Available online at www.sciencedirect.

com
Available online at www.sciencedirect.com
ScienceDirect
ScienceDirect
Procedia Manufacturing
Available 00 (2018) 000–000
Availableonline
onlineatatwww.sciencedirect.com
www.sciencedirect.com
Procedia Manufacturing 00 (2018) 000–000 www.elsevier.com/locate/procedia
www.elsevier.com/locate/procedia
ScienceDirect
ScienceDirect
Procedia Manufacturing 26 (2018) 1242–1254
Procedia Manufacturing 00 (2017) 000–000
www.elsevier.com/locate/procedia
46th SME North American Manufacturing Research Conference, NAMRC 46, Texas, USA
46th SME North American Manufacturing Research Conference, NAMRC 46, Texas, USA
IMPlanner-MAS: A Multiagent System for Distributed
IMPlanner-MAS: A Multiagent System for Distributed
Manufacturing
Manufacturing Engineering Process
Society International Planning
Conference 2017, MESIC 2017, 28-30 June
Manufacturing Process Planning
2017, Vigo (Pontevedra), Spain
Dusan N. Sormazaa* Arkopaul Sarkaraa, Subhabrata Ghosalaa
Dusan N. Sormaz * Arkopaul Sarkar , Subhabrata Ghosal
Costing models for capacity
Ohio University,
a
optimization
Department of Industrial in Industry
and Syserms Engineering, Athens, OH 45701,4.0:
USA Trade-off
Ohio University, Department of Industrial and Syserms Engineering, Athens, OH 45701, USA
a

between used capacity and operational efficiency

* Corresponding author. Tel.: +17405931545; fax: +17405930778.
E-mail address:author.
* Corresponding Tel.: +17405931545; fax: a+17405930778.
[email protected]
E-mail address: [email protected] A. Santana , P. Afonsoa,*, A. Zaninb, R. Wernkeb
a
University of Minho, 4800-058 Guimarães, Portugal
Abstract b
Unochapecó, 89809-000 Chapecó, SC, Brazil
Abstract
The paper reports development of the multiagent-based framework for distributed manufacturing process planning. The research
The paper
builds reports development
on previous of the multiagent-based
results in developing framework planning
intelligent manufacturing for distributed manufacturing
platform and extendsprocess
it into planning.
agent-basedThemodeling.
research
builds on previous
Both, design
Abstract results in developing
and manufacturing tasks are intelligent
defined as manufacturing
sequences of steps planning
to be platform
performedandby extends it into agent-based
highly intelligent autonomousmodeling.
agents.
Both, design and
Autonomous manufacturing
agents are classifiedtasks
into are
twodefined as sequences
categories: task agents, of steps
whichtoare
be responsible
performed by forhighly intelligent
completing designautonomous agents.
or manufacturing
Autonomous
tasks, and agents
service are
agents,classified
which into
provide two categories:
services for task
individualagents,
steps which
of are
those responsible
tasks. For for completing
multiagent
Under the concept of "Industry 4.0", production processes will be pushed to be increasingly interconnected, design
implementation, or manufacturing
we have chosen
tasks, and
JADE service agents,
framework and which
both taskprovide
agents services
and for individual
service agents steps
have of those
been tasks. Forby
implemented multiagent
extending implementation,
agents from we have
JADE. chosen
Thus, the
information based on a real time basis and, necessarily, much more efficient. In this context, capacity optimization
JADE
JADE’sframework and both
message passing task agents
mechanism hasand
beenservice
adopted agents have beencommunication
for inter-agent implemented byand extending
appropriateagents from JADE.
ontologies for theThus, the
message
goes
JADE’s
beyond
message
the traditional aim of capacity maximization, contributing also forand
organization’s profitability and value.
passing have beenpassing mechanism
built from planninghas beenand
classes adopted for The
objects. inter-agent communication
brief overview of IMPlanner appropriate ontologies
functionalities forprovided
has been the message
and
