Resilient BPMN: Robust Process Modeling in Unreliable
Communication Environments
Frank Nordemann
, Ralf T
and Elke Pulverm
Faculty of Engineering and Computer Science, Osnabr
uck University of Applied Sciences, Osnabr
uck, Germany
Institute of Computer Science, University of Osnabr
uck, Osnabr
uck, Germany
{f.nordemann, r.toenjes},
Business Process Modeling, Language Extension, Meta Modeling, Unreliable Communication Environments,
Dynamic Process Adaption, Process Robustness Verification, BPMN, DMN, QoS, OppNets, DTNs, rBPMN.
Process modeling languages help to define and execute processes and workflows. The Business Process Model
and Notation (BPMN) 2.0 is used for business processes in commercial areas such as banks, shops, production
and supply industry. Due to its flexible notation, BPMN is increasingly being used in non-traditional business
process domains like Internet of Things (IoT) and agriculture. However, BPMN does not fit well to scenarios
taking place in environments featuring limited, delayed, intermittent or broken connectivity. Communication
just exists for BPMN - characteristics of message transfers, their priorities and connectivity parameters are not
part of the model. No backup mechanism for communication issues exists, resulting in error-prone and failing
processes. This paper introduces resilient BPMN (rBPMN), a valid BPMN extension for process modeling
in unreliable communication environments. The meta model addition of opportunistic message flows with
Quality of Service (QoS) parameters and connectivity characteristics allows to verify and enhance process
robustness at design time. Modeling of explicit or implicit, decision-based alternatives ensures optimal pro-
cess operation even when connectivity issues occur. In case of no connectivity, locally moved functionality
guarantees stable process operation. Evaluation using an agricultural slurry application showed significant
robustness enhancements and prevented process failures due to communication issues.
Process Modeling Languages (PMLs) are used for
the definition of business processes and workflows.
Examples include traditional flow charts, UML Ac-
tivity diagrams and event-driven process chains from
ARIS (Dumas et al., 2018). However, the de-facto
standard is BPMN 2.0: due to its flexibility, expres-
siveness and usability for domain experts, its non-
proprietary and extensible design, mature tool sup-
port and ISO-Certification (ISO, 2013), BPMN fits
many different use cases of business processes. Even
in non-traditional business domains such as IoT and
agriculture, BPMN is a promising choice for model-
ing and executing process definitions (cf. Section 2).
In terms of communication, BPMN 2.0 may
not fulfill the requirements of a domain perfectly
though. Communication between actors/system parts
can be realized by message flows in collaboration di-
agrams. Specific views for the cooperation between
actors/system parts can be modeled using choreogra-
phy and conversation diagrams. However, commu-
nication properties such as available bandwidth, la-
tency and failure probability are not in the focus of
BPMN process modeling and execution. This is a
problem for scenarios in which solid and faultless
communication does not exist. Unreliable commu-
nication environments feature limited, delayed, inter-
mittent or broken connectivity in dynamically chang-
ing network topologies. Many scenarios combine
infrastructure-based (e.g. cellular, WiFi in access-
point mode) and infrastructure-free (ad-hoc) commu-
nication technologies, forming hybrid networks based
on node’s connectivity capabilities and mobility char-
acteristics. In particular, this applies to Opportunis-
tic Networks (OppNets) and Delay Tolerant Networks
(DTNs) (Fall, 2003), (Pelusi et al., 2006). Examples
for unreliable communication environments include
use cases in developing and rural areas (e.g. agri-
culture, road construction, wildlife observation), IoT
scenarios with connectivity limited devices, use cases
in untrustworthy infrastructure and disaster scenarios.
Modeling of non-functional communication as-
pects and alternatives for failing connectivity is dif-
ficult and inflexible in BPMN. Mechanisms for on-
demand process adaption based on connectivity pa-
rameters are missing. Robustness issues are identified
not before process execution fails at runtime.
This paper introduces resilient BPMN (rBPMN),
a domain specific modeling language for unreli-
able communication environments. Using this valid
BPMN extension, domain experts may model pro-
cesses and verify their robustness in graphically un-
derstandable, not overloaded collaboration diagrams.
The main research contributions of the paper are
1. a lightweight BPMN meta model extension for
opportunistic message flows,
2. a mechanism for process robustness verification at
design time,
3. methods for explicit and implicit, dynamic oppor-
tunistic alternatives and decision taking for opti-
mal process operation and
4. a method for moveable functionality in case of no
An example for an unreliable communication en-
vironment is provided by the agricultural domain.
Agriculture takes place in rural and sparsely popu-
lated areas using a small amount of mobile machines.
Communication between machines, farms, personnel
and Internet services is often limited or broken.
The paper starts with related work (Section 2),
followed by the presentation of rBPMN (Section 3)
with its domain requirements, equivalence check, the
domain extension model and robustness verification.
Next is a use case featuring an agricultural slurry ap-
plication in Section 4 and a comparison to robust pro-
cess modeling in plain BPMN 2.0 in Section 5. Con-
clusion and future work is presented in Section 6.
BPMN 2.0 has been extended in various ways since
its definition by the Object Management Group
(OMG) in 2011 (OMG, 2011). BPMN was designed
following a model driven approach. Its metamodeling
architecture defines the structure, meaning and behav-
ior of objects and is based on the Meta Object Facility
(MOF) of OMG (OMG, 2014). The integrated exten-
sion mechanism allows to extend the modeling lan-
guage for special requirements of new use cases and
domains. However, according to (Braun and Esswein,
2014a) less than 20 percent of available extensions
implement the provided mechanism. This is probably
caused by syntactical and methodical misunderstand-
ings as argued by (Braun, 2015). Valid extensions en-
able model interchangeability and remain compatible
with the core of BPMN, an important aspect when us-
ing existing tools and runtime engines. A selection of
extensions is presented subsequently.
