicit granularity
and response.
in each layer allows for a functional hierarchy with nested
In addition to providing an environmentally rugged imple-
capabilities.
mentation, robustness is addressed by accounting for design un-
Spacecraft autonomy has been demonstrated with the Deep
certainties and unmodeled dynamics. Fault management is an
Space 1 mission. Deep Space 1 was launched in October 1998 as a
important consideration in achieving robustness. Fault manage-
test platform to validate high-risk advanced technologies in space
ment involves techniques such as fault avoidance, fault removal,
[6]. In addition to demonstrating autonomous navigation of the
fault tolerance, and fault forecasting. Additionally, robustness can
spacecraft, a principal experiment involved demonstration of the
also involve self-maintenance or self-healing. This capability is
Remote Agent AI system for on-board planning and execution of
promoted through means such as captured design knowledge and
spacecraft activities.
self-correcting features, prognostics to identify incipient failure,
Finally, an approach for fault-tolerant control of the SP-100
and fault detection and isolation.
reactor system was developed by Upadhyaya et al [29] and dem-
Optimization implies rapid response to demands, minimal de-
onstrates the feasibility of applying this method for space ssion
viation from target conditions, and efcient actuator actions.
reactors, either for propulsion or as an energy source.
Optimized control can be facilitated by self-tuning and other forms
of adaptation. Flexibility and adaptability are enabled by diverse
3. Autonomous control functional denition
measurements, multiple communication options, and alternate
control solutions. Functional recongurability facilitates the effec-
3.1. The nature of autonomy
tive use of these systems options, whereas an inherent redesign
capability permits adaptation to unanticipated conditions.
There is a distinction between automated control and autono-
The characteristics discussed above represent the possibilities of
mous control. Consideration of the Greek root words illustrates the
autonomy, but they do not constitute a necessary set. Therefore,
difference.
Automatos
means
self-acting,
whereas
autonomos
autonomous control can be viewed as providing a spectrum of
means independent. Similarly, automated control involves self-
capabilities
with
automated
control
representing
the
lowest
action, whereas autonomous control involves independent action.
extreme or baseline of the continuum. The incorporation of
Autonomous control implies an embedded intelligence. Although
increasing intelligence and fault tolerance moves the control ca-
automation includes at least a limited inherent authority within the
pabilities further along the spectrum. The higher degrees of au-
control system, automated control often consists of straightforward
tonomy are characterized by greater fault management, more
automatic execution of repetitive basic actions. It is clear that
embedded planning and goal setting, and even self-healing. The
autonomous control encompasses automated control.
realization of full autonomy involves learning, evolving, and stra-
Automated control provides control actions that result from
tegizing independent of human interaction or supervision.
xed set of algorithms with typically limited global state determi-
nation. As a result, automated control is often implemented as
3.2. Near-autonomous SMR plant control
rigidly dened individual control loops rather than as fully inte-
grated process/plant control. Although automated control requires
Autonomous control functions for an advanced reactor can be
no real-time operator action for normal operational events, most
dened based on the expected operational modes, which include
signicant decision-making is left to the human rather than
startup, normal power operation, reactor protection, contingent
incorporated as part of the control system. In contrast, autonomous
operation, and end-of-cycle shutdown. As a minimum requirement
control integrates control, diagnostic, and decision capabilities. A
of autonomy, the SMR plant control system must be able to switch
exible functional architecture provides the capability to adapt to
between normal operational modes automatically (i.e., automatic
evolving conditions and operational constraints and even support
control). Additionally, reactor protective action must be available if
self-maintenance over the control system lifetime. While auto-
the desired operational conditions cannot be achieved.
mated control is common in numerous applications, autonomous
The phases of poweroperation include powerascension, steady-
control is more difcult to achieve, and the experience base is very
state power and load following, and power reduction. Under
limited.
normal conditions, power operation can be relatively simple, with
Autonomy extends the scope of primary control functions. Such
inherent feedback effects serving to maintain stability and provide
capabilities can consist of automated control during all operating
the means for load following in response to minor uctuations.
modes,
process
performance
optimization
(e.g.,
self-tuning),
Thermal load transients (e.g., turbine failure, loss of heat sink) can
continuous monitoring, and diagnosis of performance indicators
be treated as off-normal events. Other off-normal events include
as well as trends for operational and safety-related parameters,
load/power interruptions, actuator degradation or failure, actuator
diagnosis of component health, exible control to address both
signal interruption or interference, heat removal system degrada-
anticipated and unanticipated events and to provide protection of
tion or damage, control processor fault, rare-event software error,
HƯỚNG DẪN DOWNLOAD TÀI LIỆU

Bước 1:Tại trang tài liệu slideshare.vn bạn muốn tải, click vào nút Download màu xanh lá cây ở phía trên.
