Repairing Process Models to Match Reality
Description
Presentation on repairing process models to conform to observed process executions, given at BPM 2012 in Tallinn, Estonia, September 2012.
Transcript
Dirk Fahland Wil van der AalstRepairing Process Models Situation: model vs. reality process model Conformance Check Is this a good model for the real process? event log PAGE 1 Discovery vs. Repair original Whyrepaired processdiscovered not use discovery? PAGE 2 Problem Statement repaired process repair process model model Conformance Check Is this a good model for the real process? event log PAGE 3 Conformance process model fitness precision model can model behavior replay the log close to the log generalization simplicity event log model is as simple model allows more than as possible just the log PAGE 4 Here: focus on fitness process model fitness precision preserve during repair post-processing [Fahland, Aalst, BPM2011] generalization simplicity event log pre-process log (filter noise, etc.) PAGE 5 Problem Analysis repaired process repair process model model fitness replays all traces repaired discoveredreplays some traces original model distance to original model PAGE 6 Approach to Model Repair repaired process remove process model non-required model conformance deviations checker add missing event log PAGE 7 Approach to Model Repair repaired process remove process model non-required model conformance deviations checker add missing event log PAGE 8 Align Log and Model p2 p4 p1 Bmodel: A p3 p5 E p6 C DABFBDCE log PAGE 9 Align Log and Model p2 p4 p1 Bmodel: A p3 p5 E p6 C DABFBDCE log alignment PAGE 10 Align: Synchronous Move on A p2 p4 p1 Bmodel: A p3 p5 E p6 C D AABFBDCE A p2 p3 log alignment PAGE 11 Align: Synchronous Move on B p2 p4 p1 Bmodel: A p3 p5 E p6 C D A BABFBDCE A B p2 p4 p3 p3 log alignment PAGE 12 Align: Log Move on F p2 p4 p1 Bmodel: A p3 p5 E p6 C D A BABFBDCE A B F p2 p4 p3 p3 log alignment PAGE 13 Align: Log Move on B p2 p4 p1 Bmodel: A p3 p5 E p6 C D A BABFBDCE A B F B p2 p4 p3 p3 log alignment PAGE 14 Align: Model Move on C p2 p4 p1 Bmodel: A p3 p5 E p6 C D A B CABFBDCE A B F B p2 p4 p4 p3 p3 p5 log alignment PAGE 15 Align: Synchronous Move on D p2 p4 p1 Bmodel: A p3 p5 E p6 C D A B C DABFBDCE A B F B D p2 p4 p4 p4 p3 p3 p5 p3 log alignment PAGE 16 Align: Synchronous Move on C p2 p4 p1 Bmodel: A p3 p5 E p6 C D A B C D CABFBDCE A B F B D C p2 p4 p4 p4 p4 p3 p3 p5 p3 p5 log alignment PAGE 17 Align: Synchronous Move on E p2 p4 p1 Bmodel: A p3 p5 E p6 C D A B C D C EABFBDCE A B F B D E p2 p4 p4 p4 p4 p6 p3 p3 p5 p3 p5 log alignment PAGE 18 Complete Alignment p2 p4 p1 Bmodel: A p3 p5 E p6 C D A B C D C E firing sequenceABFBDCE A B F B D C E trace p2 p4 p4 p4 p4 p6 visited markings p3 p3 p5 p3 p5 of the model log alignment PAGE 19 Diagnostic Information FB sublog @ {p4,p3} need to replay FBwhen in marking {p4,p3} need to skip C to replay trace A B C D C E firing sequence A B F B D C E trace p2 p4 p4 p4 p4 p6 visited markings p3 p3 p5 p3 p5 of the model PAGE 20 Approach to Model Repair repaired process make optional / process model remove model activities that have to be skipped/removed conformance add checker subprocesses discover discover event log sublogs of events that cannot be replayed PAGE 21 Sublogs: Join by Events and Location p2 p4 p1 B A p3 p5 E p6 C D GH @ {p2,p3} FB @ {p4,p3}conformance checker GH @ {p4,p3} BF @ {p4,p5} sublogs @ locations PAGE 22 Sublogs: Join by Events and Location p2 p4 p1 B A p3 p5 E p6 C D GH FB @ {p3} @ {p4} GH BF sublogs @ locations PAGE 23 Sublog of Events Add Subprocess B p2 p4 F p1 B A p3 p5 E p6 C D GH FB @ {p3} @ {p4} GH BF PAGE 24 Sublog of Events Add Subprocess B p2 p4 F p1 B A p3 p5 E p6 C D H G GH FB @ {p3} @ {p4} GH BF PAGE 25 Events to Skip B p2 p4 F p1 B A p3 p5 E p6 C D H G allow to skip Cconformance checker PAGE 26 Events to Skip B p2 p4 F p1 B A p3 p5 E p6 C D H G allow to skip Cconformance checker PAGE 27 Approach to Model Repair repaired process make optional / process model remove model activities that have to be skipped/removed conformance add checker subprocesses discover discover event log sublogs of events that cannot be replayed PAGE 28 Implemented: ProM > Uma > Repair Model municipality process + log model moves: 3327 log moves: 310 deviations per trace: 2-49 PAGE 29 Implemented: ProM > Uma > Repair Model municipality process + log model moves: 3327 log moves: 310 deviations per trace: 2-49 PAGE 30 Implemented: ProM > Uma > Repair Model municipality process + log model moves: 3327 log moves: 310 deviations per trace: 2-49 PAGE 31 Implemented: ProM > Uma > Repair Model municipality process + log (filtered) model moves: 681 log moves: 229 deviations per trace: 1-12 PAGE 32 Implemented: ProM > Uma > Repair Model municipality process + log (filtered) model moves: 681 log moves: 229 deviations per trace: 1-12 PAGE 33 Implemented: ProM > Uma > Repair Model PAGE 34 Implemented: ProM > Uma > Repair Model PAGE 35 Case Study: Deviations vs. Similarity distance to original [Dijkman et al., BPM 2009] 1 reference model 1 10 logs .8 • 180-481 traces, up to 82 events per trace .6 • 2-49 deviations per trace rediscovered .4 + simplified results • distance to original < .2 .2 • better than rediscovery repaired • stable in # of deviations 0 2 4 6 8 10 12 avg. deviations per trace PAGE 36 Conclusion: new tool in the box PAGE 37 Conclusion effective technique for model repair • alignment sublogs of missing events subprocess future work: • quality of alignment quality of repair • allow to change ordering of tasks • repair for precision, generalization, … • many more … PAGE 38 Dirk Fahland Wil van der AalstRepairing Process Models Conformance 4 quality measures process model fitness precision “flower model can model” behavior replay the log not explained by log is small generalization simplicity event log model is as simple model allows more than as possible just the log PAGE 40 Conformance 4 quality measures process model fitness precision “each trace model can separately” behavior replay the log not explained by log is small generalization simplicity event log model is as simple model allows more than as possible just the log PAGE 41 Conformance 4 quality measures process model fitness “spaghetti” precision model can behavior replay the log not explained by log is small generalization simplicity event log model is as simple model allows more than as possible just the log PAGE 42
Leave a Comment
You must be logged in to post a comment.
