Transcript BPM course part III

Dutch-Belgian Database Day 2007
The Challenges of Process Mining
A.J.M.M. Weijters (and many others)
/faculteit technologie management
/faculteit technologie management
Content
• Process mining
• ProM
• Challenges
/faculteit technologie management
Process Mining: basic idea
• The basic idea of process mining is to extract
process knowledge from a registration what
happens during the execution of a process (a so
called event log).
• Process mining provide techniques and tools for
discovering process, control, data, organizational,
and social information from event logs.
• Information about the real behavior within a
process, not the expected behavior.
/faculteit technologie management
supports/
controls
information
system
operational
process
records
refers to
models
configures
process
discovery
(un)desired
properties
process
models
conformance
testing
log-based verification
/faculteit technologie management
event
logs
Process discovery: Reversing the process
Prepare
shipment
Register
order
Ship
goods
(Re)send
bill
Archive
order
Receive
payment
process
discovery
/faculteit technologie management
Contact
customer
Conformance testing
Prepare
shipment
Register
order
Ship
goods
(Re)send
bill
Archive
order
Receive
payment
Requirement
for material
has arisen
Purchase
Requisition
Contact
customer
Requisition
released
for scheduling
agreement
schedule/SA release
Purchase
requisition
released
for purchase
order
Purchasing
Decide To Buy Computer
Order Machine
Goods
received
Choose Model
[desktop]
[bad reviews]
Inbound
delivery
entered
Purchase
order
created
Material
is released
Invoice
received
Save Money
Read Test Reviews
Order Screen
[laptop]
[enough]
Receive Machine
Choose Operating System
Receive Screen
[reviews ok]
[laptop]
[desktop]
[windows]
Order Windows
Goods
Receipt
[linux]
Goods
receipt
posted
Download Linux
Work Hard
Receive Windows
Check Bank Account
Set Up And Connect
Open Lid
[not enough]
Warehouse/
Stores
Plug In And Power On
Install Operating System
/faculteit technologie management
TO item
confirmed
without
differences
Transfer
order
item
is confirmed
Invoice
Verification
Payment
must
be effected
Log based verification
formula
four_eyes_principle
(a1:activity,a2:activity)
:=
forall[p:person |
(!(execute(p,a1)) \/
!(execute(p,a2)))];
/faculteit technologie management
ProM framework
• ProM is open source and uses a plug-able architecture, e.g.
(www.processmining.org)
• people can add new process mining techniques by adding
plug-ins without spending any efforts on the loading and
filtering of event logs and the visualization of the resulting
models.
• ProM 4.2 provides six different types of plug-ins, and in total
more than 200 plug-ins. This makes ProM a practical and
versatile tool for process analysis and discovering.
/faculteit technologie management
Event log
•
•
•
•
•
•
•
Case identifier (Case 33)
Activity (Test if repair is OK)
Time information or ordering
Event type (start, complete, …)
Recourse (John)
Task data (repair = OK)
Case data (telephone type = T1, ...)
/faculteit technologie management
Simple
Example
/faculteit technologie management
Problems: bad performance NL (overtime work/quality)
/faculteit technologie management
Event log
XML format
- <ProcessInstance id="1" description="">
- <AuditTrailEntry>
- <Data>
<Attribute name="phoneT">T1</Attribute>
<Attribute name="countryT">B</Attribute>
</Data>
<WorkflowModelElement>Bregistration</WorkflowModelElement>
<EventType>complete</EventType>
<Timestamp>2006-01-01T00:01:00.000+01:00</Timestamp>
<Originator>Badmin</Originator>
</AuditTrailEntry>
- <AuditTrailEntry>
<WorkflowModelElement>Banalyse</WorkflowModelElement>
<EventType>complete</EventType>
<Timestamp>2006-01-01T00:02:00.000+01:00</Timestamp>
<Originator>BT2</Originator>
</AuditTrailEntry>
/faculteit technologie management
Need for more details
• Performance B seems better than performance
NL, but
– differences between the two sub-processes (B-NL)
– what is/are the bottleneck(s)
– number of cases NL en B
– number of re-repairs
– workload of resources
– difference in performance of the human resources
– ....
/faculteit technologie management
Process Mining (PM) can be used to discover
•
•
•
•
•
•
general log information
a control-flow model
performance information
bottlenecks
social models
extensions like decision rules for an XOR split in
the model
• ...
/faculteit technologie management
/faculteit technologie management
/faculteit technologie management
/faculteit technologie management
/faculteit technologie management
Result of one of the mining techniques
(Heuristics Miner)
Many other control-flow mining techniques
available in ProM:
- α-miner
- Genetic mining algorithm
- Association rules miner
- Region miner
- Fuzzy miner
- ...
/faculteit technologie management
/faculteit technologie management
/faculteit technologie management
/faculteit technologie management
Organization miner
But
NL: 851/11=77.4 ph/w
B: 549/7=78.4 ph/w
/faculteit technologie management
Performance Analysis
/faculteit technologie management
Performance Analysis
B
NL
Explanation for the long
waiting times:
Cases arrives in batches
/faculteit technologie management
Use LTL checker to select cases
• eventually_activity_A=NLrestartRepair or
eventually_activity_B=BrestartRepair
– cases with a restart = 524
– cases without a restart 876
• eventually_person_P=Jan ... etc.
– Jan / Piet / Renate / Els = 78 / 163 / 83 / 153 
477/851 = 0.56 (851 is number of NL cases)
– Ties Sjef Lieve = 74 / 131 /194 = 399/549 = 0.73 (549
is the number of B cases)
/faculteit technologie management
It is always possible to perform mining/analysis on
selected cases. Example: mining and performance
analysis on the 194 directly correct repaired cases of
Lieve!
/faculteit technologie management
Many other performance indicators
• Performance Sequence Diagrams
• Doted Chart
• ...
/faculteit technologie management
Performance Sequence Diagrams
/faculteit technologie management
Doted Chart
/faculteit technologie management
How to get an event log
• Prom Import
/faculteit technologie management
Staffware
FLOWer
Websphere
YAWL
ADEPT
ARIS PPM/SIM
Outlook
Caramba
SAP
PeopleSoft
InConcert
IBM MQSeries
CPN Tools
CVS
Oracle BPEL
UML SD
company specific systems
...
Practical experiences
Rijkswaterstaat:
Loops to get
pay permission
Heusden city hall: errors in
workflow implemantation
/faculteit technologie management
CJIB
UWV
Rijkswaterstaat
ASML
AMC hospital
Catharina hospital
Eindhoven
Heusden
ING Bank
Philips medical
systems
...
Lessons learned
• Business Intelligence (BI) tools are NOT very
intelligent!
• Logs are everywhere!
• Process mining is possible and provides valuable
insights.
• Process mining triggers process improvement.
• Most processes do not conform.
• Reality is much more complicated than people like
to believe!
/faculteit technologie management
/faculteit technologie management
/faculteit technologie management
/faculteit technologie management
Challenges
• Data in SAP ERP systems into XML event-log
format
• Mining less structured data with many different
tasks and complex splits and joins and very large
(hospitals)
• Visualization of results (Process Oriented OLAPtools)
• On-line monitoring (process optimization,
prediction)
/faculteit technologie management
Challenges
• Measuring the quality of mined process models
• Development of Benchmark event-logs
• ...
/faculteit technologie management
/faculteit technologie management