PreprintPDF Available

Realizing A Digital Twin of An Organization Using Action-oriented Process Mining

Authors:

Abstract and Figures

A Digital Twin of an Organization (DTO) is a mirrored representation of an organization, aiming to improve the business process of the organization by providing a transparent view over the process and automating management actions to deal with existing and potential risks. Unlike wide applications of digital twins to product design and predictive maintenance, no concrete realizations of DTOs for business process improvement have been studied. In this work, we aim to realize DTOs using action-oriented process mining, a collection of techniques to evaluate violations of constraints and produce the required actions. To this end, we suggest a digital twin interface model as a transparent representation of an organization describing the current state of business processes and possible configurations in underlying information systems. By interacting with the representation, process analysts can elicit constraints and actions that will be continuously monitored and triggered by an action engine to improve business processes. We have implemented a web service to support it and evaluated the feasibility of the proposed approach by conducting a case study using an artificial information system supporting an order handling process.
Content may be subject to copyright.
Realizing A Digital Twin of An Organization
Using Action-oriented Process Mining
Gyunam Park and Wil M.P. van der Aalst
Process and Data Science Group (PADS)
Department of Computer Science, RWTH Aachen University,Aachen,Germany
{gnpark, wvdaalst}@pads.rwth-aachen.de
Abstract—A Digital Twin of an Organization (DTO) is a mir-
rored representation of an organization, aiming to improve the
business process of the organization by providing a transparent
view over the process and automating management actions to
deal with existing and potential risks. Unlike wide applications
of digital twins to product design and predictive maintenance, no
concrete realizations of DTOs for business process improvement
have been studied. In this work, we aim to realize DTOs
using action-oriented process mining, a collection of techniques
to evaluate violations of constraints and produce the required
actions. To this end, we suggest a digital twin interface model as
a transparent representation of an organization describing the
current state of business processes and possible configurations
in underlying information systems. By interacting with the
representation, process analysts can elicit constraints and actions
that will be continuously monitored and triggered by an action
engine to improve business processes. We have implemented a
web service to support it and evaluated the feasibility of the
proposed approach by conducting a case study using an artificial
information system supporting an order handling process. Key-
words—Digital Twin, Action-Oriented Process Mining, Process
Improvement, Hybrid Intelligence, Action Patterns
I . I NTRODUCTION
Organizations are required to continuously improve their
business processes to react to dynamically changing situations
in business environments [1]. While the necessity of continuous
process improvement is hardly questioned, it is non-trivial to
effectively analyze business processes and identify improve-
ment points due to the complexity of business processes
and information systems supporting them [2]. For instance,
a standard business process of an organization supported by an
SAP ERP system (e.g., Order-to-Cash and Procure-to-Pay) may
contain 80,000 variants of process instances, making analysis
of the process challenging.
A digital twin is a digitally mirrored, transparent represen-
tation of a complex real-life object or process [3]. A Digital
Twin of an Organization (DTO) is a digital twin extended
to the whole organization where such objects and processes
belong [4]. DTOs are used to facilitate analysis of business
processes in the organization and continuously improve the
process with the automated execution of management actions.
Despite its tremendous benefits for continuous process improve-
ment, a concrete realization and implementation of DTOs are
missing both in research and in practice [5].
In this paper, we aim to realize DTOs using action-oriented
process mining, a collection of techniques to continuously
monitor violations of constraints in operational processes and
automatically trigger actions based on the monitoring results
in such a way that they update the configuration of information
systems [6]. If delays (are expected to) happen for approving
orders (i.e., constraint), the approval can be temporarily skipped
for upcoming orders by adjusting the system configuration (i.e.,
action). The techniques include conformance checking [7] and
predictive process monitoring [8].
To that end, we suggest a digital twin interface model
as a representation of a business process and its supporting
information systems in an organization. The representation
provides transparent views of the current state of the process
(e.g., a bottleneck in an activity) and possible configurations
in the system (e.g., skipping the activity). Interacting with it,
process analysts can elicit constraints by analyzing the state
of business processes and define actions based on the possible
configurations. Subsequently, an action engine continuously
monitors the constraints and automates corresponding actions.
Figure 1:
An overview of the approach to realize DTOs based on action-
oriented process mining
Fig. 1 describes the realization more in detail. A digital twin
interface model replicates business processes supported by in-
formation systems, using Object-Centric Petri Nets (OCPN) [9]
as formal representations. Event data, recorded by the system
during the execution of the process (e.g., Object-Centric Event
Log (OCEL) [10], are used to construct a digital twin interface
model along with some expert knowledge. An operational
view describes the current status of the process, being updated
using streaming event data. It describes which objects reside
in which phase of the process (e.g., o1 and o2 waiting for
send notification) and various diagnostics (e.g., the average
waiting time for two days is 2 hours). A control view specifies
the current configuration of the system. For instance, valve1
is currently set to 200. Based on the operational and control
view, process analysts can define action patterns specifying the
common pattern between constraints and actions. Using the
pattern, an action engine monitors the process and produces
necessary actions that update the configuration of the system.
In order to evaluate the feasibility of realizing DTOs, we
have implemented a web application to support the digital twin
interface model and action engine. In addition, we have tested
the implementation with an artificial information system that
supports an order handling process. The details and sources
of the implementation and the information system are publicly
available via https://github.com/gyunamister/dtween.
The remainder is organized as follows. We discuss the
related work in Sec. II. Next, we present the preliminaries
in Sec. III. In Sec. IV, we define a digital twin interface
model along with its operational and control views. Next,
Sec. V presents action patterns and an action engine. Afterward,
Sec. VI introduces the implementation of the web application
and Sec. VII provides a case study based on a simulated
business process. Finally, Sec. VIII concludes the paper.