Bước 2: Tại liên kết tải về, bạn chọn liên kết để tải File về máy tính. Tại đây sẽ có lựa chọn tải File được lưu trên slideshare.vn
Bước 3: Một thông báo xuất hiện ở phía cuối trình duyệt, hỏi bạn muốn lưu . - Nếu click vào Save, file sẽ được lưu về máy (Quá trình tải file nhanh hay chậm phụ thuộc vào đường truyền internet, dung lượng file bạn muốn tải)
Có nhiều phần mềm hỗ trợ việc download file về máy tính với tốc độ tải file nhanh như: Internet Download Manager (IDM), Free Download Manager, ... Tùy vào sở thích của từng người mà người dùng chọn lựa phần mềm hỗ trợ download cho máy tính của mình  
2 lần xem

An autonomous control framework for advanced reactors. The long-term economic viability of these advanced reactor plants depends on significant reductions in plant operations and maintenance costs. To accomplish these goals, intelligent control and diagnostic capabilities are needed to provide nearly autonomous operations with anticipatory maintenance. A nearly autonomous control system should enable automatic operation of a nuclear power plant while adapting to equipment faults and other upsets. It needs to have many intelligent capabilities, such as diagnosis, simulation, analysis, planning, reconfigurability, self-validation, and decision..

Nội dung

Nuclear Engineering and Technology 49 (2017) 896e904 Contents lists available at ScienceDirect Nuclear Engineering and Technology journal homepage: www.elsevier.com/locate/net Invited Article An autonomous control framework for advanced reactors Richard T. Wood*, Belle R. Upadhyaya, Dan C. Floyd Department of Nuclear Engineering, University of Tennessee, 309 Pasqua Engineering Building, Knoxville, TN 37996-2300, USA a r t i c l e i n f o a b s t r a c t Article history: Received 1 July 2017 Accepted 1 July 2017 Available online 8 July 2017 Keywords: Autonomous Control Instrumentation and Control System Small Modular Reactor 1. Introduction Several Generation IV nuclear reactor concepts have goals for optimizing investment recovery through phased introduction of multiple units on a common site with shared facilities and/or reconfigurable energy conversion systems. Additionally, small modular reactors are suitable for remote deployment to support highly localized microgrids in isolated, underdeveloped regions. The long-term economic viability of these advanced reactor plants depends on significant reductions in plant operations and maintenance costs. To accomplish these goals, intelligent control and diagnostic capabilities are needed to provide nearly autonomous operations with anticipatory maintenance. A nearly autonomous control system should enable automatic operation of a nuclear power plant while adapting to equipment faults and other upsets. It needs to have many intelligent capabilities, such as diagnosis, simulation, analysis, planning, reconfigurability, self-validation, and decision. These capabilities have been the subject of research for many years, but an autonomous control system for nuclear power generation remains as-yet an unrealized goal. This article describes a functional framework for intelligent, autonomous control that can facilitate the integration of control, diagnostic, and decision-making capabilities to satisfy the operational and performance goals of power plants based on multimodular advanced reactors. © 2017 Korean Nuclear Society, Published by Elsevier Korea LLC. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/). primary system that is entirely or substantially fabricated within a factory, and (3) a primary system that can be transported by truck Advanced reactors encompass the Generation IV nuclear reactor concepts as well as small modular reactors (SMRs). Generation IV reactor concepts include both thermal and fast spectrum reactors using coolants such as gas (helium, carbon dioxide), liquid metal (sodium, leadebismuth), molten salt (fluoride salts with dissolved fuel), and supercritical water. SMRs include water-cooled integral primary system reactors as well as nonwater-cooled integral and loop reactor system designs. The former types of reactors are generally referred to as near-term SMR designs, whereas the latter types of reactors are identified as advanced SMR designs. The subsequent discussions in this article will focus on SMRs because many Generation IV nuclear reactor concepts adopt or are suitable for the SMR approach to optimizing investment recovery through phased introduction of multiple small units on a common site with shared facilities and/or reconfigurable energy conversion systems. Additionally, nonwater-cooled SMR designs are subsets of larger-scale Generation IV nuclear reactor concepts. An SMR is generally characterized by: (1) an electrical gener-ating capacity of less than 300 MWe (megawatt electric), (2) a * Corresponding author. E-mail address: rwood11@utk.edu (R.T. Wood). or rail to the plant site. In addition to suitability for factory fabri-cation, modularity of SMRs also refers implementation of multiple modules (i.e., reactor units) at a plant site. These reactors can pre-sent lower capital costs than large reactors, allow for incremental additions to generation capacity at a centralized power park, and support multiple energy applications (e.g., process heat, desalina-tion, hydrogenproduction, and electricitygeneration). Additionally, SMRs can serve as a highly reliable foundation for smaller grids and even be remotely deployed to support highly localized microgrids in isolated, underdeveloped regions. Two critical factors for the economic competitiveness of SMRs are (1) the up-front capital cost to construct the plant and (2) the day-to-day cost of plant management. The capital cost competi-tiveness factor is primarily dependent on the size and complexity of the components that must be fabricated and the methods of installation. In this area, SMRs have a clear advantage over large plants. Because of their small size and, in many cases, simplified nuclear island configurations, it is expected that capital costs will be much lower for SMRs compared to those of large, Generation IIIþ light-water reactors. Advanced SMRs, which use coolants other than water as the primary heat transport medium, introduce several passive safety concepts and controllability features that http://dx.doi.org/10.1016/j.net.2017.07.001 1738-5733/© 2017 Korean Nuclear Society, Published by Elsevier Korea LLC. