1. Model Transformations for Business Process Analysis and Execution Marlon Dumas University of Tartu

2. 2 Business Process Management (BPM) Collection of methods and techniques to design, analyze, execute and monitor business operations involving humans, software, information and physical artifacts using process models.

3. 3 OK, so what is a process model? Collection of inter-dependent activities whose collective performance is intended to achieve a goal such as delivering a product or a service. E,g. order-to-cash, procure-to-pay, issue-to-resolution Animated example

4. 4 Process models serve many purposes… Process Documentation Workflow Management Enterprise Architecture Document Management Knowledge Management Process Improvement Process Cost Analysis / Simulation Enterprise Systems Compliance / Risk Management Software Evaluation/ Selection

5. 5 … they have many faces Abstract Models EPC, BPMN Communication, simulation, activity- based costing… Detailed Models BPEL, State Machines… Data types, conditions, data mappings, fault handling… Integration, testing, deployment…

6. 6 …and can be seen from many perspectives Control-flow Data Resource Operational

7. 7 Many in one, or one in many? “The thing can be many in one sense, but also can be one in another sense.” Imam Ghazali Revival of Religious Sciences

8. 8 Process Modelling: Dealing with Multiplicity Multiple modelling languages (meta-models) Multiple modelling viewpoints –Control-flow view vs. data view –Public views (protocols) vs. private views Multiple abstraction levels –High-level: tasks, performance metrics... –Low-level: data transformations, application bindings...

9. 9 Process Modelling Notations Business Process Modelling Notation (BPMN) Event-driven Process Chains (EPC) Business Process Execution Language (BPEL) State machines and variants (e.g. IBM Business State Machine, WWF) Petri nets (and variants, e.g. YAWL)

10. 10 Model transformations BPMN-to-BPEL –Purpose: Transform models produced by analysts into models for developers (and vice-versa) –Commonly supported in commercial tool, but in a limited manner BPEL-to-Petri nets –For analysis & verification BPMN-to-Petri nets –For analysis & verification (e.g. deadlock-freeness)

11. 11 BPMN from 10 000 miles…

12. 12 BPMN: A more detailed view

13. 13 Quick BPMN example

14. 14 BPEL from 10 000 miles Basic activities:,, Sequential flow:,, Block-structured parallel flow: Graph-oriented (parallel) flow: ** Event-action rules: ** Other constructs not relevant to this talk… ** Only partially supported by some tools

15. 15 A quick BPEL example

16. 16 BPMN-to-BPEL: Some monsters!

17. 17 BPMN-to-BPEL (from 10 000 miles) Repeat until reduction to a single node 1.Identify a structured & quasi-structured SESE region  Fold into a BPEL structured activity 2.Identify irreducible SESE regions without parallelism  Apply Goto  While transformations, repeat from 1 3.Identify acyclic fragments with concurrency  Fold into a BPEL activity with control links 4.Identify minimal unstructured components  BPEL event handlers

18. 18 BPMN-BPEL: Structured Components

19. 19 BPMN-BPEL: Structured Components

20. 20 BPMN-BPEL: Structured Components

21. 21 Example: Only Structured Components

22. 22 BPMN-to-BPEL: Acyclic component Proposition: Every acyclic BPMN component with a single entry point and a single exit point that is 1-safe and sound can be mapped to a BPEL Flow with links

23. 23 Example Structured+Acyclic Components

24. 24 Minimizing the use of control links

25. 25 For the rest… Identify a minimal SESE region that is neither structured nor acyclic For each action in the component, retrieve: –All actions that immediately precede it –All actions that immediately follow it For each action, code the following behaviour using event-action rules: –Wait for a suitable combination of predecessors to complete –Perform action –Notify completion to all successors

26. 26 BPEL-to-Petri net Application to Conformance Checking Services abstract BPEL process Petri net model SOAP Messages Event Log Services Monitoring / Correlation Conformance? Conformance Checking! Translation

27. 27 From BPEL to WF-nets

28. 28 Petri net-based Conformance Checking Objectives: –quantitatively measure conformance –locate deviations Conformance?

29. 29 Conformance Checking – Fitness

30. 30 Conformance Checking – Fitness

31. 31 Measuring fitness: Log replay analysis missing tokens remaining tokens = 0 = 1 consumed tokens produced tokens = 0

32. 32 Measuring fitness: Log replay analysis = 1 = 2 = 0 missing tokens remaining tokens consumed tokens produced tokens

33. 33 Measuring fitness: Log replay analysis = 2 = 4 = 0 missing tokens remaining tokens consumed tokens produced tokens

34. 34 Measuring fitness: Log replay analysis = 3 = 5 = 0 missing tokens remaining tokens consumed tokens produced tokens

35. 35 Measuring fitness: Log replay analysis = 5 = 6 = 0 missing tokens remaining tokens consumed tokens produced tokens

36. 36 Measuring fitness: Log replay analysis = 6 = 7 = 0 missing tokens remaining tokens consumed tokens produced tokens

37. 37 Measuring fitness: Log replay analysis = 7 = 0 missing tokens remaining tokens consumed tokens produced tokens

38. 38 Measuring fitness: Log replay analysis = 9 = 0 missing tokens remaining tokens consumed tokens produced tokens

39. 39 Measuring fitness: Log replay analysis = 9 = 0 missing tokens remaining tokens consumed tokens produced tokens

40. 40 Measuring fitness: Log replay analysis = 0 = 1 = 0 missing tokens remaining tokens consumed tokens produced tokens

41. 41 Measuring fitness: Log replay analysis = 1 = 2 = 0 missing tokens remaining tokens consumed tokens produced tokens

42. 42 Measuring fitness: Log replay analysis = 2 = 4 = 1 = 0 missing tokens remaining tokens consumed tokens produced tokens

43. 43 Measuring fitness: Log replay analysis = 3 = 5 = 1 = 0 missing tokens remaining tokens consumed tokens produced tokens

44. 44 Measuring fitness: Log replay analysis = 4 = 6 = 1 = 0 missing tokens remaining tokens consumed tokens produced tokens

45. 45 Measuring fitness: Log replay analysis = 6 = 7 = 1 = 0 missing tokens remaining tokens consumed tokens produced tokens

46. 46 Measuring fitness: Log replay analysis = 7 = 8 = 1 = 0 missing tokens remaining tokens consumed tokens produced tokens

47. 47 Measuring fitness: Log replay analysis = 8 = 1 missing tokens remaining tokens consumed tokens produced tokens

48. 48 Measuring fitness: Log replay analysis = 8 = 1 missing tokens remaining tokens consumed tokens produced tokens

49. 49 Measuring fitness: Log replay analysis

50. 50 Measuring fitness: Log replay analysis f = 1.0f ≈ 0.540f ≈ 0.955

51. 51 Conformance Checking - Appropriateness 100 % fitness but not sufficiently specific from behavioral point of view. 100 % fitness but not represented in structurally suitable way.

52. 52 Ongoing work Reversible BPMN ↔ BPEL transformation Two-way transformation: BPEL ↔ FSM –Application to automated service composition –BPEL  FSM not too difficult –FSM  BPEL more exciting, e.g.

53. 53 References