A View on Advanced Standby Control in Industry from a Knowledge
Engineering Perspective
Andreas W. Mueller and Joern Peschke
Siemens AG, Process Industries and Drives, Technology and Innovations, Nuremberg, Germany
Keywords: Energy Efficiency, Energy Management, Standby Control, Knowledge Acquisition, Knowledge
Engineering, Energy Engineering.
Abstract: In improving the energy efficiency of industrial plants, advanced standby control depicts a strategic
instrument subject to various influences and with impacts on models and software tools along the entire
plant lifecycle. In order to effectively switch plants and components to different levels of energy
consumption, the required knowledge needs to be properly engineered and applied. This paper presents an
overview of this knowledge and its interrelations in the industrial domain, with special focus on the
requirements for automated generation of energy state models and switching paths. For these tasks,
integrative approaches are proposed.
1 INTRODUCTION
Reducing energy-related expenses without negative
effects on productivity is a prime goal in many
manufacturing industries. Yet, these expenses are
often a result of the way plants are operated:
Machinery is kept ready for production, even during
non-productive phases (Hübner, 2011). In such
cases, situatively switching components to a lower
energy consumption state and back to production
state offers a large savings potential. This is a facet
of “intelligent plant control tailored to situative
needs”, which has been identified as the technical
trend with both the highest potential for making
improvements in the field of energy efficiency and
the highest need for R&D (Bründl et al., 2015).
However, in terms of the energy transition, as
promoted by the German government (BMWi,
2014), the importance of this goal will likely
decrease as decentralized power generation and
storage is increasingly emphasized. Thus, the
currently primary advantage of standby control (SC)
will lose in importance in favor of other motivators
for standby. First, the concept of “carbon footprint”
is a main value in the energy balance of various
industries (IEA, 2007). Further, components are
subject to wear. Standby phases might affect
underlying processes, and timely powering down a
component might therefore help prolong its lifetime.
In coupling SC to indicator values, and in cases
where the course of wear is actively influenceable,
standby is another tool in predictive maintenance
(Mobley, 2002). Another potential use comes from
standby as a redundancy concept in interplay with
energy consumption. This addresses component
availability while simultaneously reducing costs,
which is the object of reliability-centered
maintenance (Moubray, 1997). For this, parameters
like maintenance schedules are needed.
An advanced SC system is vital for the pursuit of
such motivators. This requires a consistent and
integrated model of the domain and plant, together
with representations of standby strategies to be
employed, linked to manufacturing execution
systems and asset management systems. In light of
the demands for integrated plant IT systems (Sauer,
2010), SC systems will eventually become an
integrated functionality. Based on existing
technologies of SC, we propose a basic approach
that permits the generic integration and combination
of standby-related knowledge to dynamically
generate switching sequences and thus flexibly
apply standby strategies.
This paper is organized as follows: Section 2 lists
work related to this topic. Section 3 describes
exemplary SC technologies. Section 4 highlights the
knowledge necessary to operate SC, and Section 5
presents the basic approach. Section 6 concludes the
paper.
364
Mueller, A. and Peschke, J..
A View on Advanced Standby Control in Industry from a Knowledge Engineering Perspective.
In Proceedings of the 7th International Joint Conference on Knowledge Discovery, Knowledge Engineering and Knowledge Management (IC3K 2015) - Volume 2: KEOD, pages 364-369
ISBN: 978-989-758-158-8
Copyright
c
2015 by SCITEPRESS Science and Technology Publications, Lda. All rights reserved
2 RELATED WORK
Schöfberger et al. (2010) point out the variety of
aspects involved in SC in industrial production
plants, such as equipment types, interdependencies
among components as well as between components
and processes, in order to describe and evaluate
energy consumption. Along with transparency for
energy-related data, standardized tools and plant
models are therefore prerequisites.
Wolff et al. (2013) elaborate on the requirements
for the engineering of such SC systems, showing the
demand for digital models and integrated toolchains.
Emphasis is placed on the distinction between static
information models for intrinsic properties of the
plant and models for dynamic behavioral aspects.
Focused on the domain of machine tools, the
research group ECOMATION presents an approach
to the description and exchange of energy-related
knowledge using a special Energy Information
Description Language (EIDL) (Schlechtendahl et al.,
2013). Here, consumers in machine tools and their
interplay are described by interrelated models and
linked to the underlying physical components.