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/ licenses/by-nc-nd/4.0/). R.T. Wood et al. / Nuclear Engineering and Technology 49 (2017) 896e904 897 further reduce the complexity of primary system designs by elim-inating redundant components and systems. The latter competitiveness factor for SMRs (i.e., plant manage-ment costs) is strongly affected by the loss of economy of scale. The most significant controllable contributor to day-to-day costs arises from operations and maintenance (O&M) activities, which heavily depend on staffing size and plant availability. The operation of a nuclear power plant is labor intensive. The O&M staff at a plant is composed of operator teams for each shift at each unit, and on-site maintenance personnel can involve a large number of technicians and specialists. The current US nuclear industry average for O&M staff is roughly one person per every 2 megawatts of generated power. Staffing size is affected by regulatory constraints, which establish minimum licensed operator and senior operator staffing requirements for each reactor unit. These staffing requirements are primarily driven by resource demands to respond to transients and accidents and are based on traditional operational models with limited automation. Without a significantly higher degree of automation than is customary for current nuclear power plants, high staffing levels relative to unit power production will pose the threat of unsustainable O&M costs for SMRs. The benefits of SMRs can include reduced financial risk, opera-tional flexibility, and modular allocation of power production ca-pacity. Achieving these benefits can lead to a new paradigm for plant design, construction, and management to provide for multi-unit, multiproduct-stream generating stations while addressing the need to compensate for reduced economy-of-scale savings. How-ever, there are technology needs that must be addressed to resolve challenges to establishing this new paradigm [1]. Automation to the point of near autonomy is the enabling technology that can support achievement of the desired operational and staffing efficiencies (i.e., the economy of automation). 2. State of the technology To support a technology assessment, the authors conducted an investigation of autonomous control. Control systems with varying levels of autonomy have been employed in robotics, transportation, spacecraft, and manufacturing applications. For manufacturing and robotics [2], much of the work involves augmenting automation of routine tasks with the capability to diagnose and adapt to varying conditions, often based on a constrained, predefined set of responsive actions. Robotic applications can also employ un-manned maneuverable platforms to enable transit within harsh or remote environments. The basis for this autonomy is equivalent to that of unmanned vehicles (both aerial and ground) [3], which involve autonomous capabilities as part of guidance, navigation, and control systems. In transportation, recent developments have focused on self-driving automobiles [4,5]. Deep-space robotic missions have been the primary focus of autonomous flight control for space exploration [6]. Although the level of autonomy and the specific control algo-rithms differ, each case illustrates key characteristics and a high-level functional framework to enable autonomy. Overviews of autonomous control characteristics, capabilities, and applications were found that establish the existing experience and current technology readiness [7e13]. The desirable characteristics of autonomous control include intelligence, robustness, optimization, flexibility, adaptability, and reliability. Although various degrees of autonomy have been demonstrated in the cited application domains, autonomous control has not been implemented for an operating nuclear power plant nor have it been extensively developed for any emerging advanced emerging advanced reactor concept. Current automated control technologies for nuclear power plants are reasonably mature, and highly automated control for an SMR is clearly feasible under optimum circumstances. Autonomous control is primarily intended to ac-count for the nonoptimum circumstances when degradation, fail-ure, and other off-normal events challenge the performance of the reactor, and the capability for immediate human intervention is constrained. There are clear gaps in the development and demon-stration of autonomous control capabilities for the specific domain of nuclear power operations. 2.1. Advanced control in nuclear power applications In the nuclear power industry, single-input, single-output clas-sical control has been the primary means of automating individual control loops. The use of multivariate control, such as three-element controllers for steam generators, has been employed in some cases. In a few cases, efforts were made to coordinate the action of individual control loops, based on an overall control goal, and extend the range of automated control. Current Generation IIIþ reactor designs involve a substantial increase in the use of digital I&C technology, but their control systems maintain traditional control strategies. One of the most fully digital plants currently in operation in the United States is the Oconee Nuclear Station [14]. The three units at Oconee have digital reactor protection systems and a digital integrated control system (ICS). The digital ICS coordinates the main control actions of mul-tiple control loops through an integrated master controller that establishes feedforward control demands based on desired overall core thermal power. The ICS also has provisions for supplementary support actions among control loops to facilitate optimized performance. The application of most advanced techniques for nuclear power control has primarily been the domain of universities and national laboratories. Some of the techniques employed in controls research for both power and research reactors include adaptive robust control for the Experimental Breeder Reactor II, fuzzy logic control for power transition, H-infinity control and genetic algorithm-based control for steam generators, and neural network control for power distribution in a reactor core, and model predictive control to enable fault tolerance and reconfiguration features for primary power control of advanced reactors. Proceedings of past International Topical Meetings on Nuclear Plant Instrumentation, Control and HumaneMachine Interface Technologies provide a useful compendium of findings from such research activities [15e23]. Aspartof the AdvancedLiquid MetalReactor(ALMR)Programfor the US Department of Energy, the Oak Ridge National Laboratory developed the concept of supervisory control for multimodular advanced reactors [24,25]. Recent activity on the DOE (U.S. Depart-ment of Energy) Advanced Reactor Technologies Program has extended that concept for advanced multimodular SMR plants [26]. 