Indeed,
passing leanbeen
have management and continuous
built from planning classes agents improvement
and objects. The brief approaches
overview suggest capacity
of IMPlanner optimization
functionalities instead andof
development and implementation of service for several typical manufacturing planning tasks havehas been
been provided
explained. The
maximization.
development The study of
and implementation
proposed framework capacity
has been executed optimization
of service
in theagents
contextfor and
of severalcosting models
typical model
the simulation is an
manufacturing important
planning
for evaluation research
tasks have
of dynamic topic that deserves
beenpolicies.
routing explained. The
contributions
proposed frameworkfrom has
bothbeen
the executed
practicalinand theoretical
the context of theperspectives.
simulation model Thisfor
paper presents
evaluation and discusses
of dynamic routing apolicies.
mathematical
© 2018 for
model Thecapacity
Authors. management
Published by Elsevier
based on B.V. different costing models (ABC and TDABC). A generic model has been
© 2018 The Authors. Published by Elsevier B.V. B.V.
developed
Peer-review and it was used to
under responsibility analyze
responsibility of ofthe idle
thescientificcapacity
scientificcommittee and to
committee design strategies towards the maximization of organization’s
Peer-review under ofofthe
NAMRI/SME.
46th SME North American Manufacturing Research Conference.
value. The under
Peer-review trade-off capacity ofmaximization
responsibility vs operational
the scientific committee efficiency is highlighted and it is shown that capacity
of NAMRI/SME.
optimization might hide
Keywords: Manufacturing operational
process inefficiency.
planning, CAPP, Agent based modeling, Distributed planning, JADE framework
Keywords:
© 2017 TheManufacturing process planning,
Authors. Published CAPP,
by Elsevier Agent based modeling, Distributed planning, JADE framework
B.V.
Peer-review under responsibility of the scientific committee of the Manufacturing Engineering Society International Conference
1. Introduction
2017. and execute a task as a whole entity. There are various
1. Introduction and execute awhich
parameters task as acontrol
whole entity. There are various
the efficiency of a
Keywords: Cost Models;
A distributed ABC; TDABC;
system Capacity Management;
is a collection of devices parameters
Idle Capacity;distributed system like network traffic, latencyofetc.a
which
Operational Efficiency control the efficiency
A distributed
operating systemlocations
from remote is a collection of devices
autonomously and distributed system like
Also the behaviors network
of the traffic,
entities of alatency etc.
distributed
operating
connected fromby remote
a network locations autonomously
having a distributedand Also the behaviors of the entities of a
system majorly affect the performance of the system. distributed
1. Introduction
connected
middleware by a helps
which network
them to having a distributed
keep record of events system majorly
This brings us toaffect the performance
the concept of a frameworkof theofsystem.
mobile
middleware which helps them to keep record of events This brings us to the concept of a framework of mobile
The cost of idle capacity is a fundamental information for companies and their management of extreme importance
in modern©production
2351-9789 systems.
2018 The Authors. In general,
Published it isB.V.
by Elsevier defined as unused capacity or production potential and can be measured
in several©under
2351-9789
Peer-review ways: tons of production,
2018responsibility
The Authors. Published
of available
by Elsevier
the scientific B.V.hours
committee of manufacturing, etc. The management of the idle capacity
of NAMRI/SME.
Peer-review under Tel.:
* Paulo Afonso. responsibility
+351 253of the761;
510 scientific committee
fax: +351 253 604of741
NAMRI/SME.
E-mail address: [email protected]