Numerous extensions provide solutions to inte-
grate resources (Stroppi et al., 2011a), (Braun and Es-
swein, 2014b), (Bocciarelli et al., 2016), (Betke and
Seifert, 2017) and performance criteria (Bocciarelli
and D’Ambrogio, 2011), (Bocciarelli et al., 2014)
into BPMN.
Many publications present solutions for an inte-
gration of IoT and Cyber Physical Systems (CPS) into
the BPMN meta model. Work often concentrates on
concepts for physical entities, sensors and actuators,
adding new task types, events and attributes to the
meta model (Meyer et al., 2013), (Meyer et al., 2015),
(Graja et al., 2016), (Bocciarelli et al., 2017). Quality
of Information (QoI) of physical resources is focused
in (Martinho and Domingos, 2014).
QoS in web service process models and SOA-
based service discovery is part of (Bocciarelli and
D’Ambrogio, 2014). (Mazzola et al., 2017) provide
a cloud-based environment with compensations for
faulty tasks and QoS-based process optimization. In
both publications, QoS is referring to non-functional
aspects such as performance, workload and reliabil-
ity of individual process tasks and resources. QoS-
aspects like bandwidth, latency and failure probabil-
ity of the actual transport network are not considered.
Furthermore, the solutions do not address unreliable
communication scenarios and robustness verification.
A couple of publications address process, task
and resource reliability (Bocciarelli et al., 2014),
ıcio and Domingos, 2015), (Domingos et al.,
2016). Other extensions include mobility (Kozel,
2010), clinical pathways (Braun et al., 2014), com-
posite applications (Kopp et al., 2012), mobile context
orndorfer and Seel, 2017) and ubiquitous comput-
ing (Yousfi et al., 2016).
To the best of the authors’ knowledge, none of
the available extensions focuses on robust process
modeling for unreliable environments, none features
a robustness verification mechanism. The solutions
do not address integration of communication require-
ments and scenario-based connectivity descriptions
into BPMN. Straight-forward mechanisms for mod-
eling of communication-issue-related alternatives by
domain experts are missing.
A good practice to design a BPMN extension was in-
troduced by (Stroppi et al., 2011b) and (Braun et al.,
2014). It starts with a requirements analysis of the
Created by model
rules from CDME
1. Analyse
2. Perform
4. Create
5. Create
XML syntax
Created by model
rules from
3. Model
Focus of this paper
Figure 1: Good practice of BPMN extension development,
cf. (Stroppi et al., 2011b), (Braun et al., 2014).
target domain and an equivalence check to compare
the requirements with existing BPMN concepts. Both
steps help to identify missing parts of BPMN for un-
reliable environments. Following steps include mod-
eling of the Conceptual Domain Model of the Exten-
sion (CDME) to address the missing parts, modeling
the extended BPMN meta model (BPMN+X) to re-
main valid with BPMN and creation of the concrete
syntax of XML schemas and files (5 steps shown in
Figure 1).
This approach is also used for the development
of rBPMN in this paper. Focus is on domain re-
quirements, equivalence check and CDME, the cre-
ation of BPMN+X and XML syntax using trans-
formation rules is described extensively in (Stroppi
et al., 2011b). The following subsections will present
rBPMN and its robustness verification mechanism.
3.1 Step 1: Domain Requirements
Subsequent paragraphs will identify the requirements
for robust process modeling. Agriculture will serve
as an example for unreliable communication environ-
In agriculture, processes to till and harvest fields
need to be defined, organized, controlled, monitored,
adapted and documented. Many processes require
collaboration of different actors (e.g. farmers, agri-
cultural contractors, supporting machines, digital ser-
vice providers, data hubs and authorities). While
collaboration is required, modeling shall be time-
efficient and respect trade secrets. Thus, the derived
requirements are:
Req. 1: Ability to model collaborative agricul-
tural processes including different actors
(e.g. humans, machinery).
Req. 2: Ability to split collaborative agricultural
work into reusable subsegments and to
model them as black boxes.
The work needs to be controlled and monitored.
Automatic and manual adaptions may be required,
e.g. for automatically adapting machine parame-
ters or manually deciding on machine failure replace-
ments. Derived requirements:
Req. 3: Ability to adapt processes automatically
based on predefined variables / events.
Req. 4: Ability to manually adapt running pro-
cesses based on process state / proposed
Agriculture happens in rural, sparsely populated
environments. The areas often lack cellular coverage,
communication is delayed, intermittent or broken.
Quality of actors’ connectivity may change rapidly.
Derived requirements:
Req. 5: Ability to define QoS requirements for
communication and actors’ connectivity
Req. 6: Ability to ensure robustness even dur-
ing intermittent or broken communica-
tion between actors.
Regardless of the challenging communication en-
vironment, processes shall operate optimally within
given limits. Process robustness shall be verifiable be-
fore runtime. Derived requirements:
Req. 7: Ability to explicitly model optimal pro-
cess operation at design time.
Req. 8: Ability to identify optimal process oper-
ation in a dynamically changing scenario
at runtime.
Req. 9: Ability to verify process robustness for a
given scenario at design time.
Finally, the extension shall follow the BPMN ex-
tension mechanism to be BPMN compliant. Derived
Req. 10: The extension shall comply with the
BPMN meta model and semantics.
3.2 Step 2: Equivalence Check
An equivalence check was performed to compare
existing BPMN concepts with the requirements to
model executable process diagrams for unreliable
communication environments. The support level (SL)
to fulfill the requirements is indicated by + full
support, o limited support and - no support in
Table 1.
