Fairtrace
A Semantic-web Oriented Traceability Solution Applied to the Textile Traceability
Bruno Alves
1
, Michael Schumacher
1
, Fabian Cretton
1
, Anne Le Calv
´
e
1
, Gilles Cherix
2
,
David Werlen
2
, Christian Gapany
2
, Bertrand Baeryswil
3
, Doris Gerber
4
and Philippe Cloux
4
1
IIG, University of Applied Sciences Western Switzerland, TechnoArk 3, Sierre, Switzerland
2
Institut Icare, TechnoArk, Sierre, Switzerland
3
Fairtrace, Technoark, Sierre, Swtizerland
4
Importexa SA, 1095 Lutry, Switzerland
Keywords:
Traceability, Ontology, Semantic Web, Textile Industry.
Abstract:
This paper presents solutions that leverage Semantic Web Technologies (SWT) to allow pragmatic traceability
in supply-chains, especially for the textile industry. Objectives are the identification of the supply-chain, order
management, tracking and problem reporting (such as dangerous substance detection). It is intended to be a
generic platform supporting potentially any kind of industrial supply-chain, to be usable in harsh environments
(mobile appliances) without any kind of communications possibility and to be fully usable to non-IT people,
including for the modeling of the production processes. The developed solutions also allow the consumer to
benefit from the traceability through information pages available by scanning the QR codes available on the
finished products (clothes, clocks, etc.). This paper presents: i) the methodology applied to achieve those
functionalities, ii) the design and implementation choices, and iii) the test results. The main value of this
paper is the usage of the Semantic Web in real-world industrial traceability solutions, which were tested in real
supply-chains in Switzerland and India. The commercialization of the developed solutions is in preparation.
1 INTRODUCTION
In 2007, a new European regulation on chemicals
called REACH
1
initiated the creation of a catalog
of potentially dangerous substances actively used in
everyday consumer goods. Many of these have been
identified as potentially threats to human health and
are therefore forbidden on the European territory.
REACH puts the responsibility on the industry for
evaluating and managing the risks about chemicals
they use or import. As a consequence, industrial
actors henceforth have the duty of tracing all
substances in use in the products they manufacture or
import in Europe. Unfortunately, in practise this is
barely the case. As an example, during the EURO
2012 soccer championship, Europe’s Consumer
Watchdog revealed unusually high concentrations of
dangerous chemical substances in several team shirts
1
REACH - Registration, Evaluation, Authorization and
Restriction of Chemical Substances - http://ec.europa.eu/
environment/chemicals/reach/reach˙intro.htm
that could potentially be harmful to fans’ health
2
.
Independent tests highlighted high concentrations
of lead, nickel and organotin, a chemical that
can irreversibly damage the human nervous system.
That’s just one example among many others, but it
shows a major failure of brands and the clothing
industry in their capacity of fully capturing or even
understanding their own manufacturing processes.
When a problem is finally discovered, it is often
already too late and the cost of any corrective measure
is usually too high.
The case shown above is unfortunately too
common and stands as a motivation driver for the
work described in this paper. Bringing supply-chain
traceability to the end consumer and to the economic
partners (resellers or brands) is a difficult task.
Information related to the manufacturing process is
usually not made available (obfuscated on purpose
or by lack of sufficient means) or extremely opaque.
2
http://news.stepbystep.com/euro-2012-football-fans-
warned-against-buying- toxic-shirts-313/
36
Alves B., Schumacher M., Cretton F., Le Calvé A., Cherix G., Werlen D., Gapany C., Baeryswil B., Gerber D. and Cloux P..
Fairtrace - A Semantic-web Oriented Traceability Solution Applied to the Textile Traceability.
DOI: 10.5220/0004440900360045
In Proceedings of the 15th International Conference on Enterprise Information Systems (ICEIS-2013), pages 36-45
ISBN: 978-989-8565-59-4
Copyright
c
2013 SCITEPRESS (Science and Technology Publications, Lda.)
However, in order to become REACH-compatible,
industrial actors will have the difficult task of
motivating each participant belonging to the process
to commit to the gathering of the necessary
information.
