Business Rules and Agent-Based Business Process Simulation

Gerd Wagner

Brandenburg University of Technology at Cottbus, Germany

This Talk Is about Three Topics

1. Business Processes
2. Agent-Based Discrete Event Simulation
3. Business Rules and Simulation (work in progress)

Business Processes

What Is a Business Process? (Classical Def.)

• A specific ordering of work activities across time and space, with a beginning and an end, and clearly defined inputs and outputs (Davenport 1993)
• A collection of activities that takes one or more kinds of input and creates an output that is of value to the customer (Hammer & Champy 1993)

What's Wrong with These Definitions?

• A specific ordering of work activities across time and space, with a beginning and an end, and clearly defined inputs and outputs (Davenport 1993)
• A collection of activities that takes one or more kinds of input and creates an output that is of value to the customer (Hammer & Champy 1993)

• Too incomplete:
  • No distinction between process (instance) and process type
  • ordering is not sufficient for defining a business process type, rather one needs AND/XOR/OR branching and merging
• Too vague:
  • What are inputs and outputs?
  • What is the value to the customer?
• Too specific:
  • Do all business process types specify an end?
  • Do all business processes have a customer?

What is a Business Process? (My Definition)

• business process: social interaction process that occurs in the context of a business system (organization) and serves a purpose of that system. Examples: a football game; a Web shop purchase
• social interaction process: interaction process that includes at least one communicative action event. Examples: a conversation; a birthday party; the Second World War; a football game; a Web shop purchase.
• interaction process: process that includes at least one perception event and one action event perceived and performed by agents that participate in it. Examples: someone turning on the light in the office when it becomes dark outside; a conversation; a birthday party; the Second World War; a football game; a Web shop purchase.

What is a Process?

• process: complex event that is composed of other events by means of event composition operators.
• Examples:
  • a parallel occurrence of two explosions;
  • a heart attack;
  • a conversation;
  • a birthday party;
  • the Second World War;
  • a football game;
  • a Web shop purchase

Observations

• There are many different kinds of processes: physical, chemical, biological, social, ...
• All processes are based on events
• Business processes are based on agents and their perception events (including incoming messages) and action events (including outgoing messages)

What is BP Simulation Good For?

• For analyzing business process models with respect to performance metrics at design time
• For evaluating short-term options how to adjust a running business process in response to unforeseen circumstances (operational decision support at run time)

Agent-Based
Discrete Event Simulation

Discrete Event Systems

A discrete event system is a dynamic system consisting of objects (of various types) whose states may be changed by events (of various types) occurring at successive points in time (from a a discrete set).

For modeling such a system:

1. Describe its object types
2. Describe its event types
3. Specify the state changes of objects when events occur (which objects are affected in which way by the occurrence of an event).

Example: A Drive-Thru Restaurant

• As a car enters from the street, the driver, who we will call Fred, decides whether or not to get in line
• If Fred decides to leave the restaurant, he leaves as a dissatisfied customer.
• If Fred decides to get in line, then he waits until the menu board is available.
• At that time, Fred gives the order to the order taker. After the order is taken, then two things occur simultaneously:
  1. Fred moves forward if there is room, otherwise he has to wait at the menu board until there is room to move forward.
  2. The order is sent electronically back to the kitchen where it is prepared as soon as the cook is available.
• As soon as Fred reaches the pickup window, then he pays and picks up his food, if it is ready.
• If the food is not ready, then Fred has to wait until his order is prepared.

A Drive-Thru Restaurant as a basic DES

Possible object types to model:

• Customer, Order
• MenuBoard, MenuBoardWaitingLine
• PickupWindow, PickupWindowWaitingLine

Possible event types to model:

• CustomerArrival
• StartMenuBoardService, EndMenuBoardService, EndKitchenService
• StartPickupWindowService, EndPickupWindowService

Multi-Agent Systems (MAS)

A multi-agent system is a dynamic system where special objects, called 'agents', interact

• with each other via communication
• with their environment via perception and (state-changing) actions

Examples:

• All kinds of systems the dynamics of which is based on human actions: universities, hospitals, manufacturing plants, economic systems, ...
• Eco-systems
• Fish swarms, ant colonies, ...

A Drive-Thru Restaurant as a MAS (1)

Possible (non-agentive) object types to model:
MenuBoardWaitingLine, PickupWindowWaitingLine

Possible agent types to model:
Customer, MenuBoard, PickupWindow

Possible message types to model:
Order, ConfirmOrder, Bill

A Drive-Thru Restaurant as a MAS (2)

Possible action event types to model:
Order-Out-MessageEvent, Bill-Out-MessageEvent, PaymentActionEvent

Possible perception event types to model:
Order-In-MessageEvent, Bill-In-MessageEvent PaymentPerceptionEvent

Application Areas for Agent-Based DES

1. Management science:
   enterprise models, managment games, ...
2. Economics:
   market models (stock market, labor market, etc.)
3. Social science:
   evolution of language, norms, ...
4. Engineering:
   socio-technical systems such as material handling systems (AGVs) and traffic systems
5. Biology:
   population dynamics, eco-systems, evolution, ...