2351-9789 © 2017 The Authors. Published by Elsevier B.V.

Peer-review under responsibility of the scientific committee of the Manufacturing Engineering Society International Conference 2017.
2351-9789 © 2018 The Authors. Published by Elsevier B.V.
Peer-review under responsibility of the scientific committee of the 46th SME North American Manufacturing Research Conference.
10.1016/j.promfg.2018.07.161
 Dusan N. Sormaz et al. / Procedia Manufacturing 26 (2018) 1242–1254 1243
Sormaz, Sarkar, Ghosal/ Procedia Manufacturing 00 (2018) 000–000 2

multi-agents which can very effectively create a interact among each other to configure the shop floor
distributed network with reduced network traffic, and take intelligent manufacturing decisions. Shen and
latency etc. The autonomous behavior of the agents Norrie, in their paper, “Agent-Based System for
and their independent existence is useful to construct Intelligent Manufacturing” [4] discussed vital issues
a stronger and fault tolerant distributed environment. towards implementation of multi-agent technology for
Distributed Processing in general happens through developing manufacturing systems like supply chain
various communication protocols in which multiple management, distributed dynamic scheduling etc.
communications are involved which increases the data Application of Agent-Based system in Process
volume in the network and can jam it. However, for a Planning has been demonstrated [5] where a dynamic
multi agent framework, the entire agent can relocate in process planning system has been developed to
the environment which saves multiple facilitate dynamic changes and automatic update of
communications and decreases chances of jam. process plans. Sormaz in his previous works have
Further, as the mobile agents are able to relocate, the concentrated on distributed modeling and interactive
processes which earlier used to happen through process planning using IMPlanner [6] for efficient
network can now be achieved in a remote system, configuring and customization of products. In this
which saves a lot of Network bandwidth. This paper is paper he has designed integrative model for the
organized in the following way. In section 2, we have product design using distributed data processing
studied the earlier works on using a multi agent between different hierarchical object models to find
framework for distributed processing. Section 3 out the most efficient one for product designing. In this
described the architecture of the distributed agent paper he also focused on service oriented agent
based processing. The brief overview of the organization, the concept of which will be later
manufacturing planning system IMPlanner is given in proliferated in this paper. This framework was built
Section 4. Section 5 contains the details of IMPlanner- using a tool called space searcher which uses a search
MAS implementation and Section 6 explains the case algorithm to obtain context specific optimal process
study. The paper end with Conclusions in Section 6 sequence. Further he worked on the development of a
and the list of references. distributed multi agent framework for design and
manufacturing integration [7] where they introduced
2. Previous Work concept of autonomous agents in the form of task
agents and service agents performing design and
The evolution of Agent Based systems [1] can be manufacturing tasks.
traced back from the origin of MAS or Multi Agent
system which is a product of Distributed Artificial 3. Agent Based Distributed Processing
Intelligence. The need to implement mobile agents in Architecture
distributed computing can be observed from the very
beginning which encouraged more research work on When we are trying to set up an effective
this domain and various intelligent agent frameworks distributed network, we look into creating a system
like DAML, Json, OWL, JAT etc. have been tested which has resource sharing, openness, concurrency,
since then. The utility of a multi-agent framework in fault tolerance and transparency. The choice of Agents
distributed or parallel processing was explained very for distributed systems is guided by some very specific
lucidly by Danny Lange in [2]. He explained that the reasons. The autonomy of agents and their
major reasons behind using mobile agents for asynchronous way of communication has proven to be
distributed networks is the amount of autonomy and more robust than the standard master slave
the asynchronous activity of the agents along with the relationship model. If a task requires continuous open
low network usage required for this framework. One connection then it is very difficult for a fragile network
of the earlier researches in the field of distributed system to carry on the objective. However, for agent
processing to automate the design and manufacturing based system, the tasks can be embedded into mobile
process involves utilizing a multi-agent system to agents which can be transferred in the network thus
build a holonic manufacturing system [3], which is an being independent of the origin and can work
intelligent shop floor management technology. In this asynchronously and independently. In that case the
model, intelligent mobile agents called “holons” agents have the right to accept or reject any task
1244 Dusan N. Sormaz et al. / Procedia Manufacturing 26 (2018) 1242–1254
Sormaz, Sarkar, Ghosal/ Procedia Manufacturing 00 (2018) 000–000 3

forming an executive dependency relationship where it is going to request service from, arranged in
a task can be rejected based on some definite sequence of services. The service agents on the other
unavoidable factor. Agents can also follow obligation hand also contain contents which will interact with the
dependency relationship where they will not be able to content coming from task agents for the execution of
refuse any task but they can delay the tasks or schedule services. The service agents maintain a history of the
the tasks according to its own rule. Agents need not services already performed and the present tasks
always be mobile. The agents which are providing which are in queue. The task agent’s lifecycle starts
service and not seeking service from other agents can with the “uninitiated” state after which it is born, then
communicate with the surroundings through starts searching for various services and once it finds
conventional means. the Service agent it goes to the “in-transit” state, then
In this educational tool for agent based processing, after entering the queue it goes in the waiting state.
we have shown both types of agents. The agents Finally, it goes in the “in-service” state when the
providing various services in a scattered distributed Service agent starts working. These cycle continues
network are autonomous static agents. They are called till all the services are completed after which the task
Service Agents. The services are requested from these Agent obtains the “complete” state. The Service
service agents by Task Agents which are remotely agents on the other hand just maintain two states,
located and they automatically find the service agents “idle” after finishing a service and “busy” when it is
as required. The service agents are in obligation performing.
dependency relationship with the Task Agents where The described agent based programming model can
they receive the various service requests from the task be implemented in different frameworks. The authors
agents and synchronize them according to their own have implemented a small-scale agent framework in
availability. The task agents sequence the various order to explore functionalities and challenges in the
requests in order before starting with the first request agent based systems [8]. However, for the work
and once one service is completed the finished result presented in this paper, it was decided that more
is automatically transmitted to the next service. advanced framework developed by other researchers
Figure 1 shows the basic structure of agent based would be more suitable. Several criteria were
programming in which the Task Agents and the considered and analyzed in order to decide the most
Service Agents exchange data in which the task agents suitable: message passing mechanism, ease of
sends data to the service agents which uses it to registering and finding agents, and compatibility with
execute the service and return it to the task agents. the existing IMPlanner system. JADE framework [9]
was chosen because of its ease to create ontologies for
message passing, compatibility with IMPlanner
codebase in Java, and agent user interface to monitor
development and execution.