The work described in this paper provides the
outline of a first answer to the following research
questions: what is the best way to represent data
related to a supply-chain, so that it is flexible for
future adaptations and generic enough to be applied
to several industrial domains? How to design a
software solution that helps non-IT specialists in
defining their supply-chains themselves and that
assists their navigation through the acquired data?
How to provide an easy tool-set for the acquisition of
process information ? How to give a full transparent
access to this information through the entire chain,
so that we can really support traceability? To
tackle those problems pragmatically, we initiated the
Fairtrace project, a generic traceability framework.
The prototype presented in this paper is applied to the
textile industry.
The objectives of the project are primarily the
identification of the supply-chain (activities), order
management and monitoring, as well as problem
reporting (such as dangerous substance detection). It
is intended to be a highly generic platform supporting
potentially any kind of industrial supply-chain
(clock industry, cocoa, ...); to be usable in harsh
environments (mobile appliances) without any kind of
communication facilities and to be usable by non-IT
people (including the modeling of their production
processes). Our solution must also benefit the
final consumer (buyer), allowing him/her to obtain
information on the traceability with a simple scan of
a QR code on finished products (clock, cloth, ...).
To support our objectives, we decided to leverage
Semantic Web Technologies (SWT) in the core of
the Fairtrace software to assess potential advantages
when applied to an industrial setting.
A fully functional prototype has been designed
and implemented. Field tests have been realized in
India in February 2012. We have collected real-time
information triggered by an order for an organic
fair tee-shirt. A startup company founded by our
commercial partner Importexa is currently preparing
the commercialization of a product based on Fairtrace
applied to the textile industry. Other markets will also
follow soon. A part of the technology has already
been patented (a dynamic formular creation system
based on semantic data (Werlen et al., 2012)).
This paper is structured in five parts: the first part
discusses some aspects of research in specific areas
of supply-chain management. In the second part, we
detail in descriptive terms the methodology of our
work. Then, in the third, we present the results of
the project and various discussion issues in the fourth
part. We then conclude with some future directions
and challenges still to overcome.
2 RELATED WORK
Fairtrace aims at achieving an agile traceability
system that can help in the management and
monitoring of supply-chains. Supply-chain
management (SCM) is concerned with the
coordination of activities for producing a
product demanded by a customer (Mentzer et al.,
2001)(Chandra and Kumar, 2001). Because
thoughtful management of the supply-chain often
results in substantial cost savings, the field has seen
considerable industrial and research activity. It is
no surprise then, the literature on the subject is so
abundant that all aspects are well covered. (Lambert
et al., 1998)(Lambert and Cooper, 2000) describe
issues and opportunities of SCM. Other works
propose formal SCM frameworks (Grubic and Fan,
2010), infrastructures (Fox et al., 2000)(Christopher,
2005) and models (Ye et al., 2008b). Logistics is a
very active field where research focuses especially
on optimization patterns (Bowersox et al., 2002).
Supply-chain monitoring (SCMo) is a sub-branch
of SCM that operates on information, inventory
management and cash-flow (Lambert et al., 1998).
The monitoring strives to rapidly identify problems
in the supply-chain and solve them with the help of
established procedures.
The focus of this paper is on supply-chain
traceability (SCMt), which primarily addresses the
problem of tracing goods from raw materials to
finished products. An identifiable trend on SCMt
is the use of external technologies such as RFID
to enable traceability (K
¨
arkk
¨
ainen, 2003)(Kelepouris
et al., 2007), including in the textile industry (Kwok
and Wu, 2009). Many publications also exist on
SCMt processes and implications, such as (Opara,
2003). An important research track has been also
developed on the composition of Web services in the
supply-chain in order to ensure traceability among
the partners of the chain (Kim and Jain, 2005).
These works have been extended by enriching Web
services with Semantic descriptions, in order to
avoid ambiguity among the services and allowing
an explicit meaning of the data interchange (Mocan
et al., 2006). These Semantic Web Services (Paolucci
et al., 2002) and their coordination (Schumacher
et al., 2008) allowed to improve mutual sharing of
Fairtrace-ASemantic-webOrientedTraceabilitySolutionAppliedtotheTextileTraceability
37
information for business-to-business integration.