The equivalence check identifies BPMN concepts
for the requirements 1, 2, 3 and 4. However, it also
demonstrates the need for extension concepts to fulfill
the requirements 5, 6, 7, 8 and 9 for communication,
dynamics and decision modeling.
Application of the Decision Model and Notation
(DMN) of OMG was evaluated and discarded. In
DMN, a decision is taken using inputs and a set of
rules structured by a decision table, resulting in an
output (OMG, 2019). However, deciding on alterna-
tives for broken communication requires to compare
Table 1: Equivalence check with BPMN concepts and their support-level (SL) for the identified domain requirements.
Req. Concept Semantics (and support declaration, where applicable) SL
General process modeling
1, 2 Activity / Task Part / step of a process. +
1, 2 Process Reusable container for a set / flow of chosen activities. +
1, 2 Sub-process
Call activity
Encapsulates / hides activities in processes, allows hierarchy levels, allows
1 Sequence flow Coordinates the process flow. +
3 Gateway
Business rule task
Allows autonomous process flow decisions based on defined variables. Limited
support: not designed for handling dynamics based on unreliable environments.
4 User task
Manual task
Ad-hoc sub-proc.
Allows user-based decisions for non-automated situations. +
3, 4 Event
sequence flow
Allows to react (dynamically) on messages, events, conditions, timings. Allows
to dynamically create process instances / configurations.
1, 2 Text annotation Specifies descriptive information, no influence on sequence flow. +
Collaboration modeling
1 Participant
Defines / structures different actors. +
1, 2 Collapsed pool Defines different actors, hides internals in black box. +
1 Pool lanes Separation of concerns / organization of activities within an actor. +
Communication modeling
1 Message flow Communication with other actors. Limited support: no possibility to specify
alternative message flows / to specify flows as optional based on connectivity.
1 Message
Graphical element and name for a message. Limited support: frequency, size,
relevance not included / not widely used in runable BPMN environments.
5, 7, 8, 9 Not available No support: no BPMN concepts fulfill requirements 5, 7, 8, 9. -
Modeling of dynamics
6 Not available No support: no BPMN concept fulfills requirement 6. -
Modeling of decisions
6, 7, 8 Decision table Decision taking based on inputs, static rule sets in form of a decision table
and outputs. Concept of Domain Model and Notation (DMN). No support:
alternatives need to be compared to each other based on different criteria.
Support-Level-Declaration: + full support, o limited support, - no support
different options on the basis of defined criteria with
each other.
So far, BPMN misses concepts to model QoS re-
quirements for message flows, to flexible describe al-
ternatives for failing message flows and to decide op-
timally on these alternatives during process runtime.
In addition, it is not possible to verify process robust-
ness for a given scenario at design time.
3.3 Step 3: Conceptual Domain Model
of the Extension
The third step is represented by modeling the CDME.
The core benefit of CDME development is to focus on
the design of the modeling language for the applied
domain without handling any restrictions set by the
BPMN extension mechanism. The extension mech-
anism follows the principle of extension by addition
rather than extension by generalization (OMG, 2011).
However, generalization was used in the CDME to
<<Domain Model>>
<<BPMN Con.>>
+size : double
+interval : double
<<Ext. Con.>>
+deliveryProbability : double
+maxDeliveryDelay : double
<<Ext. Con.>>
<<Ext. Con.>>
+failureProbability : double
+failurePenaltyTime : double
+meanBandwidth : double
+minBandwidth : double
+probOfMinBandwidth : double
<<Ext. Con.>>
+scenarioName : String
+characteristics : String
<<Ext. Con.>>
+priorityClass : String
<<Ext. Con.>>
<<Ext. Con.>>
+weight : double
<<Ext. Con.>>
+priority : Integer
<<Ext. Con.>>
+weight : double
<<Ext. Con.>>
+criteriaName : String
<<Ext. Con.>>
+groupName : String
+msgRequired : Boolean
<<Ext. Con.>>
<<BPMN Con.>>
+connType : ConnType
<<Ext. Con.>>
<<Ext. Con.>>
<<BPMN Con.>>
<<BPMN Ele.>>
<<BPMN Con.>>
+dynFuncList : List
<<Ext. Con.>>
+movTaskId : String
<<Ext. Con.>>
+localFuncList : List
<<Ext. Con.>>
+funcName : String
+funcDescr : String
<<Ext. Con.>>
<<BPMN Con.>>
<<BPMN Con.>>
+autoLevel : double
<<Ext. Con.>>
+movSPId : String
<<Ext. Con.>>
+movParId : String
<<Ext. Con.>>
Figure 2: Context Domain Model of the Extension (CDME) for rBPMN.
add extension concepts which will be explained sub-
sequently. This will not result in any problems since
transformation of CDME to BPMN+X in step 4 en-
sures BPMN validity.
After the missing parts for unreliable environ-
ments modeling have been identified by the domain
requirements analysis and the equivalence check, the
CDME shown in Figure 2 was designed. Figure 3
illustrates new graphical elements for the introduced
opportunistic message flows, tasks and attributes.
The CDME adds new message flow types to the
meta model to respect opportunistic communication
characteristics. Based on existing BPMN message
flows, OppMessageFlows enable the definition of
QoS requirements, message flow properties and con-
nectivity properties for specific scenarios. Using this
information, a verification mechanism can compare
required and provided connectivity to evaluate robust-
ness (cf. Section 3.4). Furthermore, priorities for de-
ciding on alternative message flows in case of con-
nectivity failures may be modeled using OppPrior-
ityFlows. Priorities are recognizable by a number
within the circle of a message flow.