I I . R E L AT E D WO R K
A. Digital Twin
A digital twin is a digital replication of a real system, e.g., a
production process or even a whole organization. Conceptual
frameworks of various digital twins have been widely studied.
In [11], a conceptual framework for product design, product
manufacturing, and product service has been suggested, while
[12] providing a conceptual model of digital twins for smart
manufacturing service. [4] and [5] propose conceptual frame-
works of DTOs by extending a conventional digital twin in a
manufacturing domain to an entire organization.
Contrary to various conceptualizations of digital twins in
extensive fields, concrete realizations only exist for product
designs, product developments, and productions, which focus
on digitalizing physical assets using IoT sensors [4]. For in-
stance, NASA applies the digital twin in developing vehicles to
predict the future status of vehicles using simulation models [3].
However, applications of DTOs for continuous improvements
of business processes have not been studied. In this work,
we realize DTOs aiming at improving business processes by
providing transparent views to the process and automating
necessary actions.
B. Action-Oriented Process Mining
We deploy techniques for action-oriented process mining
to realize DTOs, including 1) ones for monitoring violations
of constraints and 2) ones for triggering actions based on
monitoring results. First, extensive literature exists for mon-
itoring techniques [7]. [13] suggests a technique to monitor
(temporarily) satisfied, (temporarily) violated constraints based
on Linear Temporal Logic (LTL) and colored automata. In [14],
constraints are formulated into Petri net patterns, which are
then evaluated by alignment-based conformance checking.
Predictive monitoring techniques enable violations to be
detected before they actually happen [8]. In [15], a scenario-
based predictive approach is proposed to predict the future
behavior of a business process by deploying system dynamics.
Recently, techniques based on deep neural networks have been
widely studied, e.g., [16] suggests a predictive method based
on Long-Short Term Memory (LSTM) networks.
Next, several techniques have been proposed to automatically
generate specific types of actions to improve business processes.
In [17], the resource allocation is optimized by predicting the
risk of process instances. In [18], a prescriptive alarm system
has been proposed to generate alarms by computing trade-offs
among different interventions.
Instead of focusing on a specific type of action, [19] provides
support to turn diagnostics into extensive actions. It generates
signals by analyzing event data and executes different actions
corresponding to the signals to source systems. In [6], a more
systematic approach is proposed to transform the process-
centric diagnostics into management actions.
III. PRELIMINARIES
In this work, we use object-centric Petri nets as a core formal
representation of a digital twin interface model. First, a Petri
net is a directed graph having places and transitions as nodes,
and flow relations as edges. A labeled Petri net is a Petri net
where the transitions are labeled.
Definition 1
(Labeled Petri Net)
.
Let
Uact
be the universe of
activity names. A labeled Petri net is a tuple
N=(P, T , F, l)
with
P
the set of places,
T
the set of transitions,
PT=
,
F(P×T)(T×P)
the flow relation, and
lT6→ Uact
a labeling function.
A marking
MN∈ B(P)
represents the state of a Petri net as
a multiset of places. A transition
tT
is enabled in marking
MN
if its input places contain at least one token, and it may
fire by removing one token from each of the input places and
producing one token for each of the output places.
In an OCPN, each place is associated with an object type,
enabling it to represent interactions among different object
types. Moreover, variable arcs are used to represent the con-
sumption/production of a variable amount of tokens in one
step.
Figure 2: An example of object-centric Petri nets
Definition 2
(Object-Centric Petri Net)
.
Let
Uot
be the
universe of object types. An object-centric Petri net is a tuple
ON =(N, pt , Fvar )
where
N=(P, T , F, l)
is a labeled Petri net,
pt PUot
maps places onto object types, and
Fvar F
is the subset of variable arcs.
Fig. 2 shows an OCPN:
P={o1, . . . , o5, i1, . . . , i5}
,
T={t1, t2, . . . , t6},F={(o1, t1),(i1, t1),(o2, t2), . . . },l(t1)
=place order
,
l(t2)=send invoice
, etc.,
pt(o1)=Order
,
pt(i1)=Item
, etc., and
Fvar ={(i1, t1),(t1, i2),(i4, t6),(t6, i5
)}, where Order, I tem Uot .
A marking represents the state of an OCPN.
Definition 3
(Marking)
.
Let
Uoi
be the universe of object
identifiers. Let
ON =(N, pt , Fvar )
be an object-centric Petri
net, where
N=(P, T , F, l)
.
otyp Uoi Uot
assigns object
types to object identifiers.
QON ={(p, oi)P×Uoi |
otyp(oi)=pt (p)}
is the set of possible tokens. A marking
M
of ON is a multiset of tokens, i.e., M∈ B(QON ).
For instance, marking
M1=[(o1,0092),(i1,10085),(i1,100
86),(i1,10087)]
denotes four tokens among which place o1
has one token referring to object 0092 and
i1
has three tokens
referring to objects 10085,10086, and 10087.
The concept of binding is used to explain the semantics of an
OCPN. A binding describes the execution of a transition con-
suming objects from its input places and producing objects for
its output places. A binding
(t, b)
is a tuple of transition
t
and
function
b
mapping the object types of the surrounding places
to sets of object identifiers. For instance,
(t1, b1)
describes the
execution of transition
t1
with
b1
where
b1(Order)={0092}
and
b1(Item)={10085,10086,10087}
, where Order and Item
are the object types of the surrounding places of
t1
(i.e.,
o1, i1, o2, i2).
A binding
(t, b)
is enabled in marking
M
if all the objects
specified by
b
exist in the input places of
t
. For instance,
(t1, b1)
is enabled in marking
M1
since 0092, 10085, 10086, and 10087
exist in its input places, i.e., o1, i1.