In order to define a common understanding
of supply-chain models, several research works
proposed the use of supply-chain ontologies (Grubic
and Fan, 2010). For instance, (Ye et al., 2008a)
proposed quite a complex ontology that is not
industry specific. Those work show that Semantic
Technologies can have many advantages for the
interoperability of the business partners in the chain.
Research works also exist to use Semantics with
RFID solutions (Virgilio et al., 2010). (Bechini
et al., 2008) also proposes semantics description to
ensure traceability. Other interesting research papers
describe the usage of autonomous agents to deal
with the information in the supply-chain (Fox et al.,
2000), including its combination with Semantic Web
descriptions (Datta et al., 2007).
In relation to the research above, the work
presented in this paper can be specified as follows:
Fairtrace is meant to trace all activities of the entire
manufacturing process, including documents and
products. It is able to identify the raw materials which
were used, certificates, waste, etc. It can also be
used to identify roles and responsibilities and to keep
track of who did what. It does not currently include
optimization techniques for the logistics. Fairtrace
can track many variables, which can be used as
performance indicators. A customizable chain of
validators also allows to track problems arising along
the supply-chain, such as the detection of dangerous
chemical substances. Businesses can thus react
very quickly to any unexpected condition. From
a technological perspective, Fairtrace was conceived
with a focus on Semantic Web Technologies (SWT).
Every piece of data in the system is related to a
concept in a semantic data repository. All models
use semantic descriptions and traceability information
can be displayed to the user hierarchically.
The next sections explain thoroughly how
Fairtrace was conceived, and its main results.
3 METHODOLOGY
The scoping phase of the project consisted in
capturing and understanding the requirements of
traceability. An analysis of a typical textile
manufacturing chain was made to get a rough idea on
the process. A consultant specialized in the textile
industry was commissioned to analyze Importexa’s
own supply-chain in India. Her mission was to
visit factories, identify and document all activities
throughout the manufacturing process. She captured
information about certificates, delivery challans,
cotton lots, mixing lots and many other documents,
but also took pictures of places and production
machines. Our task was thereafter to analyze all
those paper resources in order to sketch out a formal
description of the longitudinal process. The very
first model was a transcription of the complete,
moderately detailed manufacturing process using the
primitives of BPMN
3
.
The process was complemented by several
additional documents more accurately describing the
information to be captured on the supply-chain. A
critical path — a minimal path from the starting
activity (order) to the end (ginning)— was defined.
The method consisted in searching through pairs
of documents such as orders, challans or bills for
matching identifiers to find a complete and continuous
traceability path up to the origin of the cotton. The
base model was augmented with additional attributes
(GSM, weight, ...) extracted from the documents that
we analyzed. This extra information helps tracking
quantities and waste.
3.1 Requirements
Fairtrace objectives focus on usability, supply-chain
data transparency, genericity and adaptability.
As a consequence of the first requirement,
business partners in the supply-chain needed a simple
way to enter data about their production directly into
our system. Among all alternatives considered, a
system of minimally constrained web formulars was
retained. On one hand, it was important to give users
the freedom to design themselves these formulars
— possibly by drag-dropping GUI components
— without having to resort to any particular IT
knowledge or skills. On the other hand, formulars
needed to the bound to the underlying data model.
Web formulars coupled to a flexible data binding
system offer the kind of usability that was required
in this project. Web forms can be implemented in
pure HTML and are supported natively by all kinds
of devices. They can be adapted to any sort of display
they are affected to, with relatively few lines of CSS
code. They can be edited and pre-filled with data
coming from the repository and customers can mark
fields optional or required at wish.
The formular system was also designed to be
adaptable. Each formular is dynamically linked to the
underlying data model by a dynamic binding system
that uses specific binding names to associate fields
from the formulars to particular objects and properties
in the model. Coupled to an instantiation engine that
3
BPMN - Business Process Modeling and Notation -
http://www.bpmn.org
ICEIS2013-15thInternationalConferenceonEnterpriseInformationSystems
38
creates model instances based on the data acquired,
the whole system provides a very generic framework
that can adapt to almost any model or domain, without
requiring recompilation.