More sophisticated decision taking on alternatives
is possible using OppDecisionFlows. OppDecision-
Flows annotate alternatives with alphabetic charac-
ters, displayed within the circle of the message flows.
In combination with additional extension concepts for
decision taking, different criteria can be weighted and
combined in a matrix, allowing a decision engine to
choose the optimal alternative in a flexible and dy-
namic manner. If necessary, other approaches for de-
cision taking may be integrated into rBPMN.
Opportunistic messaging and traditional BPMN
message flows may be used in a single diagram, es-
pecially helpful for scenarios with reliable communi-
cation segments.
A small number of new task types has been in-
MessageFlow (BPMN)
Figure 3: rBPMN message flows, extension tasks and at-
troduced. Moveable and opportunistic tasks represent
the basis to locally move functionality between ac-
tors/system parts: MovTasks and MoveSubProcesses
offer functionality that may be moved to other ac-
tors/system parts. OppTasks allow to locally store,
execute and to integrate moved functionality into pro-
cesses. OppDynTasks add more flexibility by the
dynamic identification of alternatives that have not
been explicitly modeled. Function descriptions help
to evaluate appearing actors/system parts as possible
Visual recognition of locally moved functionality
is supported by a task autonomy attribute shown by
Figure 3. The connectivity type of actors and system
parts is identifiable by pool attributes for seamless and
opportunistic communication to the cloud/Internet.
Design of the CDME focuses on a lightweight, but
powerful set of extension concepts. rBPMN may be
combined with other extensions. Usage of BPMN
elements with low practical usage is avoided [e.g.
message / item def. in BPMN runtime engines such
as (Camunda, 2019), more details in (Geiger et al.,
2018)]. Additional remarks about the extension con-
cepts and their semantics are summarized in Table 2.
Using the CDME and a set of transformation
rules, BPMN+X (step 4) as well as the XML schemas
and files (step 5) may be created. Due to space lim-
itations, these steps are omitted in this paper and the
reader is referred to transform illustration in (Stroppi
et al., 2011b).
Table 2: Extension concepts modeled to fulfill the requirements of unreliable communication environments.
Req. Concept Semantics
Collaboration modeling
6 MovPraticipant Participant offers moveable functionality to other actors / system parts.
Communication modeling
5 OppMessageFlow Possibly intermittent or broken communication with other actors / system parts.
Allowed to be used with existing BPMN activities / tasks / participants.
7 OppPriorityFlow Opportunistic message flow with explicitly defined priorities for alternatives
modeling. A number within the message flow circle states the priority.
7, 8 OppDecisionFlow Opportunistic message flow with implicit, criteria-based decision taking for al-
ternatives. An alphabetic character within the message flow circle states the
decision group.
9 MessageFlow-
Describes message properties (e.g. frequency, size, relevance).
9 QoSRequirements Defines QoS requirements for a message flow.
9 QoSPriorityClass Defines a QoS hierarchy, to be used by QoSRequirements.
9 Connectivity Defines a type of connectivity (seamless, opportunistic) for a participant.
9 Connectivity-
Describes connectivity at the time of a message flow.
9 Connectivity-
Allows to group ConnectivityProperties to different scenarios.
Modeling of dynamics
6 Autonomy Allows to define a level of autonomy in case of broken connectivity for BPMN
tasks (e.g. 4 OppMessageFlows, 3 with local functionality autonomy level
of 75 %).
6 Functionality Labels functionality, used to describe moveable process parts and to identify
dynamic alternatives at runtime.
6 MovTask Defines tasks that allow to move functionality to other actors / system parts.
6 MovSubProcess Defines sub-processes that allow to move functionality to other actors / system
6 OppTask Task that is capable of accepting / operating local functionality.
6 OppDynTask An OppTask that dynamically identifies alternatives (using functionality de-
scriptions) which have not been modeled explicitly at design time.
Modeling of decisions
7, 8 OppMessageGroup Group of OppMessageFlows, where a decision on calling actors / system parts
is needed.
6, 7, 8 OppDecisionEngine Engine to choose OppMessageFlows based on decision criteria and engine pa-
6, 7, 8 DecisionCriteria Decision criteria used by decision engine. Three criteria have been defined:
ConnectivityDecision, PiorityDecision, FeatureDecision.
3.4 Robustness Verification
Stable and failure-free execution of business pro-
cesses is of high importance for most business do-
mains. rBPMN prevents failing process executions by
verifying the process robustness at design time. This
section introduces rBPMN’s verification mechanisms
for a worst and average case robustness calculation in
unreliable communication environments.
Equ. 1 determines the number of data packet
frames N
required to send a message ( packet frag-
mentation). The message size M
is divided by the
payload size of a frame F
. The result is rounded up
to the nearest whole number.
Transferring messages requires protocols to orga-
nize and control communication. Protocol overhead
of a data packet frame is represented by F
and in-
cludes all protocol headers used within the frame.
Since overhead created by the protocol stack can be
several times higher than small amounts of applica-
tion payload data, it is important to include F
in the
following calculations. Summarizing F
and F
sults in the total size of a data packet frame F.
In addition to M
, N
and F
, a minimum band-
width BW
> 0 and a maximum delivery delay T
should be provided to calculate a worst case robust-
ness. Equ. 2 determines the worst case message
transfer time T
. The message transfer is robust for
, potentially not robust for T
> T
and cer-
tainly not robust for BW
= 0.
+ N
Equ. 3 and 4 provide a robustness calculation
more likely for many scenarios. The bandwidth BW is
identified by BW
, an average bandwidth BW
the probability for BW
BW min
BW min
+ BW
(1 P
BW min
) (3)
Subsequently, a failure probability P
[0, 1] and
a failure penalty time T
are included in Equ. 4.