If
t1
fires with
(t1, b1)
,0092 is removed from
o1
and added to
o2
. Besides,
10085
,
10086
, and
10087
are
removed from
i1
and added to
i2
, resulting in new marking
M2=[(o2,0092),(i2,10085),(i2,10086),(i2,10087)]
. For
more details of OCPNs, we refer readers to [9].
I V. D I G I T A L TWIN IN T E R F A C E MO D E L
In this section, we first introduce a Digital Twin Interface
Model (DT-IM) with its core components. Next, we explain
an operational view of the DT-IM describing the current states
of business processes. Afterward, we introduce a control view
of the DT-IM depicting the current configurations and possible
actions defined over target information systems.
A DT-IM represents process behaviors using OCPNs as for-
mal representations. Moreover, it describes the routing/resource
allocation rules of the process using guards. A guard is
a formula defined over attributes using relational operators
(
,
,
=
) as well as logical operators such as conjunction (
),
disjunction (
), and negation (
¬
). Associated with a transition,
a guard extends the semantics of OCPNs by allowing the
transition to fire not only if the transition is enabled in marking
M
, but also if the guard evaluates to
true
. We denote with
F(X)
the set of such formulas defined over a set
X
of attributes.
Valves are configuration and options in information systems and
can be used in guards of the OCPN (e.g., minimum threshold
to send notifications).
Definition 4
(Digital Twin Interface Model (DT-IM))
.
Let
Uattr
be the universe of attribute names. A digital twin
interface model, denoted as
DT
, is a tuple
(ON, V, G)
where
Figure 3:
(a) a DT-IM replicating an order handling process (b) an operational
state of the DT-IM (c) a configuration of the DT-IM
ON =(N , pt, Fvar)
is an object-centric Petri net, where
N= (P, T , F, l),
VUattr is a set of valves, and
GT(F(Uattr ){true})
associates transitions with
guards.
Fig. 3(a) is an example of DT-IMs (
DT1
) replicating the
information system that supports an order handling process.
The OCPN in Fig. 3(a) describes the control flow of the order
handling process where four different types of objects (i.e.,
order in red, item in blue, route in green, and resource in ivory)
interact. Note that some arcs connecting
p1
and transactions,
e.g.,
(p1, t2)
and
(t2, p1)
, are omitted for better representations.
DT1
uses valves, such as po-group,po-price, and capacity to
define guards. For instance, g1 describes a rule that a resource
from the sales department with the role of valve po-group
is eligible to perform the activity. g2 describes that only
orders with a price higher than or equal to valve po-price
can be further processed if the current capacity is below valve
capacity.
Conceptually, a guard
G(t)
may consist of two parts: one
formulating the resource allocation rules,
Gres(t)
, and the other
formulating the routing rules,
Grout(t)
, i.e.,
G(t)=Gres(t)
Grout(t)
. In Fig. 3(a), g1 describes a resource allocation rule
for place order, whereas g2 describes the routing rule. The
resource allocation and routing rules form guard
G(t1)
, i.e.,
G(t1)=g1 g2
. In other words,
t1
, when enabled at marking
M
, fires only if the corresponding resource is from po-group,
the price is higher than po-price, and the capacity of the process
is below capacity.
A. Operational View of A Digital Twin Interface Model
Using DT-IMs, we can describe the current operational state
of business processes. The state is represented by marking
and diagnostics. A marking represents which objects reside
in which parts of business processes using OCPNs, while
diagnostics represent performance/compliance of business pro-
cesses, e.g., average waiting time for sending notifications in
the last two days (avg-wt-sn-2d). The diagnostics are computed
using data-driven, model-based analysis such as token-based
replay [20]. Many diagnostics are available in a DT-IM, but
we leave them deliberately vague, denoting
DT
to be the set
of all possible diagnostics of digital twin interface model
DT
.
Definition 5
(Operational State of A DT-IM)
.
Let
Uval
be the
universe of attribute values. Let
Uvmap =Uattr 9 Uval
be the
set of all partial functions mapping a subset of attribute names
onto the corresponding values. Let
DT =(ON , V, G)
be a DT-
IM, where
ON =(N, pt , Fvar )
. An operational state of
DT
is
a tuple OS=(M, ovmap, dmap)where
M∈ B(QON )is a marking of ON ,
ovmap OUvmap
is an object value assignment
where O={oi|(p, oi)M}, and
dmap DT 9 R
is a diagnostics assignment such
that, for any
diag DT
,
dmap(diag) =
if
diag /
dom(dmap).
We denote DT to be the set of all possible states of DT .
Fig. 3(b) describes state OS1=(M1, ovmap1, dmap1)wh-
ere M1=[(p1,Nick),(p4,O-56),(p5,I-97)] denotes resource
token
Nick
, order token
O-56
, and item token
I-97
residing in
p1
,
p4
and
p5
, respectively. Besides,
ovmap1(Nick)(group)=
sales
,
ovmap1(O-56)(price)=599
,
ovmap1(I-97)(material)=
iPad
, etc., and
dmap(avg-wt-sn-2d)=2 (hour)
where
avg-wt-sn-2d DT1.
Operational states of a DT-IM change according to events
from information systems. Here, we abstract from the mecha-
nism to update states based on events. In Sec. VI, we introduce
an implementation based on replaying events on OCPNs.
Definition 6
(Operation Engine)
.
Let
DT =(ON , V, G)
be
a DT-IM. An operation engine of
DT
,
opDT
, updates the
operational states of DT , i.e., opDT DT D T .