To provide transparency on the data, it is
important that information can be accessed without
any technological barriers. A dashboard system
providing supply-chain monitoring support and
capable of providing easy data exploration was
designed as a consequence. Furthermore, we wanted
the information to be presented to the customer both
horizontally (information for a specific time frame)
and vertically (hierarchy of data). Such a system
would allow anyone entitled to do so, to descend into
any part of the traceability data chain providing them
with a full transparent access to its contents.
3.2 Domain Modelling
The data model in Fairtrace was designed to capture
many different aspects of the supply-chain, including
details on the process itself and on the products
deriving from it. We divided the modeling task into
creating models for both.
To improve on usability, we wanted to empower
the customers (or any one entitled to do so) to design
graphically their supply-chains. We had to provide
the necessary primitives (activity, flow ) to allow it.
For that reason, we did not hard-code the process
into a model for each different customer, but decided
instead to specify a language partly derived from
BPMN describing business process concepts such as
activities, flows, roles, users, partners, authorizations
and collection points. Supply-chains would then be
specified in terms of these primitives.
Currently, our model only provides support for a
tiny subset of BPMN-like constructs. We added a
few more custom primitives (collection points) that
are used by the security granting mecanism. Before
resigning ourserves to model everything from scratch,
we tried on different approaches. We had a look to the
PSL
4
specification and ontogy, M3PO(Haller et al.,
2006) and BPMNO
5
. Unfortunately, the complexity
of the semantics and the verbosity of those models
seemed a bit ”overkill” for our needs.
The product model is strongly bound to the
industrial domain it describes; in this case, the domain
was the textile supply-chain. We searched on-line
for existing ontologies describing domain knowledge
on textile products, but unfortunately, we could not
4
PSL - Process Specification Language - http://
www.mel.nist.gov/psl/
5
BPMO - BPMN Ontology - https://dkm.fbk.eu/
index.php/BPMN˙Ontology
find any that more or less suited our needs. We
decided thus to also model the product domain from
scratch. To do so, we’ve literally analyzed dozens of
documents brought back from India by the consultant.
3.3 Semantic Technologies
The leveraging of Semantic Web Technologies was
a prerequisite in this project. We didn’t have an
extensive knowledge on the subject when the project
started. It was thus important for us to stay simple
and practical at all times. In order to cope with this
requirement at every step, we did a lot of testing on
example datasets. We only started implementing the
ontologies after we felt confident enough that there
would be no uncontrolled side effects (i.e inferences
that were not expected).
Our models didn’t require a high level of
expressiveness, since they were almost one-to-one
mappings from an object-oriented hierarchy. We
didn’t use elaborated constructs like Restrictions,
because we wanted to be sure of the decidability.
We also needed the language to be sound, complete
(all logical consequences are drawn) and monotonic
(i.e all statements remain valid after inserting
new knowledge). We translated the conceptual
models (process, products) to RDFS instead of a
more expressive OWL, not only because it was
easier to work with, but also because semantics
were easier to grasp. Our models are based
on subsumption hierarchies that use sub-class
and sub-property relations (rdfs:subClassOf,
rdfs:subPropertyOf) extensively.
The process model does not put any additional
requirements on the predicates it declares. No
particular semantics are specified, because this
model is very similar to what would be done on an
object-oriented programming language. The product
model however, needs more expressiveness in order
to model the existing links between objects of
different classes, but also to compensate for potential
gaps due to the lack of sufficient data. We had
to model inverse properties (owl:inverseOf) and
property chains axioms (only available in OWL 2).
In order to avoid the extra cost of using OWL, we’ve
decided to take an intermediate route and create extra
entailment rules for these very few specialized cases.
Here are some examples in predicate logic:
Simulating owl:inverseOf:
inOrder(X,Y), Type(X,Batch),
Type(Y,Order) => hasBatch(Y,X)
Simulating property chain axioms:
Fairtrace-ASemantic-webOrientedTraceabilitySolutionAppliedtotheTextileTraceability
39
hasFabric(X,Y), hasYarn(Y,Z),
Type(X,Order), Type(Y,Fabric),
Type(Z,Yarn) => hasYarn(X,Z)
3.4 Infrastructure
The prototype infrastructure was designed to meet
requirements for the three types of customers we
wanted to target: business partners, end users and
supply-chain partners.