The resulting value T
represents the message trans-
fer time for an average case.
+ N
+ P
For recurring message transfers, Equ. 5 deter-
mines the number of messages N
that fit into the re-
quested message interval T
. Afterwards, N
is com-
pared with the requested delivery probability P
[0,1]. The recurring transfers are robust for N
and potentially not robust for N
< P
Domain experts may have difficulties estimating
the required parameters for robustness calculations.
Provisioning of representative bandwidth values for
poor and average connectivity in the business domain
helps to improve calculations. Guidelines for typical
protocol stacks with their frame header and payload
size can be provided (e.g. for TCP, IP, WiFi 802.11).
Tools for monitoring and evaluating communication
characteristics at process runtime may identify more
sophisticated parameter values for future robustness
Table 3: QoS and connectivity settings for messages.
QoS req.
GPS corr.
0.5 1 1 0.9 1 0.9
10m 30s 5s 2s 1d 2s
Message properties
(KB) 500 1000 250 10 5000 10
(s) 60 - - 5 - 5
Connectivity properties
0.3 0.3 0.3 0.3 0.3 0.1
(s) 10 10 10 10 10 5
(kbps) 10 10 10 10 10 100
(kbps) 500 500 500 500 2500 1000
BW min
0.5 0.5 0.5 0.5 0.3 0.2
Declaration: - not applicable
This section illustrates using rBPMN in an agricul-
tural slurry scenario featuring different actors. Slurry
is applied onto predefined subsections on a field based
on its ingredients in the sense of precision farming.
This approach ensures to comply with official regula-
tions (e.g. distance to streams, protected areas).
A central process management (MGMT) located
in the cloud controls the slurry application and assigns
a task to a slurry spreader (SP). SP drives to the field
and starts the application of slurry with the help of an
online slurry analysis service (OSAS) and an online
GPS correction service (OGCS). During slurry appli-
cation, SP continuously reports its status to MGMT
and may receive instructions back. This is certainly
true if SP is facing operational issues and sends a fail-
ure message to MGMT. Lastly, SP transfers a task
log back to MGMT after finishing the slurry appli-
cation. Figure 4 (a) displays the process in plain
BPMN. The process would fail badly in many real-
world-scenarios due to connectivity issues. However,
a user would recognize the failing robustness not be-
fore the process is being executed.
Figure 4 (b) shows the slurry process using
rBPMN after robustness verification. Red/gray parts
indicate robustness issues, pointing out a non-stable
slurry application process. Unreliable communication
is recognizable by tightly dashed lines compared to
non-problematic, standard BPMN message flows (cf.
Figure 3). Pool attributes show the connectivity type
Slurry spreader (SP)
Setup machine,
drive to field
Analyse slurry
GPS, apply
slurry to field
Create slurry
task log
Transfer status,
adapt operation
Online slurry analysis service (OSAS) Online GPS correction service (OGCS)
Deploy task
to machine
Control slurry
slurry task
Transfer task
Status Instructions Transfer logFailure
Slurry spreader (SP)
Setup machine,
drive to field
Analyse slurry
GPS, apply
slurry to field
Create slurry
job log
Transfer status,
adapt operation
Online slurry analysis service (OSAS) Online GPS correction service (OGCS)
Deploy task
to machine
Control slurry
slurry task
Transfer task
Status Instructions Transfer logFailure
Figure 4: Slurry scenario (a) in plain BPMN and (b) with OppMessageFlows and robustness verification.
Add QoS
Process robust?
Figure 5: Adding robustness to processes with rBPMN.
of actors. Some actors have a seamless connection
to the Internet (cloud sign), others may face interrup-
tions or no Internet connectivity (warning sign).
This result was achieved by adding QoS and ex-
pected connectivity parameters for opportunistic mes-
sage flows as described by Figure 5. QoS and con-
nectivity parameters shown in Table 3 have been de-
fined as an example for this paper. It is possible to
model varying protocol overhead for different mes-
sage transfers in rBPMN. Typical values for a com-
bination of TCP, IP and WiFi 802.11 have been set
in this example, resulting in a frame header size F
of 82 byte and a frame payload size F
of 2230 byte
(Kliazovich and Granelli, 2008).
The agricultural domain expert excluded the ini-
tial task transfer from MGMT to SP, because SP is
located at a farm with solid connectivity at the time
of transfer. Thus, the model in Figure 4 (b) features
reliable and unreliable communication segments.
A worst case robustness verification identified
several connectivity-related issues in Figure 4 (b), in-
dicated by the red/gray color of tasks and message
flows. While missing up to 50 percent of status re-
ports from SP to MGMT is acceptable (P
in Table
3), transfer of crucial failure information and instruc-
tion messages would fail. In addition, the actual slurry
application could not operate: functionality calls to
OSAS and OGCS would fail. At this point, rBPMN
is able to enhance process robustness by
: adding alternatives for message flows explic-
itly at design time, by
: finding appropriate alternatives dynamically at
runtime and by
: moving (limited) functionality to SP.
All three robustness enhancement options have been
used in Figure 6 (a):
Concerning E
: A local GPS station at the field
border (LGCS) was added to emit correction signals.
Connectivity of LGCS to SP is solid during slurry ap-
plication (cf. Table 3) due to its close proximity to
Concerning E
: Dynamic alternatives for the
slurry ingredients analysis of SP have been included
implicitly by choosing an OppDynTask. Since dy-
namic alternatives appear at runtime, they are not
found in the graphical process model at design time.