For instance, an event of send notification about order
O-56
updates
OS1
to
OS2=(M2, ovmap2, dmap2)D T1
where
M2=[(p1,Nick),(p6,O-56),(p5,I-97)]
,
ovmap2(O-56)(notif-
ied)=true, and dmap2(avg-wt-sn-2d)=2.5(hour).
B. Control View of A Digital Twin Interface Model
A DT-IM is also used to describe controls of the information
system. The control is characterized by the value assignment of
valves, called configuration. Note that the configuration does
not enforce the behaviors of business processes since deviations
may happen in business processes.
Definition 7
(Configuration)
.
Let
DT =(ON , V, G)
be a DT-
IM. A configuration
conf VUval
assigns values to
valves. ΣDT is the set of all possible configurations of DT .
Fig. 3(c) describes the configuration,
conf1ΣDT1
where
conf1(sn-price)=200,conf1(caw-qty)=10, etc.
Definition 8
(Action)
.
Let
DT =(ON , V, G)
be a DT-IM. An
action
act ΣDT ΣDT
updates the configuration.
ADT
denotes the set of all possible actions defined over DT .
For instance, skip-sn
ADT1
increases valve sn-price
to infinity so that no orders go through send notifica-
tion, i.e.,
skip-sn(conf1)=conf2
where
conf2ΣDT1
and
conf2(sn-price)=.
V. A CTION PA T T E R N S A N D ACTION ENGINES
Using operational and control views of DT-IMs, process
analysts can define action patterns. An action pattern describes
the automated execution of actions to recurring problems in
business processes. Each pattern consists of a constraint that
specifies an undesired situation and a corresponding action to
resolve the situation. A constraint is a formula defined over
diagnostics using relational operators and logical operators.
Definition 9
(Action Pattern)
.
Let
DT
be a DT-IM.
ΠDT =F(∆DT )×ADT
is the set of all possible action patterns
of DT .
Given
DT1
in Fig. 3, we can define action pattern
ap1
that skips send-notification if the average waiting time
for the activity in the last two days is higher than 2
hours, i.e.,
ap1=([avg-wt-sn-2d>2h],skip-sn)ΠDT1
where
avg-wt-sn-2d DT1skip-sn ADT1.
Definition 10
(Action Engine)
.
Let
DT =(ON , V, G)
be a DT-
IM. An action engine
aeDT DT ×DT ×ΣD T )ADT
generates actions given action patterns, operational states, and
configurations, i.e., for any
ap=(constr, act)ΠDT
,
OS
DT
, and
conf ΣDT
,
aeDT (ap, OS, conf )=act
if
constr
evaluates to true w.r.t.
OS
and
conf
.
aeDT (ap, OS, conf )=
otherwise.
In the remainder, we introduce the taxonomy of constraint
violations and management actions. Note that our goal is not
to provide completely exhaustive taxonomies, but to facilitate
the elicitation of various action patterns by providing possible
constraints defined over operational states of a DT-IM and
actions based on its configuration.
A. A Taxonomy of Constraint Violations
Fig. 4 shows the taxonomy of constraint violations in
UML class diagram. First, violations are categorized by their
orientations: compliance and performance [21]. Compliance-
oriented violations refer to the violations of rules during the
execution of business processes. In the process shown in
Fig. 3(a), skipping place order is considered as compliance-
oriented violations. Performance-oriented violations concern
undesired performances that are measured with various process
performance metrics. The high average waiting time for the
notification to customers in the order handling process is a
performance-oriented diagnostic.
We further distinguish the compliance-oriented violations
based on their types, i.e., existence and non-existence. The
existence concerns unnecessary executions of activities in
business processes, while the non-existence refers to skipping
necessary executions of activities. In Fig. 3, notifications are
sent for the orders higher than
$
200. Sending notifications
for orders lower than
$
200 is concerned with the existence,
Figure 4: A taxonomy of constraint violations
whereas skipping such notifications for orders higher than
$
200
is related to the non-existence. Furthermore, the violations
can consider either a single activity or a set of activities (i.e.,
sub-process). In Fig. 3, a violation of the existence can be
defined not only for sending notifications, but also together
with collecting payments.
Performance-oriented violations are further divided into
temporal and non-temporal ones. The former involves abso-
lute/relative time information in its definition, whereas the
latter is not concerned with time information. For instance,
the high average sojourn time for sending notifications relates
to the time information, whereas the high frequency of sending
notifications does not involve time information.
B. A Taxonomy of Management Actions
Fig. 5 shows the taxonomy of management actions to miti-
gate risks resulting from violations of constraints in UML class
diagram. First, inflow controls the execution of certain activities
in business processes by blocking or allowing the execution.
In Fig. 3, we can block all executions of sending notifications
by increasing the minimum threshold to infinity. Besides, we
can block some of the executions by configuring the threshold
to the value higher than the current one, e.g.,
$
300. In contrast,
we may allow the execution of activities by augmenting valves.
Suppose that ap-quantity in Fig. 3(a) is set to 0, which means
all items need approvals for checking availability. We can allow
some skips by increasing ap-quantity.
Second, routing is to control the routing of objects in busi-
ness processes, including parallelism,extra,alternative, and
skipping. First, parallel routing is to parallelize the execution
of activities. In case that a parallel execution exists in the
process model (e.g., a choice between parallelizing send invoice
and send notification and sequentially processing them as
described in Fig. 3), we can adjust the routing to enable parallel
executions for some/all objects. Next, extra routing is to execute
extra activities for certain objects (e.g., further approval steps
for specific types of items), while alternative routing is to adopt
alternative control flows for certain objects (e.g., customer
pick-ups of items without routes). Finally, skip routing is to
deliberately bypass the execution of certain activities. For
instance, we can skip the notification for the orders having
a price higher than $200 by setting sn-price to 200 in Fig. 3.