2.2. Autonomy in space exploration National Aeronautics and Space Administration has pursued autonomy for spacecraft and surface exploration vehicles (e.g., ro-vers) to reduce mission costs, increase efficiency for communica-tions between ground control and the vehicle, and enable independent operation of the vehicle during times of communi-cations blackout. For rovers, functional autonomy addresses navi-gation, target identification, and science package manipulation. For spacecraft, functional autonomy has focused on automated guid-ance, navigation, and control. Autonomy for rovers has progressed during the last two decades with prominent examples from efforts to explore the surface of Mars. The Mars Pathfinder rover, Sojourner, explored the Martian 898 R.T. Wood et al. / Nuclear Engineering and Technology 49 (2017) 896e904 terrain beginning in July 1997 [27]. The Sojourner had very limited autonomy to enable navigation and provide for resource manage-ment and contingency response. Because it only provided super-vised autonomy, repetitive ground monitoring was required. In January 2004, Spirit and Opportunity, the twin Mars Exploration Rovers (MERs), began a surface exploration mission that has continued into 2006. These rovers employ expanded autonomy over what was feasible for Sojourner and provide model-based re-covery, resource management, and autonomous planning capabil-ities in addition to autonomous obstacle detection and navigation. The integration software architecture used to facilitate MER au-tonomy is the Coupled Layer Architecture for Robotic Autonomy (CLARAty) [28]. CLARAty provides a dual-layer architecture con-sisting of a decision layer for artificial intelligence (AI) software and a functional layer forcontrols implementations. Implicit granularity in each layer allows for a functional hierarchy with nested capabilities. Spacecraft autonomy has been demonstrated with the Deep Space 1 mission. Deep Space 1 was launched in October 1998 as a test platform to validate high-risk advanced technologies in space [6]. In addition to demonstrating autonomous navigation of the spacecraft, a principal experiment involved demonstration of the Remote Agent AI system for on-board planning and execution of spacecraft activities. Finally, an approach for fault-tolerant control of the SP-100 reactor system was developed by Upadhyaya et al [29] and dem-onstrates the feasibility of applying this method for space fission reactors, either for propulsion or as an energy source. 3. Autonomous control functional definition 3.1. The nature of autonomy There is a distinction between automated control and autono-mous control. Consideration of the Greek root words illustrates the difference. Automatos means self-acting, whereas autonomos means independent. Similarly, automated control involves self-action, whereas autonomous control involves independent action. Autonomous control implies an embedded intelligence. Although automation includes at least a limited inherent authority within the control system, automated control often consists of straightforward automatic execution of repetitive basic actions. It is clear that autonomous control encompasses automated control. Automated control provides control actions that result from fixed set of algorithms with typically limited global state determi-nation. As a result, automated control is often implemented as rigidly defined individual control loops rather than as fully inte-grated process/plant control. Although automated control requires no real-time operator action for normal operational events, most significant decision-making is left to the human rather than incorporated as part of the control system. In contrast, autonomous control integrates control, diagnostic, and decision capabilities. A flexible functional architecture provides the capability to adapt to evolving conditions and operational constraints and even support self-maintenance over the control system lifetime. While auto-mated control is common in numerous applications, autonomous control is more difficult to achieve, and the experience base is very limited. Autonomy extends the scope of primary control functions. Such capabilities can consist of automated control during all operating life-limited components (such as batteries and actuators), adapta-tion to changing or degrading conditions, and validation and maintenance of control system performance. Key characteristics of autonomy include intelligence, robust-ness, optimization, flexibility, and adaptability. Intelligence facili-tates minimal or no reliance on human intervention and can accommodate an integrated, whole system approach to control. It implies embedded decision-making and management/planning authority. Intelligence in control provides for anticipatory action based on system knowledge and event prediction. To support control and decision, real-time diagnostic/prognostic capabilities are important for state identification and health/condition moni-toring. Additionally, self-validation is an aspect of intelligence that addresses data, command, and system performance assessment and response. In addition to providing an environmentally rugged imple-mentation, robustness is addressed by accounting for design un-certainties and unmodeled dynamics. Fault management is an important consideration in achieving robustness. Fault manage-ment involves techniques such as fault avoidance, fault removal, fault tolerance, and fault forecasting. Additionally, robustness can also involve self-maintenance or self-healing. This capability is promoted through means such as captured design knowledge and self-correcting features, prognostics to identify incipient failure, and fault detection