However, they appear in the decision matrix for the
example slurry application illustrated in Table 4. Data
provided by an ingredients laboratory (LAB) or a
near-infrared spectroscopy sensor (NIRS) may be
used, if they are present. Decision matrix explana-
tions follow subsequently.
Concerning E
: An autonomous control unit of
MGMT and a limited calculation module of OSAS
(OSAS-LF) have been designed as moveable func-
tionality, illustrated by a moveable sub-process and
a moveable black-box pool encapsulating functional-
ity. The autonomy attributes on appropirate SP tasks
indicate that functionality was moved locally.
Alternatives come with the challenge of decision
taking. Priorities may be modeled explicitly with
OppPriorityFlows, implicitly and more dynamically
with OppDecisionFlows. In this example, the domain
expert decided to use an OppTask with explicit Opp-
PriorityFlows to get the GPS signal correction: first
choice is LGCS, second is OGCS.
In contrast, the expert used an OppDynTask with
OppDecisionFlows for the slurry ingredients analy-
sis. Here, not all alternatives are graphically visible
in the model. A decision matrix with different criteria
(cf. Table 4) is used to define and dynamically decide
for the optimal alternative at runtime. Alternatives in-
clude LAB, NIRS, OSAS and OSAS-LF. For this ex-
ample, the decision engine is configured to choose the
alternative with highest accuracy while having con-
Slurry spreader (SP)
Setup machine,
drive to field
Create slurry
job log
Transfer status,
adapt operation
Analyse slurry
GPS, apply
slurry to field
Online slurry analysis service (OSAS) Online GPS correction service (OGCS)
Deploy task
to machine
Control slurry
slurry task
Local GPS correction service (LGCS)
Transfer task
Status Instructions Transfer logFailure
Online slurry analysis service (OSAS) Online GPS correction service (OGCS)
Deploy task
to machine
Control slurry
slurry task
Local GPS correction service (LGCS)Near-infrared spectroscopy sensor (NIRS)
Slurry spreader (SP)
Setup machine,
drive to field
Create slurry
job log
Analyse slurry
GPS, apply
slurry to field
Transfer status,
adapt operation
Transfer task
Transfer log
InstructionsStatus Failure
Figure 6: Slurry scenario (a) with robustness optimizations and (b) after final robustness verification.
Table 4: Decision matrix using features and connectivity.
Accuracy (%) 90 75 65 55
Error ratio (%) 1 10 15 15
Calculation time delay (ms) - 10 2000 -
Local performance cost (class) - - - 1
Cost of usage (EUR) 25 10 2 5
Connectivity aspect
Connectivity check x X (X) X
Connectivity prognosis x X x X
Cancellation timeout (s) 0 1 5 0
Locally moved functionality - - - X
Declaration: - not applicable, x failed,
(X) limited success, X successful
nectivity. More complex decision matrices and rules
may be created, if required. For instance, the decision
could also include feature properties such as an error
ratio or a calculation time, connectivity aspects could
include a connectivity prognosis as displayed by Ta-
ble 4.
Figure 6 (b) presents the final slurry robustness
verification. Communication with MGMT, OSAS
and OGCS is still unreliable. However, the slurry pro-
cess is supposed to run stable and under optimal con-
figuration: without having laboratory data on-hand,
the ingredients alternatives decision is made for the
dynamically appearing NIRS. It is rated as best avail-
able alternative by the decision engine (cf. Table 4).
The backup alternative is the local functionality pro-
vided by OSAS. MGMT might receive some status
reports initiated by SP. Since the required QoS is
not guaranteed for status, failure and instruction mes-
sages, local functionality of MGMT is picking up on
SP in case of connectivity failures. This illustrates an
important mechanism: communication attempts with
other actors are possible even in rapidly changing net-
work environments - after a defined timeout, local
components take over operation. An optimal process
operation is guaranteed.
rBPMN represents a domain specific modeling lan-
guage for unreliable communication environments.
Mechanisms such as modeling alternatives for un-
available message flows are not exclusive to the ex-
tension and can be modeled within plain BPMN 2.0.
This section compares robust process modeling using
rBPMN with three approaches in plain BPMN that
have been modeled for this paper.
The comparison is based on modeling the slurry
ingredients analysis with multiple alternatives. Deci-
sion taking in the rBPMN implementation is shown
in Table 4 and includes the alternatives LAB, NIRS,
OSAS and OSAS-LF. Configuration can be changed
by decision engine rules, the decision matrix stays un-
touched. Again, the engine is configured to decide on
best accuracy while having connectivity.
BPMN-EV presented in Figure 7 is the first plain
BPMN solution and uses error events to model al-
ternatives. Alternatives are handled in fixed prior-
ity, one after another, and got encapsulated within an
additional sub-process. If an alternative is unavail-
able due to connectivity issues or no data, an error
event is thrown and the next option is addressed. Al-
though timeouts are not modeled, the solution already
requires high modeling effort and complicates the ex-
pressiveness of the diagram and its functionality. Pri-
Slurry spreader (SP)
No timeouts
Correct GPS
position, apply
slurry to field
Analyse slurry ingridients
from laboratory
OSAS local
Online slurry analysis service (OSAS)Near-infrared spectroscopy sensor (NIRS)
No laboratory
Figure 7: Robustness modeling with BPMN error events
Slurry spreader (SP)
No timeouts
Correct GPS
position, apply
slurry to field
Analyse slurry ingridients
OSAS local
Online slurry analysis service (OSAS)Near-infrared spectroscopy sensor (NIRS)
Figure 8: Robustness modeling with BPMN inclusive gate-
ways (BPMN-GW).