Figure 5:
A taxonomy of management actions. Note that we only specify
All/Some for Block/Decrease, but they exist for all the classes in the same
level.
Third, resource allocation is to control the assignment of
resources to activities based on different attributes of resources,
e.g., group,role, and experience. First, we can change the
resource allocation by adjusting group information. In Fig. 3,
we can limit the execution of place order only by resources
from sales department. Moreover, the assignment rule can be
changed according to the role and experience of resources. In
Fig. 3, by changing po-role to staff, the resource allocation for
place order can be updated. Likewise, by increasing the level
of po-exp, the allocation is accordingly updated.
Finally, one can control the capacity of business processes
either by increasing or decreasing the maximum amount of
process instances, e.g., manufacturing products, based on the
situation of the process. For instance, we can increase the
capacity of the order handling process by decreasing valve
capacity. Contrary to that, we can decrease the capacity by
increasing valve capacity.
V I . I MPLEMENTATION
A cloud-based web service
1
is implemented to support the
digital twin interface model and action engine with a dedicated
user interface. The implementation consists of four major
functional components: (1) building DT-IMs, (2) updating
states of DT-IMs, (3) visualizing DT-IMs, and (4) evaluating
action patterns. All four functional components are imple-
mented as Python packages. Being containerized as a Docker
container, the functionality of the framework and its functional
components is structured into a coherent set of microservices
that are deployable on any platform.
In the following, we briefly explain each component. We
refer readers to the tool manual1for more details.
A. Building Digital Twin Interface Models: This compo-
nent aims at building DT-IMs given event data and user inputs.
The main inputs of this component are event data, guards,
and valves, whereas the output is a DT-IM. The event data
are stored as standard OCEL format [10], while the guards
1sources & manuals available at https://github.com/gyunamister/dtween
and valves are stored as JSON-based formats
1
. The discovery
technique introduced in [9] is deployed to discover an OCPN
from the event data. User-provided guards and valves are used
to enrich the discovered OCPN, completing a DT-IM.
B. Updating States of Digital Twin Interface Models:
The goal of this component is to update states of a DT-IM
accordingly to the updates of information systems. The main
inputs include streaming event data in the standard OCEL
format and a DT-IM built using the component in Sec. VI.
The token-based replay [20] technique is used to compute
diagnostics using the streaming event data in user-defined
intervals (e.g., every 24 hours). Likewise, markings and object
value assignments are computed, completing operational states
of the DT-IM.
C. Visualizing Digital Twin Interface Models: This com-
ponent provides visual information to users to facilitate the
elicitation of action patterns. First, it provides the control view
of a DT-IM, describing the routing and resource allocation rules
and possible controls. Second, it provides the operational view
showing the current states of a DT-IM. Fig. 6 is a screenshot
of the web service showing the operational state. The OCPN
describes the process model and the green box specifies the
marking at place order3 where four objects reside at 11:22:54,
17.June.2021. The table on the right hand side describes the
object value assignment, e.g., object o22 is an order having
price
$
174. Using operational states in the operational view and
configurations in the control view, users can define constraints
and actions, which are used to compose action patterns.
D. Action Engine: This component aims to continuously
evaluate action patterns by monitoring the business process.
The main inputs include states of a DT-IM computed by
the component in Sec. VI, current configurations, and user-
defined action patterns. Based on the states, action engine
evaluates the constraint specified in the action pattern and
produce the corresponding action. The generated actions are
applied to update the configuration of the DT-IM, which in turn
is consumed by the target system to update its configuration.
VII. PR O O F O F CO N C E P T
To evaluate the feasibility of realizing DTOs, we conduct a
case study using an artificial information system that supports
the order handling process introduced in Sec. IV. The imple-
mentation of the information system is publicly available
1
to be
used for demonstrations. 14 resources are available in total at
any point in time, each of them being responsible for multiple
activities in the process. Orders are randomly placed and the
resources process the orders and the corresponding items and
routes based on First-in-First-out rule.
Multiple configurations are available in the information
system. For instance, sn-price determines the minimum price of
the orders that require the notification for payments. Moreover,
caw-price determines the minimum price of the items requiring
approvals for checking availability. The information system
generates event logs recording the execution of the process in
the standard OCEL format.
Using the implementation, we 1) construct a DT-IM for the
information system, 2) define action patterns, and 3) evaluate
action patterns by monitoring the process.
A. Constructing A Digital Twin Interface Model
We first build a DT-IM (
DT 0
) that replicates the behavior
of the information system. The OCEL containing events
from the information system is used to discover an OCPN.
Fig. 7(a) shows a screenshot of the control view of the DT-
IM. As described in Fig. 3(a), guards are defined with the
valves of the information system. Fig. 8(a) is a screenshot
of the operational view showing diagnostics of the process
at 23:59:59, 02.May.2021. The average/median sojourn time
of send notification for the last two days (
avg-st-sn-2d
DT 0
), i.e., between 00:00:00, 01.May.2021 and 23:59:59,
02.May.2021 is 16 hours. The routing probability to check
availability with approval (
rout-prob-caw-2d DT 0
) for the
last two days is 0.3.