and isolation. Optimization implies rapid response to demands, minimal de-viation from target conditions, and efficient actuator actions. Optimized control can be facilitated by self-tuning and other forms of adaptation. Flexibility and adaptability are enabled by diverse measurements, multiple communication options, and alternate control solutions. Functional reconfigurability facilitates the effec-tive use of these systems options, whereas an inherent redesign capability permits adaptation to unanticipated conditions. The characteristics discussed above represent the possibilities of autonomy, but they do not constitute a necessary set. Therefore, autonomous control can be viewed as providing a spectrum of capabilities with automated control representing the lowest extreme or baseline of the continuum. The incorporation of increasing intelligence and fault tolerance moves the control ca-pabilities further along the spectrum. The higher degrees of au-tonomy are characterized by greater fault management, more embedded planning and goal setting, and even self-healing. The realization of full autonomy involves learning, evolving, and stra-tegizing independent of human interaction or supervision. 3.2. Near-autonomous SMR plant control Autonomous control functions for an advanced reactor can be defined based on the expected operational modes, which include startup, normal power operation, reactor protection, contingent operation, and end-of-cycle shutdown. As a minimum requirement of autonomy, the SMR plant control system must be able to switch between normal operational modes automatically (i.e., automatic control). Additionally, reactor protective action must be available if the desired operational conditions cannot be achieved. The phases of poweroperation include powerascension, steady-state power and load following, and power reduction. Under normal conditions, power operation can be relatively simple, with inherent feedback effects serving to maintain stability and provide the means for load following in response to minor fluctuations. modes, process performance optimization (e.g., self-tuning), Thermal load transients (e.g., turbine failure, loss of heat sink) can continuous monitoring, and diagnosis of performance indicators as well as trends for operational and safety-related parameters, diagnosis of component health, flexible control to address both anticipated and unanticipated events and to provide protection of be treated as off-normal events. Other off-normal events include load/power interruptions, actuator degradation or failure, actuator signal interruption or interference, heat removal system degrada- tion or damage, control processor fault, rare-event software error, R.T. Wood et al. / Nuclear Engineering and Technology 49 (2017) 896e904 899 sensor failure, sensor signal interruption or interference, sensor drift, signal conditioning electronics drift, sensor noise increase, and communication failures or retransmissions. The most likely immediate protective action for a significant event would consist of a rapid power runback. Contingent operation occurs when SMR operation may be restricted because of power system limitations, such as component failures, degradation or loss of heat sink, or station blackout. The response to off-normal events is where autonomy becomes especially relevant. The autonomous response includes a reflexive element and a deliberative element. The first element addresses reactor protection. Unlike conventional reactor operational con-cepts, in which the primary defense against potentially adverse conditions resulting from off-normal events is to scram the reactor, the objective of autonomous control is to limit the progression of off-normal events and minimize the need for shutdown. This is especially true in situations where the nuclear power plant is the stabilizing generation source on a small electric grid. Thus, an enhanced, layered reactor protection can be provided through di-versity and defense-in-depth to anticipate potential challenges to power operation. A limitation system is one means of protecting the reactor while minimizing the risk of costly scrams. This is accomplished by defining acceptable operational regimes and overriding control actions that would drive the reactor conditions to acceptable operational states that do not violate the limitation boundaries. In effect, the limitation system acts as a bounding system whose primary purpose is to provide a check against op-erations outside of analyzed conditions. The principle response of the limitation system would be to run back the reactor power to assume a safe low-power conditionwhen necessary. In cases driven by the operational objective that power should remain available to support critical power needs and ensure grid stability, the SMR plant control system must provide the capability to address off-normal events over an extended range of operating conditions without challenging the safety boundary of the reactor and, thus, triggering reactor scram. The second element of the response to off-normal events ad-dresses availability assurance. The deliberative nature (i.e., deter-mination and decision) of this element contributes the most relevant attribute of autonomous control that distinguishes it from conventional automation. In the operational control context, the autonomous control functionality involves detection and immedi-ateresponse todegraded or failureconditions. Fault management is a crucial part of this element of autonomous control, which pro-vides for detection, diagnosis, and adaptation (or reconfiguration) given changing plantorequipment conditions. An additional aspect of this deliberative element is the monitoring, diagnosis, and vali-dation of control system and reactor performance. Through this capability, the plant control system is able to identify incipient events (transients or failures) for anticipatory rather than reac-tionary action, determine measures to protect life-limited or vulnerable components, and ensure continued dependable opera-tion of the power plant. As noted, autonomous control functionality revolves around automated control for normal operational modes. In essence, the primary function of the control system is command generation to achieve the desired operational state. Additional functionality to support