4. Functionalities of IMPlanner
Figure 1. Basic Structure of Agent Based
Programming
The IMPlanner systems is a manufacturing process
The various characteristic features of Task agent
planning system built and developed over the years to
and Service Agent are discussed as follows. The main
perform various functionalities required for
action of the Task Agent is to search for the server
manufacturing planning [10] [11]. The details of all
which will accept the job in sequence. The transferring
the processes are beyond the scope of this paper. We
of data happens through a series of services located
will give short insight into all the processes that have
remotely by searching their Internet Protocol address.
been developed. The different functionalities of the
The Service agent’s main function is to process the
IMPlanner system are as follows:
jobs which are waiting in their queue. The Service
• Feature Mapping
agent follows the normal first in first out queue policy.
• Feature Precedence Network Generation
The Task Agent on the other hand follows the policy
• Process Selection
of a rule based system.
• Process Planning Network
The task agent contains contents along with the IP
• Process Simulation
address and port number of the various service agents
 Dusan N. Sormaz et al. / Procedia Manufacturing 26 (2018) 1242–1254 1245
Sormaz, Sarkar, Ghosal/ Procedia Manufacturing 00 (2018) 000–000 4

In this section we give a brief overview of the information, (i.e. the information about the type,
procedures with references to published work that dimensions, tolerance etc. of the feature), the process
contains more details. information (i.e. the different types of processes like
drilling, reaming, grinding etc. and their respective
4.1. Feature Mapping specifications), the machine and the tool information
(i.e. the available machine and tool information with
Feature mapping is the process of detecting the their respective capabilities). A rule based system
features present in a CAD file and map them with their makes decisions just like normal human reasoning
respective dimensions, type, quality tolerance and where based on the requirement of certain features,
surface finish requirements as manufacturing features. appropriate process candidates are selected. When the
The object model for a CAD file is named rule based system is run, the feature facts are generated
MfgPartModel. The IMPlanner system takes .prt files using the machine facts and the tool facts, which are
generated by the NX software as input and detects the populated in the working memory of the system. These
features present in the file using a programming facts in turn trigger certain rules which again trigger
interface with NX software . For each feature different other rules. This process continues until appropriate
tag values are assigned in which they can be detected. process candidates for all features are obtained.
The manufacturing design is then redrawn in the Sormaz et al. [14] [15] designed the rule based system
IMPlanner software palate with the different features using JESS language and integrated it with the Java
highlighted. programming platform of IMPlanner. When the rule
based system is executed, for each feature, it creates a
4.2. Feature Precedence Network Generation: template with dimensions, tolerances and type of the
feature. This template is matched with various
The feature precedence analysis system takes the available templates with given tolerances and
CAD feature model as input and produces feature dimension requirements generated from the rule based
relationship information as output. This information is system. After the entire process is complete, each
used to create precedence constraints which are feature in the feature list of a part model is populated
presented in the feature precedence network. In this with process candidates, each having corresponding
process the features which are related to each other are machine name, tool information and processing time.
selected and are arranged in the order in which they
should be machined. Queries are performed on the 4.4. Process Planning Network
CAD model to detect face adjacency which in turn
provides the feature adjacency. Once the adjacencies In this application we developed a model for the
are determined and combined in a feature interaction generation of alternative process plans. We proposed
graph, analysis of features are done pairwise using a an explicit process plan network (PPN) representation
feature interaction analysis engine [12] [13]. This by applying feature precedence constraints and
process is executed on all the interacting features and including alternatives on individual features.
finally the precedence information is arranged into a This work is based on results shown in [16] where
feature precedence network (FPN) for the part model. a space search based algorithm for the generation of
alternative process plans has been described. The
4.3. Process Selection previous results have convinced us that it is necessary
to build an explicit network representation for reasons
For each of the manufacturing features mapped of flexibility. That means that a process plan network
from the CAD model, the IMPlanner system finds provides a means for storing the result of process
process candidate (i.e. manufacturing processes which planning, and that any change in the design may result
are capable of generating those features). This is done only in adjustment to the network without need to redo
using a rule based system (jess.rbpp). A rule based process planning. The procedure is tightly linked with
system is one that finds solution to given problems on the feature modelling application and consists of three
the basis of rules generated from certain information. steps: (a) FPN validation, in which an FPN generated
For generating process candidates for different in the feature modelling application is a check for
features, the rule based system requires feature validity (for example, existence of circular constraints
1246 Dusan N. Sormaz et al. / Procedia Manufacturing 26 (2018) 1242–1254
Sormaz, Sarkar, Ghosal/ Procedia Manufacturing 00 (2018) 000–000 5

