Title: Preprocessing support for large scale process mining of SAP transactions
1Preprocessing support for large scale process
mining of SAP transactions
- Jon Espen Ingvaldsen
- Jon Atle Gulla
2Overview
- Motivation
- Process mining SAP data
- EVS ModelBuilder
- Preprocessing steps
- Algorithm for constructing process instances
- Mapping to MXML
- Discussion Experiences
3Process mining project
4Process mining SAP data
- Documents that refer to each other
- Example
- Manual document operations are done by use of
transactions - ME21 Create Purchase Order
- VA02 Change Sales Order
- All document operations are logged with user
reference and timestamp
OC
PO
PR
5Process mining SAP data
- n to n mapping between business processes
transaction codes - n to n mapping between transaction codes and
tasks - Large amounts of data
- Multiple of database tables involved
6EVS ModelBuilder
- Developed as a set of plugins to Eclipse
- Tested and applied at Nidar AS to investigate
their purchase agreement processes
7Preprocessing process
- Define a process mining project
- Define access properties for involved database
sources - Define event types
- Define business object types
- Define relationships between event types and
business object types
8How do we define event types?
- Type name
- Key attributes
- Ensure unique identification
- Timestamp attribute and format
- Tells when the event occurred and how to parse
the timestamp values - Relationships to BO types
- Tells which BOs that were involved in the event
- Discriminators
9Example of an event description
- Type name
- ALTER PURCHASE ORDER
- Key attributes
- CDHDR.CHANGENR
- Timestamp attributes
- CDHDR.UDATE (ddMMyyyy) and CDHDR.TIME (hhmmss)
- Discriminators
- CDHDR.OBJECTCLAS EINKBELEG
10Examples of constructed events
Type ALTER PURCHASE ORDER Key CDHDR.CHANGENR
00021 Timestamp 1. June 2007, 140123
Type ALTER PURCHASE ORDER Key CDHDR.CHANGENR
00024 Timestamp 4. June 2007, 115204
11How do we define Business Object types?
- Type name
- Key attributes
- Ensure unique identification
- Value attribute
- Provides a textual description of the BO
- Analysis features
- Information that describe the BO
- Relationships to other BO types
- Discriminators
12Examples of Business Object descriptions
- Type name USER
- Key attributes USR01.BNAME
- Value attribute USR01.NAME1
- Analysis features (NONE)
- Discriminators (NONE)
13Examples of constructed business objects
Type USER Key USR01.BNAME JACKSON Value
Michael Jackson
Type USER Key USR01.BNAME SIMPS Value Homer
Simpson
14How do we define relations?
- Relations consist of
- Two end points
- Textual description
- Mapping attributes / path
15Example of a relation
Type USER Key USR01.BNAME JACKSON Value
Michael Jackson
Relation name Has user /
Executes Mapping attributes
CDHDR.USERNAME USR01.BNAME
Type ALTER PURCHASE ORDER Key CDHDR.CHANGENR
00021 Timestamp 1. June 2007, 140123
16Example of an event and its relations
Type DEPARTMENT Key USR01.ABTLG Los
Angeles Value Los Angeles
Type USER Key USR01.BNAME JACKSON Value
Michael Jackson
Type ALTER PURCHASE ORDER Key CDHDR.CHANGENR
00021 Timestamp 1. June 2007, 140123
Type PURCHASE REQUISITION Key EBAN.BANFN
007866 Value 007866
Type TRANSACTION Key TSTC.TCODE ME21N Value
Create Purchase Order
Type PURCHASE ORDER Key EKKO.EBELN
002121004 Value 002121004
17Construction of process instances
18Process instances
19MXML
- The input format for ProM
- Requirements for events (AuditTrailEntries) in
MXML - Reference to an activity
- Reference to a process instance
- Reference to a performer
- Labeled with a timestamp
- Ordered on timestamp within the process instance
20Mapping to MXML
To construct AuditTrailEntries inMXML we need to
Type USER Key USR01.BNAME JACKSON Value
Michael Jackson
Selects the BO type that represents the
orginators
Selects the BO type that represents the workflow
element
Type ALTER PURCHASE ORDER Key CDHDR.CHANGENR
00021 Timestamp 1. June 2007, 140123
Make use of the event timestamp
Type TRANSACTION Key TSTC.TCODE ME21N Value
Create Purchase Order
21Discussion Experiences
- Still a effort is required!
- Preprocessing support have most value
- When mining longer process chains
- A lot of database tables are involved
- Stores data source descriptions, not the process
of transforming the data - Ensures reusability across process mining
projects. - Enable process mining of processes that spans
multiple information systems
22Thank you!