confirmation of control system performance includes fea-tures such as command verification, control coordination with interconnected systems, and strategyenforcement. Mechanisms for implementing these features can involve multiple diverse algo-rithms for comparison with the principal controller command, in-clusion of feedforward action or some representation of unmodeled dynamics (e.g., exogenous variables) in control algorithms, event management according to predetermined sequences of events, and adaptation of the control strategy. Performance management as part of the autonomous control functionality involves continuously assessing the condition of the control system and the reactor to identify when predetermined adjustments to the controller should be invoked. The needed as-sessments include monitoring control system effectiveness, iden-tifying the dynamic state of the plant, and determining the condition of key components. Methods that can be employed are state estimation algorithms, process system diagnostics, compo-nent condition monitoring, and control parameter adaptation. Data management and communications are related capabilities with traditional and autonomous functionality intended to support autonomy and system integration. Data acquisition and signal processing methods provide the data needed for control and monitoring, whereas signal validation adds information about data quality. For communications, the functional elements include device-level data and control signals, system-level information and commands, and plant-level status and demands. The effective integration of data and information at each level requires a well-defined functional architecture with a capable physical infrastruc-ture that supports reliable, timely information flow. Desired functionality for fault management includes detection and identification of field device faults, change tracking for system parameters, detection of off-normal transients and identification of anticipated events, and configuration control. Field device moni-toring can be accomplished through model-based and/or data-driven algorithms. Parameter tracking can involve empirical models or first principles estimation. Each capability can be used to facilitate an adjustable system dynamic model that can be used for fault prediction or control system performance validation. Finally, configuration control functions are needed to manage transitions among predefined control strategies or algorithms for the auton-omous control system. This is essential for effective fault recovery. To illustrate the autonomous functionality that can be provided for the SMR plant control system, two fault management scenarios are considered in which detection and response are described. The first scenario relates to fault adaptation in the case of sensor failure. The indicators from surveillance and diagnostic functions that the plant control system can employ include divergence of redundant measurements, conflict between predicted (based on analytical or relational estimation) and measured values, and detection and isolation of a confirmed fault. The prospective response can include substitution of a redundant measurement or utilization of a diverse measurement. An example of the latter would be using neutron flux instead of temperature (i.e., core thermal power) as a power measurement. Switching to an alternate control algorithm may prove necessary for faulted or suspect measurements. The second scenario relates to fault avoidance in the case of a degrading actuator. The indicators of an incipient failure can be prediction of actuator failure based on prognostic modeling (e.g., fault forecasting) or detection of sluggish response to commands. The prospective response can be to switch to an alternate control strategy to avoid incipient failure by reducing stress on the suspect component. An example would be utilizing manipulation of core heat removal (e.g., coolant density change) instead of direct reac-tivity insertion (e.g., control element movement) to control reactor power. 3.3. Enabling autonomous control Autonomous control must be addressed early in the design of the SMR to determine the degree of autonomy required. Opera-tional requirements, technology readiness, design trade-offs, and resource constraints will affect the autonomous capabilities to be included. The extent to which the key characteristics of autonomy 900 R.T. Wood et al. / Nuclear Engineering and Technology 49 (2017) 896e904 are realized depends on the level of responsibility that is to be entrusted to the autonomous control system and the degree of operational risk that the autonomous control system must mitigate. Several factors can influence the degree of autonomy selected for a plant control system. These factors include the potential for continuous direct human interaction (which may be limited because of shared operator supervision responsibilities over mul-tiple units or because of constrained on-site staffing at remote in-stallations), performance goals, complexity of system demands, technological constraints, operational risk considerations, and the balance between simplicity (i.e., reliability) and complexity (i.e., the capacity to detect and adapt). The trade between reliability and operational assurance profoundly affects the level of autonomy employed for plant control. Although having a highly reliable plant control system is important, that fact is of limited value if the control system cannot accommodate plant degradation without immediate human intervention or scram. In such a case, the result is a highly reliable control system that becomes ineffective because the plant has changed. Finally, as previously described, the experience base for auton-omous control is not deep. In particular, autonomous control has not been implemented for an operating nuclear power plant. The technology gaps indicated by investigation of the state of the technology for reactor control in general and autonomous control in particular indicate research, development, and demonstration (RD&D) activities that need to be accomplished to fully realize the goal of autonomous control for a SMR. Key elements of the needed RD&D effort involve establishing a suitable functional architecture, developing foundational modules to support autonomy, and levels of intelligence, into