would render the FPN invalid), (b) process network agents working in a market structure. For each of the
generation, in which, for a valid FPN and alternative operations - Feature Recognition, Process Selection,
processes on features, the PPN for the part is Process Planning, Process Simulation, corresponding
generated, (c) PPN optimization, in which a space services are generated which are executed by the
search or network optimization algorithm is applied on service agent. We first describe market-based
the PPN in order to select the optimal process plan architecture, then agent communication, and in the end
These two procedures, PPN generation and we present workflow for distributed processing in
optimization, are also implemented into a hybrid IMPlanner-MAS.
procedure that simultaneously generates the portion of
the PPN needed to select the optimal plan. Currently, 5.1. Market Architecture
we are working on incremental generation of the
network based upon any addition or deletion of Agent configuration for IMPlanner-MAS was
features, or any other changes in the feature created by using functional decomposition, in which
precedence network. one service agent governs each manufacturing
planning step. Task Agent stores different service
4.5. Process simulation needs to be processed on a part design according to the
interaction with user through a graphical interface.
Simulation model is implemented as a template Task Agent searches for registered services and sends
capable of producing different model types that request with input data to the service agent and in turn
correspond to various routing policies [17]. This receives the result from service agent. Overall
template has been developed in AnyLogic, architecture of this configuration is shown in Figure 2.
representing the logic of i) routing policy, ii) Applied approach is focused on increasing the level
dispatching rule, iii) process plan review, iv) process of autonomy, independence and intelligence of service
plan assignment, and v) system status update in terms agents. In fact, we preferred to call them freelancer
of machine utilization and queue size. There are three agents, since each of them is capable of performing a
sub-models in every simulation model, Part Arrival particular service by using a localized algorithm or
Model, Manufacturing Model, and Exit Model. rule engine, and produce result for various tasks. The
Process plan selection is triggered at every predefined main aim of designing freelancer agents is to enable
decision making points based on the current system any third party service provider to take part in the
status. bidding. This strategy also completely decouples the
Four different models for the FMS control are service agents from the task agents. Task agent (in
developed: Static Best Model, Static Random Model, presented model its name is IMPlanner agent) can
Routing Dynamic Model, and Feature Focused request service based on different user interactions
Dynamic Model. Four different dispatching policies through the GUI attached to it but doesn’t need to
are also deployed in terms of part priority selection on know which service agents are performing the service
different machines. In dynamic models, Process Plan or how they are performing the service.
Selection Module also monitors the entire systems Approach as shown in Figure 2, is based on
status and changes between different plans to avoid clustering agents into different coalitions in order to
long waiting queue at decision points in order to respond to a request from various markets. The market
increase the utilization of every available machine. in this approach is established for different stages in
Machine utilization and other output of the system product design and manufacturing: feature recognition
are recorded after running all combinations of and modeling, process selection, process sequencing,
different routing and dispatching policies, then and scheduling. However, those markets may be
analysis is performed to determine the best performing extended for other manufacturing planning tasks, such
model type and dispatching policy. as layout design, cell design, assembly planning and
others.
5. Implementation of the IMPlanner-MAS

In this work all the functionalities available in

IMPlanner are done using independent intelligent
 Dusan
Sormaz, N. Sormaz
Sarkar, et al.
Ghosal/ / Procedia
Procedia Manufacturing
Manufacturing 26 (2018)
00 (2018) 1242–1254
000–000 12476

Figure 2. Architecture of market-based coalition of distributed dealers and services

A separate market is created for each functional functional task. This extra layer of abstraction is
domain in the manufacturing planning. IMPlanner designed for two reasons. Firstly, as service agents are
agent requests a certain service to a specific market. A provided by third party and may depend on other
Dealer agent, which has its own rules for performing service agents for accessing distributed resource,
negotiations with the service agents, controls each sometimes a single freelancer agent may not be
market. Different negotiation policies like, bidding, capable of performing the complete task requested by
auctioning, tenders can be adopted while building the task agent. Secondly, this extra layer provides more
rule engine for a dealer agent. In fact, freelancers don’t decomposition and opportunity for third parties to
participate in the negotiation directly. Separate types focus on different negotiation strategy and take part in
of agents called Vendor Agents recruit different the system by designing their own vendor agents. Each
service agents required to perform a complete vendor agent forms a separate coalition by recruiting
1248 Dusan N. Sormaz et al. / Procedia Manufacturing 26 (2018) 1242–1254
Sormaz, Sarkar, Ghosal/ Procedia Manufacturing 00 (2018) 000–000 7

different service agents needed for performing the ProcessMappingService and ProcessSimulation
task. This coalition is only created when Vendor agent Service.
receives a service request and disbanded as soon as the
particular service is performed completely and result 5.2. IMPlanner-MAS Communication
is delivered. In this way a freelancer agent is freed
right after the service is performed and can take part in One of the most important aspects of Agent based
another coalition. programming is proper communication between the
The IMPlanner-MAS system has been designed agents. For communication between agents the
using JADE middleware and the Java class hierarchy language used is called ACL or Agent Communication
for the same is shown in Figure 3. Language. FIPA has certain rules and regulations on
this and any ACL which follows the FIPA regulations
are called FIPA compliant ACL. The FIPA
specifications recommend using explicit or implicit
ontology for communicating between agents. Here we
have used explicit ontology to communicate different
manufacturing related data required for different
functionalities and also to communicate the purpose of
the message to make the receiver understand what
action to take based on the message type. Different
terms are created for communicating between agents.
Figure 4 shows a few communication semantics of
IMPlanner-MAS. HaveVendor is a predicate which is
sent to dealer agents in order to make them understand
that the IMPlanner agent is requesting to locate
vendors capable of a particular service. It contains bid
value, time limit etc. - parameters which are required
for negotiation. Send_Quotes and Service are requests
sent to service agents for negotiation and final work
request respectively.
When a service agent accepts a quote, a service
semantic is used to send the content required for a
particular functionality to the service agent.
Figure 3 Class diagram of IMPlanner MAS

