Planning and Scheduling for Dynamic Workflow Management

1
Planning and Scheduling for Dynamic Workflow
Management Dr Nikolay Mehandjiev Department of
Computation, UMIST, Manchester, U.K. E-mail
mehandjiev_at_acm.org WWW www.co.umist.ac.uk/nd
m

• Aim Promoting the effective application of AI
  PS techniques to WM
• Roadmap
• similarities between PS and WM
• areas where PS can provide substantial input
  to dynamic WM
• Some related projects

2
Motivation

• Workflow Management
• Growing need for
• Business Process Modelling
• Supply Chain Execution
• Rapidly changing processes
• Static tools incapable of dynamic optimisation
• AI Planning and Scheduling
• Applied to a range of real work problems
• Mostly linked with materials and production
• Potential for Workflow Management

3
A simple workflow process

1. Customer selects laptop
2. Customer sends specs to purchasing
3. Purchasing requests quote from supplier
4. Supplier sends quote to purchasing
5. Purchasing confirm quote with customer
6. Purchasing send purchase order
7. Supplier sends laptop to customer
8. Supplier sends invoice to purchasing
9. Purchasing send cheque to supplier

4
Workflow Management
5
Mapping similarities
6
Highlighting differences
Terminology Design-time work WM humans assisted
by simple tools PS mainly automatic by
software Representations WM domain-oriented and
user-friendly but vague PS mathematically
formal and semantically precise Languages WM
scripting languages for coordination of
activities PS AI plan languages are higher level
7
Requirements
Short-term Integration of scheduling and resource
allocation in WM tools Re-planning Generating
workflow definitions from process models Process
mining Medium-term WM support for skilled
work User empowerment Visualising
work Long-term Flexibility Evolvability and
Adaptiveness Decentralised management
8
Roadmap themes
9
Human factors
State-of-art WM tailorable workflow systems PS
mixed-initiative planning Research issues User
empowerment versus centralised control Balance
human and software effort at different
stages Visualising the way planners
work Recommended actions Trans-disciplinary
workshops Prototype supporting user control
10
Infrastructure
State-of-art WM WfMC reference architecture PS
PDDL and ADL Research goals From objects through
components to agents Developing planning and
scheduling servers Recommended actions Draw up a
reference architecture Interface standards
11
Domain and business modelling

• State-of-art
• Process management ARIS, iThink, IDEF, PIF, PSL,
  WPDL
• Knowledge engineering CommonKADS, CoRE
• Ontologies CYC, Enterprise, TOVE, DAML-OIL/OWL
• AI planning ADL, PDDL, STRIPS/PDDL, GIPO
• Research goals
• How to synthesise BPM, PS and ontology modelling
  languages into a language which is
• Useable by domain experts
• Has rigorous semantics and mathematicall formal
• Executable
• Translateable into other formalisms
• Suitable for planning
• Recommended actions
• Taxonomy of languages and tools
• Organise a hands-on workshop on modelling

12
Planning and scheduling

• State-of-art
• Many planning techniques available, current
  trends to
• integrated planning and scheduling
• mixed initiative planning
• constraint-based approaches
• Some tools to support these, eg ILOG Scheduler,
  but expert-oriented
• Successful industrial applications but not in WM
• Research goals
• How to best combine human capabilities with PS?
• Create interfaces so that planners are easily
  used by domain experts.
• Recommended actions
• Definition of graduated reference problems

13
Enactment / execution
State-of-art Conditional planning Reactive
planning Research goals flexible working with
overall plan use of plan repair or re-planning
techniques for exception handling combination of
plans for multiple actors Requirements Techniques
for monitoring execution Techniques for exception
handling
14
Adaptation, optimisation metrics
State-of-art Mature optimisation techniques
coming from AI and OR Using multiple criteria for
planning and scheduling Research
goals Appropriate metrics for WM time and cost
only? Languages for providing metrics to the
system Combination of metrics Requirements User-de
finable metrics and optimisation
parameters Integration with process design and
enhancement tools Interaction with the user
15
Main recommendations

• To raise awareness of real challenges and
  constraints in workflow domain
• To make application and tool developers aware of
  what AI planning and scheduling research has to
  offer
• To address practical issues of integrating
  planning and scheduling technology into suites of
  application software, and of making the
  techniques usable by typical software engineers,
  analysts, etc.
• To form a consensus on medium and long term
  research goals.
• The RoadMap should be seen as a living document
  and be extended and updated regularly.

16
Related work at UMIST

1. Tailorable workflow systems and EUD - ECHOES
2. Allocating tasks in distributed teams SAMBA, BT
   Exact
3. Semi-automatic allocation of tasks to agents -
   RAMASD
4. Agent-based workflow support systems
   Agentcities IntLog
5. Supporting the automatic formation of Virtual
   Enterprises

17
ECHOES

• Achievements
• Users can change workflow during execution under
  an intelligent guardian
• Highly responsive mode of interaction
• Multi-aspect visual language that can evolve
• Architecture to support all this

18
SAMBA

• BT Short Term Fellowship (1998)
• Aim Provide work coordination software that
• Allows workers to control their work
• Allows managers to modify local business rules
• Informates people about the work process
• Supports team building and social interaction
  processes

19
RAMASD

• Funded by BT Exact as a PhD studentship (June
  1999 June 2002)
• Allocate behaviour to agents using
• Roles
• Role models
• Role algebra
• Method
• Use synthesis-based design process
• Allocated by constraint satisfaction algorithm
• Incorporated in the agent-building toolkit Zeuss
• Application
• Currently for the design of agent systems
• Can be used to allocate tasks to performers

20
Agentcities.ORG network snapshot taken from
http//www.agentcities.org/Network/
21
IntLog
Co-optimise production and logistics services.
Agents allocate work using extensions to
Contract Net protocol. An Agentcities prototype
related to EC-funded project MaBE
(www.mabe-project.com)
22
Summary

• Workflow Management TCU produced a Roadmap on the
  use of AI PS techniques to Workflow Management.
• This presentation summarised findings of the
  Roadmap regarding
• Similarities and differences between the two
  areas
• Open research issues and actions in the area of
  PS application to Workflow Management.
• Requirements for applying PS techniques to
  Workflow Management.
• Several projects were then described briefly as
  an example of activities relevant to the Roadmap.
• For further information regarding this
  presentation, please e-mail me on
  mehandjiev_at_acm.org