the decision layer. Essentially, the deliberative and procedural functionalities are merged into an architectural layer that parallels the functional layer and provides a common database to support decision-making. Additionally, a system granularity dimension is maintained to explicitly represent the system hierarchies of the functional layer and the multiple planning horizons of the decision layer. The functional layer is an object-oriented hierarchy that pro-vides access to the capabilities of the plant/system hardware and serves as the interface for the decision layer to the subject (robot, spacecraft, plant) under control. The interaction between the two layers depends on the relative granularity of each layer at the interface. At lower granularity, the decision layer has almost direct access to the basic capabilities of the plant/system. At higher granularity, the decision layer provides high-level commands that are broken down and executed by the intelligent control capability of the functional layer. The decision layer provides functionality to break down goals into objectives, establish a sequential task ordering based on the plant/system state and known constraints, and assess the capability of the functional layer to implement those commands. At lower granularity within the decision layer, execu-tive functions such as procedure enforcement are dominant, whereas at higher granularity, planning functions such as goal determination and strategy development are dominant. There is an architectural approach for nearly autonomous con-trol systems that has been developed through simulated nuclear power applications (see Fig. 2). As part of research into advanced multimodular nuclear reactor concepts, such as the ALMR, the In- ternational Reactor Innovative and Secure (IRIS), and representative demonstrating the integrated application of autonomous capabilities. advanced SMR concepts, a supervisory control system architecture was devised [24e26]. This approach provides a framework for autonomous control while supporting a high-level interface with 4. Functional architecture for autonomous control 4.1. Architectural approaches As observed from examples of autonomous control for nuclear and space applications, the principal functional architectures that have been employed, in most cases, involve some form of hierar-chical framework with varying distributions of intelligence. A three-level hierarchy is typical for robotic applications [8,30,31]. The three layers in top-to-bottom hierarchical order are the planner layer, the executive layer, and the functional layer. The general concept of the hierarchy is that commands are issued by higher levels to lower levels, and response data flows from lower levels to higher levels in the multi-tiered framework. Intelligence increases with increasing level within the hierarchy. Each of the three interacting tiers has a principal role. Basically, the functional operations staff, who can act as plant supervisors. The final au-thority for decisions and goal setting remains with the human, but the control system assumes expanded responsibilities for normal control action, abnormal event response, and system fault toler-ance. The autonomous control framework allows integration of controllers and diagnostics at the subsystem level with command and decision modules at higher levels. The autonomous control system architecture is hierarchical and recursive. Each node in the hierarchy (except for the terminal nodes at the base) is a supervisory module. The supervisory control modules at each level within the hierarchy respond to goals and directions set in modules above it and to data and information presented from modules below it. Each module makes decisions appropriate for its level in the hierarchy and passes the decision results and necessary supporting information to the functionally connected modules. layer provides direct control, the executive layer provides The device network level consists of sensors, actuators, and sequencing of action, and the planner layer provides deliberative planning. As previously described, autonomous control architecture, based on the CLARAty software environment, was developed to support the MER mission. The CLARAty dual-layer architecture provides an upper (decision) layer for AI software and a lower (functional) layer for controls implementations (see Fig. 1). The development of CLARAty addresses perceived issues with the three-tiered architecture [28]. Those issues are the tendency to-ward a dominant level that depends on the expertise of the developer, the lack of access from the deliberative or planner level to the control or functional level, and the difficulty in representing the internal hierarchy of each level (e.g., nested subsystems, trees of logic, and multiple time lines and planning horizons) using this representation. In one sense, the CLARAty architecture collapses the planner and executive levels, which are characterized by high communications links. The next highest level consists of control, surveillance, and diagnostic modules. The coupling of the control modules with the lower-level nodes is equivalent to an automated control system composed of controllers and field devices. The surveillance and diagnostic modules provide derived data to sup-port condition determination and monitoring for components and process systems. The hybrid control level provides command and signal validation capabilities and supports prognosis of incipient failure or emerging component degradation (i.e., fault identifica-tion). The command level provides algorithms to permit reconfi-guration or adaptation to accommodate detected or predicted plant conditions (i.e., active fault tolerance). For example, if immediate sensor failure is detected by the diagnostic modules and the cor-responding control algorithm gives evidence of deviation based on command validation against pre-established diverse control algo- rithms, then the command module may direct that an alternate R.T. Wood et al. / Nuclear Engineering and Technology 49 (2017) 896e904 901 Fig. 1. Decision and functionality layers in CLARAty architecture. CLARAty, Coupled Layer Architecture for Robotic Autonomy. controller, which is not dependent on the affected measurement variable, be selected as principal controller. The actions taken at these lower levels can be constrained to predetermined configu-ration options implemented as part of the design. In addition, the capability to inhibit or reverse autonomous control actions based on operator commands can be provided. The highest level of the autonomous control architecture provides the link to the opera-tional staff. 