The parent agent for all agents running in the JADE

framework is jade.Agent. The parent agent created
inside IMPlanner-MAS is MfgAgent which is
extended from jade.Agent. Four types of agents are
present in the market structure as discussed earlier:
The IMPlanner agent, dealer agent, service agent and
system agent. The system agent acts as the controller.
All of these are extended from the MfgAgent. Four
Figure 4 Semantics of agent communication
types of dealer agents: FeatureMappingDealer agent,
ProcessSelectionDealer agent,
The content which is sent with the messages is
ProcessPlanningDealer agent and
constructed using explicit ontology of IMPlanner.
ProcessSimulationDealer agent, all extend the
Manufacturing data in various forms (like part file,
DealerAgent class. The ServiceAgent uses the Service
machine data, feature facts, process facts, etc.) is
class to perform various services. The Service class is
required to be sent between various agents. These sets
extended by four different types of services:
of data are in general populated in object model. So a
FeatureMappingService, ProcessSelectionService,
well-developed communication data structure was
 Dusan N. Sormaz et al. / Procedia Manufacturing 26 (2018) 1242–1254 1249
Sormaz, Sarkar, Ghosal/ Procedia Manufacturing 00 (2018) 000–000 8

needed to encode and decode these messages. This is 5.3. Distributed Planning Workflow
done using concepts. Concepts are expressions which
can be used to define complex data structures using The workflow of various distributed agents which
their various attributes. They can be expressed as constitute IMPlanner-MAS is shown in Figure 6.
human readable versions and can be converted to the When the IMPlanner-MAS is run, the system agent,
object model. For developing the concepts for which is the virtual controller of all the active agents,
IMPlanner, the manufacturing planning ontology is starts running and the platform containing all the
built which is called the IMPlanner ontology. In order containers with their respective agents are created. All
to make ontology work with all the manufacturing the service agents and dealer agents are registered to
planning objects, an extra layer of inheritance is added the DF of the main container.
to the IMPlanner library where all the objects that are There are service agents capable of performing four
part of the agent communication are extended from different types of services: Feature Recognition,
MfgConcept. MfgConcept is a type of concept created Process Selection, Process Network and Process
in IMPlanner system which implements the Concept Simulation respectively. For each type two service
interface, thus enabling the ontology generation in agents are created in two different containers. For
JADE platform. All information sent between agents example, service agents F1 service and F2 service are
for different functional purposes are wrapped in an created in container 1 and container 2 for feature
object of a semantic type which inherits from recognition service. The dealer agents for the
MfgConcept. For example, feature facts are sent from corresponding services are created in the main
IMPlanner agents to process selection dealers, and container along with the DF, AMS and RMA. The
from there to Process Selection Service using a controller agent (i.e. the system agent) also resides in
wrapper called FactFile where the feature facts are the main container. Figure 7 shows the agent
stored as String attributes. This FactFile inherits distribution in the JADE platform after starting
MfgConcept so that while sending a message, IMPlanner –MAS.
ontology encrypts the message object in semantic The launch IMPlanner behavior generates an
format and the agent that receives it can decrypt it embedded window of IMPlanner system inside the
using the same protocol. Figure 5shows the class IMPlanner-MAS window. Figure 8 which is displayed
hierarchy of different concepts created in IMPlanner in Section 6 shows the generation of an IMPlanner
system to facilitate transfer of manufacturing data window which is an internal frame of the IMPlanner-
between agents. MAS window for two different parts.

Figure 5 Concepts of IMPlanner MAS

1250 Dusan N. Sormaz et al. / Procedia Manufacturing 26 (2018) 1242–1254
Author name / Procedia Manufacturing 00 (2018) 000–000 dns 9