3 STANDBY CONTROL
TECHNOLOGIES IN
INDUSTRY
3.1 Basic Terminology
The term “standby” actually subsumes several
energy-saving modes, distinguished by their levels
of energy consumption. Böde et al. (2000) name
four basic modes. In normal mode, a device fulfills
its main task at a level of 100%, while in general
standby mode it does not fulfill its main task but can
be reactivated at any time. This mode can be
subdivided into the ready(-to-use) mode with almost
no decrease in consumption, the actual standby
mode with reduced consumption, and sleep mode
with the largest decrease in consumption. Another
mode is the pseudo off mode, in which a device still
absorbs energy despite being switched off physically
(an error state). The last basic mode is the actual off,
with no power consumption at all. In this paper,
these modes will also be referred to as energy states
ε. Finally, various triggers exist for activating a
standby mode. These may be device-specific, time-
based presets as well as external schedules, e.g.
pause schedules or reactions to unplanned situations
such as faults.
3.2 Technologies
Initiated by the German automotive industry,
PROFIenergy (PE) aims to reduce power
consumption of production plants through controlled
switching to standby mode during non-productive
phases, based on the PROFINET field bus (PNO,
2010). PE serves the four use cases brief pauses,
longer pauses unscheduled pauses, and measuring
and load visualization. Any device supporting PE
needs to individually implement the responses to the
mandatory commands. The architecture is based on
one PE controller acting as managing instance for
many PE devices representing the components to be
controlled. Here, time – as requested durations for
pauses – is the only criterion: The controller receives
a standby request for a specific duration (e.g. pause
for 15 minutes), and sends Start_Pause and
End_Pause commands to devices in a timely
manner. The devices then autonomously determine
the ε they can switch to, while guaranteeing the
return to production mode when needed. For this, a
state model with mandatory and optional transitions
between states is employed (Figure 1). This covers a
spectrum of discrete ε ordered by their associated
consumption levels and the length of time required
to return to production mode.
Figure 1: PE state model (adapted from (PNO, 2013)).
Sercos Energy (sE) is an application profile for
the sercos field bus system common in the domain
of production machines (Schlechtendahl, 2012). In
addition to the also-covered PE use cases, two
scenarios are served: partial machine operation
(switching off temporarily unneeded components)
and partial load operation (adapting the machine’s
energy consumption according to production
completion dates). The architecture is also 1:n-based
with a controller and subordinate devices. A sercos
controller is called coordinating system, as it is
aware of the overall process state and its influences
on the devices, permitting standalone device control.
Underlying this functionality is a rich state model
A View on Advanced Standby Control in Industry from a Knowledge Engineering Perspective
365
with discrete, semantically divisible ε and with
mandatory and optional transitions (Figure 2)
(sercos, 2011). Unlike PE, the sE state model
permits three mixable modes of control: ε may be
directly addressed, durations for pauses may be
provided (“PE mode”), or a maximum level of
energy consumption may be specified. Switching to
off mode is hence permitted.
Figure 2: sercos Energy state model (from (sercos, 2011)).
The Energy State Controller (ESC) is a result
of a work conducted at a Siemens research facility
for manufacturing automation. Built upon existing
technologies for communication and plant control, it
can perform fine-grained switching of production
plants and subsystems to predefined ε. The ESC
employs an automaton-based approach that requires
a predefined plant model covering aggregation
hierarchy, state models per component, conditions
and costs for switching, and qualification of
transitions by temporal constraints and states of
subordinate components as preconditions. Details
are given in (Mechs, 2013). The architecture
comprises 1 controller and n devices, operating on
individually addressable state machines. Calls to
devices are coordinated by the controller, based on
current subordinate ε and process values. Unlike PE
and sE, the ESC permits arbitrary topologies of
states to be defined. Target ε or durations for pauses
can be stipulated, and the required ε for each
component are inferred from these. For this, the
sequences for reaching any state are inferred from a
static model.
4 KNOWLEDGE ENGINEERING
Today, the desired behavior for such SC needs to be
manually programmed or modeled, with the key
challenge of properly determining, applying, and
monitoring the switching actions in reaction to
triggers. These steps are complex and error-prone.
This section provides an overview of the relevant
knowledge to be engineered in order to enhance
generic integration in plant software systems.
4.1 Switching Paths
Due to interdependencies between components,
switching actions need to be performed in specific
sequences. This ordered set of actions, which is later
executed by a dedicated logic, is subsequently
referred to as the switching path σ, with three basic
phases and corresponding knowledge entities. First,
before switching, the dependencies and valid actions
need to be determined in order to generate a basic σ.
This needs to be checked for both prospective
duration and effects being in line with the specific
demands of the desired action, and whether
returning to the productive state can be guaranteed.