4.2. Framework for autonomous control functionality more constrained. Even in the event of transients or faults, the control systemwill try to drive the plant back to a known safe state. This compression of the dual layers into a truncated three-sided pyramid allows for a deeper integration of control, diagnostics, and decision to provide the necessary capability to respond to rapid events and to adapt to changing or degraded conditions. The granularity dimension is retained with more complexity shown at the lower hierarchical levels. Additionally, the informa-tion and command flow reflects granularity as well. At lower granularity, volumes of data are present. As the granularity in- creases moving up the hierarchy, the data are processed into sys- Avariation on the nuclear plant supervisory control architecture tem state and diagnostic/prognostic information that are and the CLARAty architecture for microrovers seems appropriate for consideration as the framework to support autonomy for a SMR plant control system. Fig. 3 illustrates the concept. Essentially, the approach of a hierarchical distribution of supervisory control and diagnostic functionality throughout the control system structure is adopted, while the overlaid decision functionality is maintained. It is possible to blend the decision and functional layers for this application domain because the planning regime for nuclear power system operation is much more restricted than for robotic or spacecraft applications. For example, while there are a multitude of paths that a robot may traverse as it navigates to its next site, the states and state transitions that are allowed for an SMR are much subsequently refined into status and indicator information. On the command side, the transition from the top is demands to com-mands to control signals with the resolution of the plant/system control growing increasingly more detailed. As with the supervisory control architecture, the bottom two levels of the hierarchy are the equivalent of an automated control system. The embedded functionality that enables a reliable, fault-tolerant implementation is indicated as a base intelligence. It is expected that there will be some decision capability associated with the control/surveillance/diagnostics level of that baseline system. The higher levels of the hierarchy assume greater degrees of decision capabilities. 902 R.T. Wood et al. / Nuclear Engineering and Technology 49 (2017) 896e904 Fig. 2. Supervisory control architecture for multimodular SMR plants. SMR, small modular reactor. Fig. 3. Hierarchical framework to support SMR plant control system autonomy. SMR, small modular reactor. In addition to the communications within the hierarchy, the autonomous control system must coordinate with the comparable control systems of other units with whom it is coupled or sharing plant production responsibilities. In addition, it must keep the operational staff informed. To this end, the reactor supervisor/ coordinator node must communicate information about the status of the SMR and the control system and also receive directives and commands. The information provided by the supervisor node can include SMR operational status and capability (e.g., constraints due to degradation), control action histories, diagnostic information, self-validation results, control system configuration, and data logs. Additional communication outside of the hierarchy may be required to coordinate control actions with other prospective ele-ments of the energy conversion/utilization facilities other than the traditional power conversion system. For example, provisions may be incorporated in the plant implementation for dynamic transi-tions between end-use processes, which could include an interface to an industrial user of process heat. R.T. Wood et al. / Nuclear Engineering and Technology 49 (2017) 896e904 903 The functionality that is embodied in the hierarchy can be Intelligence to confirm system performance and detect decomposed into several elements. These include data acquisition, actuator activation, validation, arbitration, control, limitation, checking, monitoring, commanding, prediction, communication, fault management, and configuration management. The validation degraded or failed conditions Optimization to minimize stress on SMR components and effi-ciently react to operational events without compromising sys- tem integrity functionality can address signals, commands, and system perfor- mance. The arbitration functionality can address redundant inputs Robustness to accommodate uncertainties and changing conditions or outputs, commands from redundant or diverse controllers, and status indicators from various monitoring and diagnostic modules. Flexibility and adaptability to accommodate failures through reconfiguration among available control system elements or The control functionality includes direct operational control of the plant as well as supervisory control of the SMR plant control system itself. The limitation functionality involves maintaining adjustment of control system strategies, algorithms, or parameters plant conditions within an acceptable boundary and inhibiting control system actions. The checking functionality can address computational results, input and output consistency, and plant/ system response. The monitoring functionality includes status, response, and condition or health of the control system, compo- nents, and plant, and it provides diagnostic and prognostic infor- The extent to which the key characteristics of autonomy are realized depends on the level of responsibility that is to be entrusted to the autonomous control system. Given anticipated operational imperatives to utilize technology with demonstrated (or at least high probability) readiness, it is not practical to strive for the high-end extreme of autonomy in first-generation SMRs. mation. The commanding functionality is directed toward Instead, modest advancement beyond fully automatic control to

Tài liệu liên quan