B. Defining Action Patterns
We define constraints and actions based on the operational
and control view, respectively. By analyzing the diagnostics
from the operational view, we conclude that it is problematic
for the average sojourn time for the previous two days to
be higher than 16 hours, whereas the problematic situation is
considered to be relieved if the average sojourn time reaches
10 hours. Based on it, the following constraints are defined:
delay-on-notification (C1) evaluates if the average sojourn
time for two days is higher than 16 hours, i.e., [avg-st-sn-
2d
>16
(hours)]. Fig. 8(b) illustrates how we define C1
using the user-interface of the implementation.
relaxed-delay-on-notification (C2) evaluates if the average
sojourn time for two days is lower than 10 hours, i.e.,
[avg-st-sn-2d <10 (hours)]
Next, using the valves in the control view, we define the
following actions:
skip-more-notification (A1) ADT 0
sets sn-price to 220,
i.e., for any
conf, conf 0ΣDT 0
,
A1(conf)=conf0
such
that
conf0(sn-price)=220
and
vdom(conf)\{sn-price}
conf0(v)=conf(v)
. Fig. 8(a) illustrates how we define
A1 using the user-interface of the implementation.
skip-less-notification (A2) AD T 0
sets sn-price to 180,
i.e., for any
conf, conf 0ΣDT 0
,
A2(conf)=conf0
such
that
conf0(sn-price)=180
and
vdom(conf)\{sn-price}
conf0(v)=conf(v).
Based on the constraints and actions, we define following
action patterns:
AP1=(C1,A1),AP2=(C2,A2)ΠDT 0
.AP1
evaluates if the average sojourn time for two days is higher
than 16 hours and generates actions to skip more notifications,
whereas AP2 generates actions to skip fewer notifications by
evaluating if the average sojourn time is lower than 10 hours.
C. Monitoring
Using action patterns and the action engine of the imple-
mentation, we analyze the event stream, which is continuously
generated by the information system. AP1 and AP2 are
Figure 6:
Operational view of the DT-IM replicating the artificial information system. The green box denotes the objects residing in place order3 and the
orange box indicates the value assignment of each object.
Figure 7:
(a) Control view of the DT-IM replicating the artificial information
system. Note that it only describes a part of the guards for a better represen-
tation. (b) Defining skip-more-notification (A1) using the user interface of the
implementation.
evaluated every 12 hours for 24 days, i.e., from 03.May.2021
to 26.May.2021. Fig. 9 reports the average sojourn time of
sending a notification for the last two days by time. The
red line denotes the average sojourn time when we trigger
the actions defined by the action patterns. At 12:00:00 on
06.May.2021, A1 was triggered since C1 was satisfied (i.e., the
Figure 8:
(a) Diagnostics between 00:00:00 01.May.2021 and 23:59:59
02.May.2021. (b) Defining delay-on-notification (C1) using the user interface
of the user interface of the implementation.
average sojourn time was higher than 16). As a result of the
action, sn-price of the DT-IM changed to 220, which in turn
updated configurations of the information system. After A1,
the average sojourn time had decreased until A2 was triggered
after the average sojourn time reached below 10. A1 was taken
again on 18.May.2021 and 25.May.2021 to deal with delays
in sending notifications.
Figure 9:
The red dotted line depicts the average sojourn time when actions
are triggered, whereas the blue line shows one when actions are not taken.
Red/orange dots indicate the execution of A1/A2.
To evaluate the effectiveness of the action, we measure the
average sojourn time when actions are not taken. The blue line
depicts the average sojourn time, showing that the execution
of actions incredibly reduces the average sojourn time after the
first execution of A1. For instance, at 12:00:00 10.May.2021,
the average sojourn time with actions is 9, whereas the average
sojourn time without actions is 20.
VIII. CONCLUSION
In this paper, we proposed a digital twin interface model
to realize DTOs based on action-oriented process mining. An
operational view of the interface model reflects the current
states of business processes with various diagnostics, while
a control view provides routing and resource allocation rules
of the process. Process analysts can define action patterns
based on the operational and control views, which are eval-
uated by an action engine to produce the required actions.
We have implemented a web service that supports building
digital twins, updating states, visualizing the digital twin, and
monitoring action patterns. To evaluate the feasibility of the
proposed concept, we conducted a case study using an artificial
information system supporting an order handling process.
However, given the practical relevance of the proposed ap-
proach, the applicability to real-life information systems needs
to be extensively evaluated. The evaluation should include the
construction of DT-IMs for real-life business processes, real-
time updates of operational states, and automated execution of
actions. As future work, we plan to conduct case studies with
real-life information systems. Moreover, we plan to develop a
feedback loop that provides suggestions to improve the existing
action patterns. Finally, we will evaluate the inter-effects of
different actions to provide recommendations for optimizing
existing action patterns.
REFERENCES
[1]
L. Reinkemeyer, “Business view: Towards a digital enabled organization,”
in Process Mining in Action: Principles, Use Cases and Outlook, 2020,
pp. 197–206.
[2]
Y. Alotaibi and F. Liu, “Survey of business process management:
challenges and solutions,” Enterprise Information Systems, vol. 11, no. 8,
pp. 1119–1153, 2017.
[3]
E. Glaessgen and D. Stargel, “The digital twin paradigm for future
nasa and u.s. air force vehicles,” in 53rd AIAA/ASME/ASCE/AHS/ASC
Structures, Structural Dynamics and Materials Conference, 2012.
[4]
R. Parmar, A. Leiponen, and L. D. Thomas, “Building an organizational
digital twin,” Business Horizons, vol. 63, no. 6, pp. 725–736, 2020.
[5]
M. Caporuscio, F. Edrisi, M. Hallberg, A. Johannesson, C. Kopf, and
D. Perez-Palacin, “Architectural concerns for digital twin of the organi-
zation,” in Software Architecture, A. Jansen, I. Malavolta, H. Muccini,
I. Ozkaya, and O. Zimmermann, Eds., 2020, pp. 265–280.
[6]
G. Park and W. M. P. van der Aalst, “A general framework for action-
oriented process mining,” in BPM 2020 Workshops, A. Del R
´
ıo Ortega,
H. Leopold, and F. M. Santoro, Eds., 2020, pp. 206–218.
[7]
L. T. Ly, F. M. Maggi, M. Montali, S. Rinderle-Ma, and W. M. P. van der
Aalst, “Compliance monitoring in business processes: Functionalities,
application, and tool-support,” Information Systems, vol. 54, pp. 209–
234, 2015.