Second, if the σ is executable, it is mandatory to
verify on the path that the next pending switching
action can be performed and, after this, whether the
action has been traceably completed according to the
expected effects. Third, after switching, it must be
checked that all affected components now inhibit the
target ε as expected. Consequently, two main
qualities are essential for any handling of a σ:
Systematic knowledge management is required to
provide the plant knowledge and suitable diagnostics
must be present to interpret actions’ results and to
identify and initiate remedial measures.
4.2 Categories to Be Covered
The knowledge required for handling ε may be
subdivided into four categories. First, there is
structure and functionality, i.e. the static and
dynamic aspects of the plant. Covering components,
their interdependencies, processes and their effects,
this depicts the basic knowledge for determination of
σ. The second category classifies components by
their switchability: Non-switchable or always on
components (e.g. safety controllers) as well as those
non-switchable by programmable logic controllers
(PLC)(e.g. lighting systems) do not contribute to
energy efficiency through SC, whereas those
switchable on-demand by PLC (e.g. conveying
systems) are the most relevant for SC as they are
switchable in dependency on a process or PLC
program. The third category covers the effects of
actions. This subsumes knowledge on components
being affected and on the delays the actions’ effects
usually require to become apparent. Finally, the
fourth category handles classifications of the
components’ states according to the energy
KEOD 2015 - 7th International Conference on Knowledge Engineering and Ontology Development
366
consumption levels. This addresses the ε-specific
energy consumptions as well as the energy demand
for transitions between ε.
4.3 Diagnostics
Effective diagnostics of SC requires a cross-level
view. While determination of σ and coordination of
their execution takes place on a plant or subsystem
level, switching actions are performed on the field
level with typically heterogeneous devices. Requests
for switching actions run from control level to field
level over various communication technologies and
field bus systems. As failures and effects may
generally involve all levels in arbitrary
combinations, full diagnostics must cope with this
issue. The knowledge described above forms the
basis for this. However, this basis is highly dynamic:
First, the plant itself is subject to change during its
lifecycle and second, all related knowledge is
subject to refinement (Figure 3; arrows symbolize
different communication technologies).
Figure 3: Interrelations of knowledge.
A special challenge arises from the infrastructure
connecting the levels, as different technologies with
own diagnostic functionalities are employed, e.g.
(PNO, 2015) or (IO-Link Community, 2013). Thus,
a great deal of knowledge needs to be captured
redundantly, due to incompatibilities. Hence, an
overall approach for diagnostics is strongly advised.
4.4 Towards an Integrated Standby
Management
In terms of the topics presented above, the following
basic methodology for knowledge organization is
proposed for creating an integrated SC system
(Figure 4). A concept for knowledge distribution
must be established initially. Strategies must be
adopted for handling the knowledge needed for
generation of σ and for coordinating its execution,
and to provide all involved components with the
necessary information. On this basis, domain models
that describe the plant in terms of structure and
function are introduced. This references aspects of
knowledge distribution as these mechanisms are part
of the plant. Energy state models are developed,
based on this. These describe the ε of the
components and processes, along with their
interconnections. An idea for this is given in
(Schlechtendahl et al., 2012). Next, switching
strategies are defined. As switching actions are
transmitted over communication channels and affect
components and processes, this information needs to
be available in order to evaluate the plausibility of a
strategy (see (Wolff et al., 2013)). Finally, models
for diagnostics are created. These contain knowledge
about the interpretation of phenomena observed
during switching actions.
Figure 4: Knowledge organization methodology.
5 A UNIFIED SWITCHING
MODEL
Having presented the knowledge relevant for SC, we
will now outline an integrative approach to the
generation and application of energy state models.
5.1 Switching Matrix
Switching to an ε is bound to certain preconditions
and has certain effects. In addition, the criterion for
switching may either be duration, a target ε, an
associated attribute, or a combination of these along
a σ. Based on their characteristics and by
interpreting state models on the graph level, we
propose the organizational means of a switching
matrix M as the central generic knowledge structure
for handling energy state models.
=
0
0
0
0
:
10
101
010
,,
1
,,
1
1
,,
1
0
10
SSSSn
SSSS
SSSS
n
nn
n
n
TTS
TTS
TTS
SSS
M
(1)
Built upon the adjacency matrix of the respective
state graph, M has the following properties:
A View on Advanced Standby Control in Industry from a Knowledge Engineering Perspective
367
A header row and column is added, with distinct
elements for entering and leaving (“
-1
”) states S,
linked by transitions T. The order of the S within
M reflects the levels of energy consumption, with
S
0
for normal mode (operational state) and S
n
for
maximum energy-saving mode (lowest standby).