AOR Modeling and Simulation

• Agent Object Relationship Modeling is a refinement of
  Entity Relationship Modeling
• An entity may be a passive object or an interactive agent (or a message or an event)
• AOR Simulation is an agent-based discrete event simulation approach
• Visit the project website: www.AOR-Simulation.org

The AOR Simulation Language (AORSL)

• Is based on the AOR Simulation metamodel, a UML/MOF model defining its abstract syntax
• Has an XML syntax, which includes opaque expressions (e.g. Java expressions)
• Is using rules for modeling causality and agent behavior

Simulation Scenarios and Simulation Models

• A simulation scenario consists of a
  1. simulation model,
  2. an initial state definition, and
  3. zero or more view definitions (for visualization).
• A simulation model consists of:
  1. an optional space model (needed for physical objects/agents);
  2. a set of entity type definitions, including different categories of event, message, object and agent types; and
  3. a set of environment rules, which define causality laws governing the environment's state changes and the causation of events.

The AOR Simulator Architecture

An AOR simulator consists of

• an environment simulator, which creates perception events and passes them to the agent simulators concerned
• zero or more agent simulators, which process any perception events received and, in response, create action events that are passed back to the environment simulator

Running an AOR Simulation Scenario

1. Validate your AORSL file
2. Generate Java code
3. Compile the Java code
4. Run the resulting Java program
5. Evaluate the statistics and/or the simulation log

Basic DES with AORSL

AORSL supports basic DES without agents by defining

• Object types for abstract objects in time
• Physical object types for objects in space and time
• Exogenous event types with some random occurrence periodicity
• Caused event types for events that are caused by other events

A Basic DES Model for a Drive-Thru Restaurant

• Four object types: Customer, Order, MenuBoard, PickupWindow
• One exogenous event type: CustomerArrival
• Five caused event types: StartMenuBoardService, EndMenuBoardService, EndKitchenService, StartPickupWindowService, EndPickupWindowService
• Nine simulation rules: CustomerArrivalWhenQueueFull_Rule, CustomerArrivalWhenQueueNotFull_Rule, StartMenuBoardService_Rule, EndMenuBoardServiceWhenPickupWindowQueueNotFull_Rule, ...

Rules (1)

An environment rule is a 5-tuple < EvtExpr, Var, Cond, UpdExpr, ResEvt > where

• EvtExpr is an event expression specifying the type of event that triggers the rule
• Var is a set of variable declarations, such that each variable is bound either to a specific object or to a set of objects
• Cond is a logical condition formula possibly containing variables
• UpdExpr is an expression specifying an update of the state
• ResEvt is a list of resulting events, which will be created when the rule is fired

Rules (2)

A rule < EvtExpr, Var, Cond, UpdExpr, ResEvt > can also be stated in the following form:

ON EvtExpr
FOR Var
IF Cond
THEN
   UPDATE UpdExpr
   SCHEDULE-EVENT ResEvt

Operational Semantics

A simulation step s = 1, 2, … of a DES system with state < S, FEvt, s > and environment rule set R consists of three sub-steps:

1. Determine the set of current events:
   FEvt_s = { e ∈ FEvt | occTime(e) = s }
2. Apply the rules triggered by the current events, resulting in an updated state S' and an updated future events list FEvt'
3. Increment the simulation step (or time) counter

Agent-Based DES with AORSL

For an agent-based model we have to define

• (Possibly physical) agent types
• Message types
• Perception event types
• Action event types

Why should We Want to Make Agent-Based Models?

• Agent-based models are (much) more complex
• But:
  • They allow to capture more of the structure of a business process (e.g. BPMN "pools"), leading to closer-to-reality models
  • They allow modeling cognitive and social phenomena

Agent Types

An agent type is defined by

1. a set of (objective) properties;
2. a set of (subjective) self-belief properties;
3. a set of (subjective) belief entity types; and
4. a set of agent rules, which define the agent's reactive behavior in response to events

Agent-Related Events

Agent Types

For a cognitive agent type we may specify:

• Belief entity types (defined by means of belief properties) for expressing an agent's beliefs about entities in its environment
• Self belief properties for expressing an agent's beliefs about itself

Business Rules

SBVR

• The Semantics of Business Vocabularies and Rules (SBVR) is an OMG standard for ("computation-independent") business modeling (as of January 2009)
• It allows to define business vocabularies and rules in controlled English (also called "structured English"), corresponding to formal logic expressions
• Why is SBVR based on controlled English?
• Because business people don't want to use formal logic, and neither do they want to use graphical languages such as UML, for expressing their business terms and rules

SBVR's Strengths

• SBVR supports the formal capture of vocabularies and rules in use within an organization
• SBVR supports the maintenance of multiple vocabularies and their modular composition
• SBVR allows the violation of certain rules (this is part of their semantics)

SBVR's Weaknesses

• There is no methodology how to develop "good" SBVR models
• There is no support for identifying those rules that can be automated in an information system
• After capturing business rules with SBVR (leading to some kind of formalization), it is often still difficult to find proper operationalizations for the rules

Implementing Business Rules

• In general, SBVR rules correspond to various kinds of integrity constraints that govern the operation of a business system
• Some of the rules captured with SBVR can be implemented in an information system, e.g. in the form of
  • user interface restrictions
  • SQL checks/assertions
  • special rule software (using a rule engine such as JBoss Rules or IBM/ILOG)
• For other rules it's not clear if and how they could be implemented with the help of IT

Business Rules and Simulation

Simulations can be used for

• Validation: does a particular implementation of a rule lead to the expected behavior of the system?
• Investigating rule violations:
  • Under which conditions do human actors violate a rule?
  • What are the consequences of rule violations?

Conclusions

• Agent-based models are ontologically more faithful ("closer-to-reality")
• Agent-based modeling and simulation may allow a deeper understanding of business systems, business rules and business processes
• The price to pay (the higher complexity of models) may be unavoidable for obtaining better models of complex systems