[8]
A. E. Marquez-Chamorro, M. Resinas, and A. Ruiz-Cortes, “Predictive
Monitoring of Business Processes: A Survey,IEEE Transactions on
Services Computing, vol. 11, no. 6, pp. 962–977, 2018.
[9]
W. M. P. van der Aalst and A. Berti, “Discovering Object-centric Petri
Nets,” Fundam. Informaticae, vol. 175, no. 1-4, pp. 1–40, 2020.
[10]
A. F. Ghahfarokhi, G. Park, A. Berti, and W. M. P. van der Aalst,
“Ocel: A standard for object-centric event logs,” in New Trends in
Database and Information Systems, L. Bellatreche, M. Dumas, P. Karras,
R. Matulevi
ˇ
cius, A. Awad, M. Weidlich, M. Ivanovi
´
c, and O. Hartig,
Eds., 2021, pp. 169–175.
[11]
F. Tao, J. Cheng, Q. Qi, M. Zhang, H. Zhang, and F. Sui, “Digital twin-
driven product design, manufacturing and service with big data,” The
International Journal of Advanced Manufacturing Technology, vol. 94,
no. 9-12, pp. 3563–3576, 2018.
[12]
Q. Qi, F. Tao, Y. Zuo, and D. Zhao, “Digital Twin Service towards Smart
Manufacturing,” Procedia CIRP, vol. 72, pp. 237–242, 2018.
[13]
F. M. Maggi, M. Montali, M. Westergaard, and W. M. P. van der Aalst,
“Monitoring business constraints with linear temporal logic: An approach
based on colored automata,” in BPM 2011, S. Rinderle-Ma, F. Toumani,
and K. Wolf, Eds., vol. 6896, 2011, pp. 132–147.
[14]
E. Ramezani, D. Fahland, and W. M. P. van der Aalst, “Where did I
misbehave? diagnostic information in compliance checking,” in BPM
2012, A. Barros, A. Gal, and E. Kindler, Eds., vol. 7481, 2012, pp. 262–
278.
[15]
M. Pourbafrani, S. J. van Zelst, and W. M. P. van der Aalst, “Scenario-
based prediction of business processes using system dynamics,” in OTM
2019, H. Panetto, C. Debruyne, M. Hepp, D. Lewis, C. A. Ardagna, and
R. Meersman, Eds., 2019, pp. 422–439.
[16]
J. Evermann, J.-R. Rehse, and P. Fettke, “Predicting process behaviour
using deep learning,” Decision Support Systems, vol. 100, pp. 129–140,
2017.
[17]
R. Conforti, M. de Leoni, M. La Rosa, W. M. P. van der Aalst, and A. H.
ter Hofstede, “A recommendation system for predicting risks across
multiple business process instances,” Decision Support Systems, vol. 69,
pp. 1–19, 2015.
[18]
S. A. Fahrenkrog-Petersen, N. Tax, I. Teinemaa, M. Dumas, M. de Leoni,
F. M. Maggi, and M. Weidlich, “Fire Now, Fire Later: Alarm-Based
Systems for Prescriptive Process Monitoring,arXiv:1905.09568 [cs,
stat], 2019.
[19]
P. Badakhshan, G. Bernhart, J. Geyer-Klingeberg, J. Nakladal, S. Schenk,
and T. Vogelgesang, “The Action Engine – Turning Process Insights into
Action,” in 2019 ICPM Demo Track, 2019, pp. 28–31.
[20]
A. Berti and W. M. P. van der Aalst, “A novel token-based replay
technique to speed up conformance checking and process enhancement,”
Trans. Petri Nets Other Model. Concurr., vol. 15, pp. 1–26, 2021.
[21]
W. M. P. van der Aalst, Process Mining - Data Science in Action, 2016.
ResearchGate has not been able to resolve any citations for this publication.
Conference Paper
Full-text available
The application of process mining techniques to real-life information systems is often challenging. Considering a Purchase to Pay (P2P) process, several case notions such as order and item are involved, interacting with each other. Therefore, creating an event log where events need to relate to a single case (i.e., process instance) leads to convergence (i.e., the duplication of an event related to different cases) and divergence (i.e., the inability to separate events within the same case) problems. To avoid such problems, object-centric event logs have been proposed, where each event can be related to different objects. These can be exploited by a new set of process mining techniques. This paper describes OCEL (Object-Centric Event Log), a generic and scalable format for the storage of object-centric event logs. The implementation of the format can use either JSON or XML, and tool support is provided.
Conference Paper
Full-text available
Process mining provides techniques to extract process-centric knowledge from event data available in information systems. These techniques have been successfully adopted to solve process-related problems in diverse industries. In recent years, the attention of the process mining discipline has shifted to supporting continuous process management and actual process improvement. To this end, techniques for operational support, including predictive process monitoring, have been actively studied to monitor and influence running cases. However, the conversion from insightful diagnostics to actual actions is still left to the user (i.e., the “action part” is missing and outside the scope of today’s process mining tools). In this paper, we propose a general framework for action-oriented process mining that supports the continuous management of operational processes and the automated execution of actions to improve the process. As proof of concept, the framework is implemented in ProM.