Each element is defined as an individual
criterion-based (c) evaluable configuration
f(S, c), f(T, c) of preconditions P (must hold for
the S, T to be considered), actions A (performed
upon valid P), and effects E (occur by
performing the A): S, T := P, A, E. An S may
comprise distinct P, A, E for entry/exit. An E
may comprise both triggering of switching
actions and changes of values. The criterion and
thus evaluation result may be duration, state, or
attribute value.
All P, A, E are rooted in the underlying plant
models through their elements referencing plant
entities with evaluable and modifiable attributes.
Missing or prohibited T as well as elements
along the main diagonal evaluate to neutral.
The potential states for each component can be
inferred from the energy-consumption properties and
the components’ interrelations in the plant model.
Ultimately, a σ depicts a trajectory in this search
space, whose determination is controlled using
policies or switching strategies, which effectively
depicts a planning problem. Hence, different modes
of control can be realized by means of M. E.g., since
PE mode requires finding the lowest ε possible for a
given t
Pause
, starting at a current state x, this can be
modeled using (c as duration omitted for brevity):
()
Pause
n
xi
iiiiii
xx
t
TfSfSfTf
SfSf
+++
++
+=
1
1,
1
,1
1
)()()()(
)()(
(2)
Switching matrices may be employed at any
level of a plant, with ideally one M per component.
Viewing an M as a container for inferred knowledge,
they may be centrally engineered on the
manufacturing execution system level and deployed
to both controller and device level, along with the
logic for σ generation. In addition, when initially
supplied for lowest (field) level components, higher-
level M may be inferred bottom-up.
5.2 Automated Generation of
Switching Matrices and Paths
Switching paths depict sequences of actions on a
dynamic knowledge corpus. Also, the domain of SC
is well-defined in terms of relevant relationships. By
taking advantage of these facts, dynamic taskflow
generation (Brecher et al., 2010) can be used for the
automated generation and population of M and for
the generation of σ. There, A are defined by
ontology-based P and E inferred via rule sets. Given
a state-based task on a plant model as parameter, the
proper sequence of A is inferred via a state-based
planning approach. This utilization is outlined in the
following.
M act as guides to each component’s energy
states and thus to the A required for switching ε.
Hence, determining these A is a prerequisite. For
this, the following information is required:
Meta-model-based rules for prioritized handling
of components according to the basic semantic
relations of is-part-of, is-a, is-related-to as well
as depends-on, and according to the roles they
occupy within the relationships: For entries/exits
of S, additional semantics can be defined.
Prioritizations can be specified either by
precedence or by temporal relationship (Allen,
1983), e.g. x, y: depends-on (x, y) after (x,
y). This was impractical for the maintenance and
service domain of the original approach, due to
the great variety of relations. Yet, this is feasible
for the domain of SC as the set of distinct
relations is much smaller.
The characterizing relationships between
components in the plant model: In combination,
these specify candidate classes for P.
The overall E to be achieved on the classes given
by the M, along with the information about the
component values that need to be set to cause the
E and whether activation is permitted.
The initial state of the model.
While iterating over an M, the
P, A, E are
inferred for each element, with initial P of the target
state not being met. The candidate classes are
determined by finding those components for which
any of the given rules would fire. In conjunction
with this, the premises of those rules identified as
valid for candidates constitute the P that are required
to hold for an A to be performed while the E are
defined as the states to be reached. With the actions
at hand, the required σ can be inferred. For this, a
reverse search strategy is applied on the search space
given by the plant model and the associated state
space. Starting with the one E representing the target
state on a considered component, the associated P
are matched with other E that cause the respective P
to hold. This search yields a raw σ, typically with
multiple relevant P and A required for an E. Using
KEOD 2015 - 7th International Conference on Knowledge Engineering and Ontology Development
368
the prioritization rules, this graph is linearized in
analogous reverse search manner, yielding the final
σ. This reasoning process can either be applied a-
priori in order to provide and deploy fully qualified
M or it can be performed on-demand. Hence,
dynamics of the plant model can be met
appropriately. In conjunction with diagnostics, this
also permits the automatic generation of bypass
solutions for σ. In addition, other scenarios based on
sequences of actions can be handled with this
approach as well.
6 CONCLUSION AND OUTLOOK
Although it appears to be a simple task at first
glance, a thorough examination of “standby” in an
industrial context requires broader consideration
beyond the original focus on energy efficiency.
Switching industrial plants to energy states during
non-productive times requires that consideration of
many details that extend into different adjacent
domains. This requires systematic and integrated
knowledge management that combines disparate
knowledge artefacts across the entire plant lifecycle.