Figure 6 Workflow of IMPlanner MAS

extent of matching between the search string and the

name of dealer agents. If the Lavenshtein’s distance
falls within a certain range, that dealer agent is selected
as one suitable for the purpose. Further details of the
search algorithm are beyond the scope of this paper.
Once the IMPlanner agent finds a feature mapping
dealer agent and service, the IMPlanner agent sends
the query message to it. The dealer agent, after
receiving the query message, sends the same message
to that service agent. The service agent decrypts the
message and sends it to the feature recognition service
where the part file is rebuilt from the string buffers
present as content in the message. This part file is fed
in to create an object model of the part file called
MfgPartModel and the feature list is populated with all
the detected features. Henceforth the instance of
Figure 7 Agent distribution in JADE platform of IMPlanner-MAS MfgPartModel will be referred as part model. The part
model is returned back from the service agent to the
Whenever the IMPAgent needs to find dealers or feature recognition dealer as a reply of the request
services from multiple available agents in the registry, message sent by the dealer, which uses the stored
search of the registry is performed The search for an address of the IMPlanner agent from where the initial
appropriate dealer is completed using the request originated, to send back the part model with
Lavenshtein’s search algorithm designed by [18] using the list of the features. This part model is saved inside
string parsing of the dealer’s name registered in the the IMPlanner agent. Figure 8 which is displayed in
DF. Based on the requirement Lavenshtein’s distance Section 0 shows the detected features enlisted inside
is calculated, which is a value used to calculate the the IMPlanner GUI after completion of the feature
recognition service.
 Dusan N. Sormaz et al. / Procedia Manufacturing 26 (2018) 1242–1254 1251
Author name / Procedia Manufacturing 00 (2018) 000–000 10

Table 1. Input and Output Format for each service evaluation of different dynamic routing policies. The
steps of process planning procedure and simulation
Service Input Output model have been executed as independent agents as
Feature CAD file Part feature model described above.
recognition
Process Jess facts Process alternatives
An example FMS that contains eight machines is
selection for each feature simulated in this FMS system. The machines are:
Process Process Process plan network CNC - Drill machines, 2 units, one with all tools
network alternatives and best routes having carbide and the other with all tools having HSS
Simulation Routes for System performance as tool material.
several parts CNC - Horizontal-Mill machines, 3 units, one with
all tools having carbide, one with all tools having HSS
Similarly, for each IMPlanner function dealer and and the other with all tools having both carbide and
service search is performed and message passing is HSS as tool material.
cascaded through the whole process. Input and output CNC - Vertical-Mill machines, 3 units, one with all
formats for each service are shown in Table 1Error! tools having carbide, one with all tools having HSS
Reference source not found.. All messages are and the other with all tools having both carbide and
created from IMPlanner model objects and serialized HSS as tool material.
for agent communication. As mentioned, for feature Process, machine and tool matrix is shown in Table
recognition service IMPlanner agent serializes the 2. Machine-tool assignment in manufacturing cell. As
complete CAD file and sends it to Service Agent shown in Table 2, machines are limited to certain
which reconstructs the file and performs feature processes. For Example, CNC- Horizontal Mill (HM)
mapping. Serialized MfgPartModel is sent back. For machine is limited to End Milling (EM) Slotting, Side
process selection service IMPlanner agent creates a set Milling (SM) Slotting, and End Milling (EM)
of Jess facts, since process selection is done using rule Peripheral, whereas CNC Drill machine is limited to
based approach, as described in Section 4.3. The result perform only drilling operation on the parts as shown
is again sent back as a set of serialized objects. in Table 2 with solid green (light shaded) blocks.
Similarly, Process network service takes serialized Machines are also constrained by Tools which can be
process alternatives for each feature, makes its replica accessed by individual machine, but this information
and generates process plan network. is omitted from the table.
The workflow for the simulation service includes The FMS produces six parts which have been run
the capability of IMPlanner agent to handle multiple through process planning and simulation evaluation
parts simultaneously in order to create a set of routings procedures, shown in Table 3. The parts range from 6
used to evaluate the performance of the system which to 28 features, representing industrial parts of average
will be described in Section 0. complexity.
The process planning agents have been executed
6. Simulation of dynamic routing policies for each part individually in order to generate routings
necessary for the simulation model . An example of
The proposed distributed framework has been the results is shown in Figure 8 which shows results of
deployed in order to develop a simulation model for executing the Feature mapping agent on two first parts.

Table 2. Machine-tool assignment in manufacturing cell

Process
Tool Material Machines
Drilling EM Slotting SM Slotting EM Peripheral
CNC - HM No Yes Yes Yes
Carbide
CNC - VM No Yes No Yes
CNC - Drill Yes No No No
CNC - HM No Yes Yes Yes
HSS CNC - VM No Yes No Yes
CNC - Drill Yes No No No
Both (Carbide & CNC - HM No Yes Yes Yes
1252 Dusan N. Sormaz et al. / Procedia Manufacturing 26 (2018) 1242–1254
Author name / Procedia Manufacturing 00 (2018) 000–000 11

HSS) CNC - VM No Yes No Yes

Table 3. Mechanical designs and their features

Name AES94 Plate Netex

Model

Number of features 10 11 7

Name USC Slider Bracket

Model

Number of features 6 18 11
 Dusan N. Sormaz et al. / Procedia Manufacturing 26 (2018) 1242–1254 1253
Author name / Procedia Manufacturing 00 (2018) 000–000 12

Figure 8 The system status after two Feature mapping agents have completed the work