Chapter
Full-text available
Employing a Digital Twin of the Organization would help enterprises to change and innovate, thus enhancing their organization’s sustainability. However, the lack of engineering best practices for developing and operating a Digital Twin of the Organization makes it difficult for enterprises to fully benefit from it. Many companies are currently investigating the potential use of it, but available solutions are often context-dependent or system-specific, and challenging to adapt, extend, and reuse. Therefore, digitalization is perceived as a slow, resource-demanding, and extremely expensive process whose outcome is uncertain. To this extent, enterprises seek solutions allowing them to gently introduce a Digital Twin of the Organization into their organization and to evolve it according to the changing needs and situations. This paper reports a first attempt on architecting a Digital Twin of an Organization, and discusses some architectural concerns to be addressed in order to facilitate its development and evolution.
Chapter
Full-text available
Many organizations employ an information system that supports the execution of their business processes. During the execution of these processes, event data are stored in the databases that support the information system. The field of process mining aims to transform such data into actionable insights, which allow business owners to improve their daily operations. For example, a process model describing the actual execution of the process can be easily extracted from the captured event data. Most process mining techniques are “backward-looking” providing compliance and performance information. Few process mining techniques are “forward-looking”. Therefore, in this paper, we propose a novel scenario-based predictive approach that allows us to assess and predict future behavior in business processes. In particular, we propose to use system dynamics to allow for “what-if” questions. We create a system dynamics model using variables trained on the basis of the past behavior of the process, as captured in the event log. This model is used to explore the effect of possibly applied changes in the process as well as roles of external factors, e.g., human behavior. Using real event data, we demonstrate the feasibility of our approach to predict possible consequences of future decisions and policies .
Article
Full-text available
The increasing digitization of the economy means that a new digital resource – a digital twin of an organization – is now feasible. We suggest five principles that assist in the construction of an organizational digital twin and show how they combine into a dynamic evolutionary process that incrementally builds and maintains the digital twin. We also discuss the organizational implications of implementing a digital twin, and how digital twins create value in an organization.
Conference Paper
Full-text available
The Action Engine from Celonis is a new web application that translates findings from automatic discovery and rule-based process analysis into recommendations for operational support during process execution. The engine continuously analyzes data across different systems and processes, communicates personalized recommendations for process-related actions to human users or the digital workforce, and directly executes actions in the source system. This approach contributes to the important field of operational process support by operationalizing insights created through process analysis. The presented capabilities are especially valuable for active process mining users requiring fast suggestions where to take action for improvement of process performance during execution.
Article
Full-text available
Nowadays, process mining is becoming a growing area of interest in business process management (BPM). Process mining consists in the extraction of information from the event logs of a business process. From this information, we can discover process models, monitor and improve our processes. One of the applications of process mining, is the predictive monitoring of business process. The aim of these techniques is the prediction of quantifiable metrics of a running process instance with the generation of predictive models. The most representative approaches for the runtime prediction of business process are summarized in this paper. The different types of computational predictive methods, such as statistical techniques or machine learning approaches, and certain aspects as the type of predicted outcomes and quality evaluation metrics, have been considered for the categorization of these methods. This paper also includes a summary of the basic concepts, as well as a global overview of the process predictive monitoring area, that can be used to support future efforts of researchers and practitioners in this research field.
Article
Full-text available
Nowadays, along with the application of new-generation information technologies in industry and manufacturing, the big data-driven manufacturing era is coming. However, although various big data in the entire product lifecycle, including product design, manufacturing, and service, can be obtained, it can be found that the current research on product lifecycle data mainly focuses on physical products rather than virtual models. Besides, due to the lack of convergence between product physical and virtual space, the data in product lifecycle is isolated, fragmented, and stagnant, which is useless for manufacturing enterprises. These problems lead to low level of efficiency, intelligence, sustainability in product design, manufacturing, and service phases. However, physical product data, virtual product data, and connected data that tie physical and virtual product are needed to support product design, manufacturing, and service. Therefore, how to generate and use converged cyber-physical data to better serve product lifecycle, so as to drive product design, manufacturing, and service to be more efficient, smart, and sustainable, is emphasized and investigated based on our previous study on big data in product lifecycle management. In this paper, a new method for product design, manufacturing, and service driven by digital twin is proposed. The detailed application methods and frameworks of digital twin-driven product design, manufacturing, and service are investigated. Furthermore, three cases are given to illustrate the future applications of digital twin in the three phases of a product respectively.
Article
Techniques to discover Petri nets from event data assume precisely one case identifier per event. These case identifiers are used to correlate events, and the resulting discovered Petri net aims to describe the life-cycle of individual cases. In reality, there is not one possible case notion, but multiple intertwined case notions. For example, events may refer to mixtures of orders, items, packages, customers, and products. A package may refer to multiple items, multiple products, one order, and one customer. Therefore, we need to assume that each event refers to a collection of objects, each having a type (instead of a single case identifier). Such object-centric event logs are closer to data in real-life information systems. From an object-centric event log, we want to discover an object-centric Petri net with places that correspond to object types and transitions that may consume and produce collections of objects of different types. Object-centric Petri nets visualize the complex relationships among objects from different types. This paper discusses a novel process discovery approach implemented in PM4Py. As will be demonstrated, it is indeed feasible to discover holistic process models that can be used to drill-down into specific viewpoints if needed.
Chapter
The business outlook is written by the editor and considers the dimensions business expectations, potentials and benefits, technological developments, market trends, and developments of a digital workforce. The chapter has been structured on a timeline from present trends to short-, mid-, and long-term outlook, concluding in a Vision of a Digital Enabled Organization. The chapter aims to initiate thought processes, stir discussions, stimulate technical developments, and further enhance the power of Process Mining.