Slurry spreader (SP)
on ingridients
Correct GPS
position, apply
slurry to field
Call choosen
Call activity
details not incl.
Online slurry analysis service (OSAS)Near-infrared spectroscopy sensor (NIRS)
Figure 9: Robustness modeling with BPMN business rule
tasks (BPMN-DMN).
ority changes require additional modeling effort. Dy-
namics do not exist, calling multiple actors in parallel
due to bad connectivity is not possible.
Parallel communication is enabled by the sec-
ond solution BPMN-GW displayed in Figure 8, again
without timeouts for communication calls. Inclusive
gateways add more flexibility compared to BPMN-
EV. The expense is hidden within the inclusive gate-
way in which all rules need to be set up carefully to
operate the process optimally. It is hardly possible to
include all parameters for optimal process operation
even with broadly defined gateway rules. Rule adap-
tion is far from easy for agricultural domain experts.
A sub-process is required to hide the modeling effort.
Dynamically appearing alternatives are not captured.
The third solution BPMN-DMN in Figure 9 uti-
lizes a business rule task and adds less graphical ele-
ments to the diagram. The business rule task is rep-
resented by a decision table of DMN, evaluating in-
puts using a set of rules to outputs. However, here we
Table 5: Comparison of robust modeling approaches.
Use of existing BPMN / DMN el-
+ + + - -
Graphical expressiveness of un-
reliable connectivity, alternatives
and priorities.
o o - + +
Avoidance of additional graphical
hierarchies for modeling of oppor-
tunistic characteristics.
o o o + +
Ability to integrate QoS require-
ments for opportunistic message
flows between actors.
- - - + +
Ability to adapt process execution
dynamically based on connectivity
(run tasks exclusively / in parallel
/ locally moved functionality).
- - o o +
Ability to model & adapt to move-
able / local / dynamically appear-
ing functionality.
- - - - +
Domain experts: simplicity of ro-
bust modeling while focusing on
problem domain.
- - - + +
Ability to verify process robust-
ness for scenarios prior to runtime.
- - - + +
Summarized points 4 5 3 11 14
Declaration: + full support / 2 points,
o limited support / 1 point, - no support / 0 points
do not have a set of inputs that is evaluated against
a static rule set. Instead, we want to compare dy-
namic alternatives with each other, choosing the most
appropriate alternative to be invoked by a call activ-
ity. Hence, the solution is unable to provide the re-
quired flexibility. Furthermore, details of alternatives
and their decision taking are hidden in the decision
Table 5 compares and evaluates the plain BPMN
solutions with the introduced extension for unreliable
communication environments. rBPMN is represented
by a minimum set (rBPMN-min) implementing op-
portunistic message flows and connectivity parame-
ters as well as a full implementation including priori-
ties, decision taking and opportunistic tasks (rBPMN-
max). All three plain BPMN solutions presented end
up with a low number of points in the comparison.
They are missing the required flexibility and do not
verify robustness prior to process runtime. Error-
prone and failing processes are the consequence.
This paper introduces rBPMN, a valid BPMN exten-
sion for unreliable communication environments with
limited, delayed, intermittent or broken connectivity.
Using rBPMN, domain experts may model robust sce-
narios with dynamic alternatives for failing message
transfers. Users may verify and enhance scenario ro-
bustness prior to runtime.
The BPMN meta model has been extended with
opportunistic message flows, enabling the definition
of QoS parameters and connectivity characteristics
for robustness verification. Alternatives may be de-
scribed explicitly with priorities or implicitly using
decision criteria to be evaluated dynamically by de-
cision engines. In case of no connectivity, locally
moved functionality of other actors/system parts guar-
antees stable process operation. rBPMN has been de-
signed as a lightweight, but powerful set of extension
concepts. It is extendable by more complex param-
eters and criteria for decision taking and robustness
verification, if required.
Further research objectives include practical eval-
uations of the proposed concepts for opportunistic
messaging, moveable functionality and dynamic de-
cision taking. In addition, research on a tool for as-
sisted troubleshooting of robustness issues seems to
be a promising help for domain experts.
The presented work is part of the OPeRAte project
(OPeRAte, 2019). OPeRAte is supported by funds
of the Federal Ministry of Food and Agriculture
(BMEL) based on a decision of the Parliament of the
Federal Republic of Germany via the Federal Office
for Agriculture and Food (BLE) under the innovation
support program.
Betke, H. and Seifert, M. (2017). BPMN for disaster
response processes. In INFORMATIK 2017, pages
1311–1324. Gesellschaft f
ur Informatik, Bonn.
Bocciarelli, P. and D’Ambrogio, A. (2011). A BPMN exten-
sion for modeling non functional properties of busi-
ness processes. In Proceedings of the 2011 Sympo-
sium on Theory of Modeling & Simulation: DEVS In-
tegrative M&S Symposium, pages 160–168. Society
for Computer Simulation International.
Bocciarelli, P., D’Ambrogio, A., Giglio, A., and Paglia, E.
(2014). Simulation-based performance and reliabil-
ity analysis of business processes. In Proceedings of
the 2014 Winter Simulation Conference, pages 3012–
3023. IEEE Press.
Bocciarelli, P., D’Ambrogio, A., Giglio, A., and Paglia,
E. (2016). A BPMN extension to enable the explicit
modeling of task resources. In CIISE, pages 40–47.
Bocciarelli, P., D’Ambrogio, A., Giglio, A., and Paglia,
E. (2017). A BPMN extension for modeling cyber-
physical-production-systems in the context of Indus-
try 4.0. In 14th International Conference on Network-
ing, Sensing and Control (ICNSC), pages 599–604.