After running all three different agents (feature Manuf., vol. 14, no. 1, pp. 7–23, Feb. 2003.
mapping, process selection, and process network) on [4] W. Shen and D. H. Norrie, “Agent-Based Systems for
each part, the final process plans of all parts are sent to Intelligent Manufacturing: A State-of-the-Art Survey,”
Knowl. Inf. Syst., vol. 1, no. 2, pp. 129–156, May 1999.
the simulation agent, which launches the simulation
model built in AnyLogic. The animation of the [5] L. Wang and W. Shen, “DPP: An agent-based approach for
distributed process planning,” J. Intell. Manuf., vol. 14, no.
simulation model run is shown in Figure 9. The 5, pp. 429–439, Oct. 2003.
detailed results of the simulation and the FMS
[6] D. Sormaz, “Distributed Modeling of Manufacturing
performance are beyond the scope of this paper and
Activities using Integrative Manufacturing Process
they can be found in [17]. Model,” Int. J. Ind. Eng. …, vol. 1, no. 1, pp. 9–18, 2010.
[7] A. Sarkar, D. N. Sormaz, and J. Huang, “Multi-Agent
framework for design and manufacturing integration,” in
International Conference on Production Research, 2012.
[8] S. Ghosal, D. Sormaz, and A. Sarkar, “Educational System
for Agent-based Distributed Task Processing,” 2014.
[9] F. L. Bellifemine, G. Caire, and D. Greenwood,
Developing Multi-Agent Systems with JADE, vol. 5. Wiley,
2007.
[10] D. N. Sormaz, J. Arumugam, and S. Rajaraman,
“Integrative process plan model and representation for
intelligent distributed manufacturing planning,” Int. J.
Prod. Res., vol. 42, no. 17, pp. 3397–3417, 2004.
Figure 9 Simulation model of the FMS in AnyLogic
[11] D. N. Šormaz, J. Arumugam, R. S. Harihara, C. Patel, and
N. Neerukonda, “Integration of product design, process
7. Conclusions and future work planning, scheduling, and FMS control using XML data
representation,” Robot. Comput.-Integr. Manuf., vol. 26,
In this paper we have presented a framework, no. 6, pp. 583–595, Dec. 2010.
development and implementation of the distributed [12] D. N. Sormaz and B. Khoshnevis, “Modeling of
agent-based manufacturing planning system. The manufacturing feature interactions for automated process
planning,” J. Manuf. Syst., vol. 19, no. 1, pp. 28–45, 2000.
previous work on IMPlanner has been extended to
provide for autonomous implementation of each [13] J. Arumugam, “Analysis of Feature Interactions and
Generation of Feature Precedence Network for Automated
process planning function, and their integration using Process Planning,” Ohio University, Athens, OH, 2004.
JADE framework. The framework is demonstrated on
the simulation model of the small FMS in which agent [14] D. N. Sormaz, P. Khurana, and A. Wadatkar, “Rule-Based
Process Selection of Hole Making Operations for
execution of individual parts is integrated into the Integrated Process Planning,” in ASME 2005 International
model which allows the performance evaluation. The Design Engineering Technical Conferences and
future work will include conversion of the rule based Computers and Information in Engineering Conference,
2005, pp. 983–988.
system to semantic databases, deployment of agents as
web services, and deployment in the cloud platforms, [15] D. N. Sormaz, J. Arumugam, and C. Ganduri, “Integration
of Rule-based Process Selection with Virtual Machining
such as AWS. for Distributed Manufacturing Planning,” in Process
Planning and Scheduling for Distributed Manufacturing,
References L. W. PEng and W. S. PEng, Eds. Springer London, 2007,
pp. 61–90.
[1] “Welcome to the Foundation for Intelligent Physical [16] D. N. Sormaz and B. Khoshnevis, “Generation of
Agents.” [Online]. Available: http://fipa.org/. [Accessed: alternative process plans in integrated manufacturing
05-Dec-2017]. systems,” J. Intell. Manuf., vol. 14, no. 6, pp. 509–526,
Dec. 2003.
[2] D. B. Lange, “Mobile objects and mobile agents: The
future of distributed computing?,” in ECOOP’98 — [17] S. Ghosal, “Object Oriented Simulation of Agent Based
Object-Oriented Programming, 1998, pp. 1–12. Flexible Manufacturing System with Dynamic Routing and
Scheduling,” Ohio University, 2015.
[3] M. Fletcher, R. W. Brennan, and D. H. Norrie, “Modeling
and reconfiguring intelligent holonic manufacturing [18] D. Sormaz and A. Sarkar, “Distributed Integration of
systems with Internet-based mobile agents,” J. Intell. Design and Planning Activities in Manufacturing using
1254 Dusan N. Sormaz et al. / Procedia Manufacturing 26 (2018) 1242–1254
Author name / Procedia Manufacturing 00 (2018) 000–000 13

Intelligent Agents,” in TMCE, 2014.