This paper has highlighted aspects of knowledge
engineering significant for the enabling of
comprehensive advanced standby control (SC). In
addition, it presented an approach to the flexible,
automated provision of the necessary energy state
models and switching paths. Future work will
address the evolution of the tools and components
involved in plant automation, with the goal of
offering advanced SC as an integrated feature. Also,
in light of the evolving general frameworks, the
implications of advanced SC as a dedicated tool
must be evaluated in the contexts of predictive
maintenance and reliability-centered maintenance so
that its potential can be classified.
REFERENCES
Allen, J.F., 1983. Maintaining Knowledge about Temporal
Intervals. Communications of the ACM, 26(11),
pp.832–843.
BMWi, 2014. The Energy of the Future - First “Energy
Transition” Progress Report -Summary, Berlin: The
Federal Ministry for Economic Affairs and Energy
(BMWi).
Böde, U., Bradke, H., Cremer, C., 2000. Detaillierung des
Stromverbrauchs privater Haushalte in der
Bundesrepublik Deutschland 1997 - 2010, Karlsruhe:
Fraunhofer-Institut für Systemtechnik und
Innovationsforschung -ISI-.
Brecher, C., Feng, J., Özdemir, D., Fayzullin, K., Herfs,
W., Müller, A., Hamadou, M., 2010. Automatische
Generierung von Arbeitsabläufen. Zeitschrift für
wirtschaftlichen Fabrikbetrieb (ZWF), (12/2010),
pp.1084–1090.
Bründl, A., Deutsch, N., Bornholdt, M., Ruhbaum, C.,
2015. Branchenmonitor Energieeffizienz 2015, Berlin:
Deutsche Unternehmensinitiative Energieeffizienz
e.V. (DENEFF).
Hübner, I., 2011. Erkenntnisse der Profienergy-Studie.
openautomation, 3/11.
IEA, 2007. Tracking Industrial Energy Efficiency and
CO2 Emissions, Paris: International Energy Agency.
IO-Link Community, 2013. IO-Link Interface and System
Specification V1.1.2.
Mechs, S.G., 2013. Model-based Engineering for Energy-
efficient Operation of Factory Automation Systems
Within Non-productive Phases, Shaker.
Mobley, R.K., 2002. An Introduction to Predictive
Maintenance, Butterworth Heinemann.
Moubray, J., 1997. Reliability-Centered Maintenance
Second Edition 2 Revised edition., New York, NY:
Industrial Press, Inc.
PNO, 2013. Common Application Profile PROFIenergy -
Technical Specification for PROFINET V1.1 Ed. 2,
PROFIBUS Nutzerorganisation e.V.
PNO, 2015. Diagnosis for PROFINET IO V1.1,
PROFIBUS Nutzerorganisation e.V.
Sauer, O., 2010. Trends in Manufacturing Execution
Systems. In Proceedings of the 6th CIRP-Sponsored
International Conference on Digital Enterprise
Technology. Springer Berlin Heidelberg, pp. 685–693.
Schlechtendahl, J., 2012. Whitepaper sercos ENERGY,
sercos International e.V.
Schlechtendahl, J., Eberspächer, P., Haag, H., Verl, A.,
Westkämper, E., 2013. Framework for Controlling
Energy Consumption of Machine Tools. In P.
Golinska, ed. EcoProduction and Logistics.
EcoProduction. Springer Berlin Heidelberg, pp. 155–
168.
Schlechtendahl, J., Sommer, P., Eberspächer, P., Verl, A.,
2012. Automated Linkage of Consumption Models
and Control Information in Control Systems. In D. A.
Dornfeld & B. S. Linke, eds. Leveraging Technology
for a Sustainable World. Springer Berlin Heidelberg,
pp. 387–391.
Schöfberger, W., Blumauer, G., Bachl, M., Traxler, R.,
Petzl, B., Leutgöb, M., Kolic, P., Bergthaler, F.,
Bäuchler, W., Aschauer, S., Grill, H., Reisinger, B.,
Preinfalk, S., 2010. Abschaltbare Fabrik: Zentrale
Leittechnik und Gesamtkonzept zu Energie-
einsparungen und zur Ressourcenoptimierung im
Standby Betrieb von Industrieanlagen, Linz, Austria.
sercos, 2011. sercos Energy Profile V1.3.1-1.1, sercos
International e.V.
Wolff, D., Hundt, L., Dreher, S., 2013. Requirements on
the engineering of advanced standby strategies in
automobile production. In Proceedings of the 11th
Global Conference on Sustainable Manufacturing.
A View on Advanced Standby Control in Industry from a Knowledge Engineering Perspective
369