Bocciarelli, P. and D’Ambrogio, A. (2014). A model-driven
method for enacting the design-time qos analysis of
business processes. Software & Systems Modeling,
Braun, R. (2015). Behind the scenes of the BPMN ex-
tension mechanism principles, problems and options
for improvement. In 3rd International Conference
on Model-Driven Engineering and Software Develop-
ment (MODELSWARD), pages 1–8. IEEE.
Braun, R. and Esswein, W. (2014a). Classification of
domain-specific BPMN extensions. In IFIP Working
Conference on The Practice of Enterprise Modeling,
pages 42–57. Springer.
Braun, R. and Esswein, W. (2014b). Entwicklung einer
BPMN-Extension f
ur ressourcen-intensive Prozesse
im Maschinenbau. Tagungsband Multikonferenz
Wirtschaftsinformatik, 2014:1574–1586.
Braun, R., Schlieter, H., Burwitz, M., and Esswein, W.
(2014). BPMN4CP: Design and implementation of
a BPMN extension for clinical pathways. In 2014
IEEE International Conference on Bioinformatics and
Biomedicine (BIBM), pages 9–16. IEEE.
Camunda (2019). Workflow and Decision Automation Plat-
form, (2019-09-03).
Domingos, D., Resp
ıcio, A., and Martinho, R. (2016). Us-
ing resource reliability in BPMN processes. Procedia
Computer Science, 100:1280–1288.
orndorfer, J. and Seel, C. (2017). A meta model based
extension of BPMN 2.0 for mobile context sensi-
tive business processes and applications. 13. Inter-
nationale Tagung Wirtschaftsinformatik (WI2017).
Dumas, M., La Rosa, M., Mendling, J., and Reijers, H. A.
(2018). Fundamentals of Business Process Manage-
ment. Springer, second edition.
Fall, K. (2003). A delay-tolerant network architecture for
challenged internets. In Proceedings of the 2003 con-
ference on Applications, technologies, architectures,
and protocols for computer communications, pages
27–34. ACM.
Geiger, M., Harrer, S., Lenhard, J., and Wirtz, G. (2018).
BPMN 2.0: The state of support and implementation.
Future Generation Computer Systems, 80:250–262.
Graja, I., Kallel, S., Guermouche, N., and Kacem, A. H.
(2016). BPMN4CPS: A BPMN extension for mod-
eling cyber-physical systems. In 25th International
Conference on Enabling Technologies: Infrastructure
for Collaborative Enterprises (WETICE), pages 152–
157. IEEE.
ISO (2013). Internation Organization for Standardization:
ISO:IEC 19510:2013, Information technology - Ob-
ject Management Group Business Process Model and
Kliazovich, D. and Granelli, F. (2008). Packet concate-
nation at the ip level for performance enhancement
in wireless local area networks. Wireless Networks,
Kopp, O., Binz, T., Breitenb
ucher, U., and Leymann, F.
(2012). BPMN4TOSCA: A domain-specific language
to model management plans for composite applica-
tions. In International Workshop on Business Process
Modeling Notation, pages 38–52. Springer.
Kozel, T. (2010). BPMN mobilisation. In Proceedings of
the European Conference of Systems: World Scientific
and Engineering Academy and Society, WSEAS.
Martinho, R. and Domingos, D. (2014). Quality of informa-
tion and access cost of IoT resources in BPMN pro-
cesses. Procedia Technology, 16:737–744.
Mazzola, L., Kapahnke, P., Waibel, P., Hochreiner, C.,
and Klusch, M. (2017). FCE4BPMN: On-demand
qos-based optimised process model execution in the
cloud. In 2017 International Conference on Engineer-
ing, Technology and Innovation (ICE/ITMC), pages
305–314. IEEE.
Meyer, S., Ruppen, A., and Hilty, L. (2015). The things of
the internet of things in BPMN. In International con-
ference on advanced information systems engineer-
ing, pages 285–297. Springer.
Meyer, S., Ruppen, A., and Magerkurth, C. (2013). In-
ternet of things-aware process modeling: integrating
IoT devices as business process resources. In Inter-
national conference on advanced information systems
engineering, pages 84–98. Springer.
OMG (2011). Object Management Group: Business Pro-
cess Model and Notation (BPMN) 2.0 Specification, (2019-
OMG (2014). Object Management Group: MetaObject Fa-
cility Specification, (2019-09-03).
OMG (2019). Object Management Group: Deci-
sion Model and Notation (DMN) 1.2 Specification, (2019-09-03).
OPeRAte (2019). Osnabrueck University of Applied
Sciences: OPeRAte project, http://operate.edvsz.hs- (2019-09-03).
Pelusi, L., Passarella, A., and Conti, M. (2006). Oppor-
tunistic networking: data forwarding in disconnected
mobile ad hoc networks. IEEE Communications Mag-
azine, 44(11):134–141.
ıcio, A. and Domingos, D. (2015). Reliability of
BPMN business processes. Procedia Computer Sci-
ence, 64:643–650.
Stroppi, L. J. R., Chiotti, O., and Villarreal, P. D. (2011a).
A BPMN 2.0 extension to define the resource perspec-
tive of business process models. In XIV Congreso
Iberoamericano en Software Engineering.
Stroppi, L. J. R., Chiotti, O., and Villarreal, P. D. (2011b).
Extending BPMN 2.0: method and tool support. In
International Workshop on Business Process Model-
ing Notation, pages 59–73. Springer.
Yousfi, A., Bauer, C., Saidi, R., and Dey, A. K. (2016).
uBPMN: A BPMN extension for modeling ubiquitous
business processes. Information and Software Tech-
nology, 74:55–68.