The business partners category is mostly
composed of brands and resellers that will license
the future Fairtrace solution for monitoring their
own supply-chain. Their needs are primarily
basic-management controls, extensive information
and problem reporting. Those requirements were
implemented into the prototype as a supply-chain
designer, a graphical formular edition interface and
a dashboard to control all operations, assign users,
roles and authorizations.
To end users we wanted to provid them with
traceability information about a product on scanning
a QR code on the finished product. We had to design
a system to map these codes to specific views on
the information collected from the data store. That
information is then formatted to target the particular
device the user browses on.
Finally, we had to provide a way for the
supply-chain partners to send data to the system.
By partners, we mean any entity working in the
supply-chain to create and finalize the product. These
can be factories, transporters, single users... The
dynamic formular system we designed was meant to
be simple and functional. It supports simple common
controls such as date pickers, file selectors, picture
boxes, text fields, lists, combo boxes and other types
of selectors. It is thus possible to integrate certificates
(such as for biological production) as PDF/A files
as well as other kind of documents. Those binary
resources are stored on the cloud and a link is kept
inside the RDF data store. The formular design
and the dynamic binding systems were thought to
be used by designers not especially proficient with
technology. We favored simplicity and wanted to hide
the technical details (URIs, JSON object, ...).
Furthermore, we also considered both mobile and
browser-enabled applications to display formulars.
The mobile application was designed to manage and
display the formulars in potentially non connected
configurations. We had to take into account that
wireless signal could be a problem in some places
where the application was meant to be used (such
as in cotton fields in India). We designed an offline
mode, that would allow later synchronization with
the Fairtrace server. The browser-enabled application
was meant to be used directly in factories, where
Internet is generally available. We considered this
option too, because it was obviously easier and faster
to input data from a keyboard rather than on a mobile.
3.5 Implementation
The actual implementation phase went surprisingly
well. We opted for an agile development process
with a relatively short feature development cycle (<1
month). We were able to finish both on time and
on budget. We spent extra resources on polishing
important features, so that the prototype would be
ready for a demonstration to potential customers.
The last step of the development was testing the
prototype in a real situation. We modeled Importexa’s
supply-chain and requested them to initiate the
process. Importexa started an order for an organic
fair cotton tee-shirt. When the actual manufacturing
began, we had the consultant in India to teach the
functioning of the system and to check that everything
was going as planned. We wanted to show off that
the prototype was able to trace all the manufacturing
activities and capture all the products of it.
4 RESULTS
This project yielded three main results, which are the
functional prototype, two ontologies created from the
process and product models and the system and the
modeling of a textile manufacturing chain.
4.1 Prototype
The prototype shown on Figure 1 is a three-tier
infrastructure composed of a data, a business and
a presentation layers and was developed in Java as
a Spring MVC application running on a Tomcat 6
servlet container. The data layer hosts the RDF store
(semantic repository). The middleware is comprised
of the business layer and acts as a message translator
and rule enforcer between the application server
and the RDF store. Finally, the presentation layer
contains both the Business To Business (B2B) and
front-end operators. A Business To Consumer (B2C)
front-end solution maps QR codes that come on the
final products with information views about the whole
upstream chain to be presented to the final consumer.
The semantic functionality (i.e. semantic store)
is provided by the BigOWLim 3.5 engine from
ICEIS2013-15thInternationalConferenceonEnterpriseInformationSystems
40
Figure 1: Overview of the architecture.
OntoText
6
running as an extension of the OpenRDF
7
Sesame 2.4.0 runtime. A set of Data Transfer
Object (DTO) classes provide seamless data access
and querying functionality. The prototype exclusively
uses SPARQL 1.0 queries, since SPARQL 1.1
UPDATE was not supported by the versions of
OWLim and Sesame used in the prototype. Data had
to be programmatically inserted, updated or removed.
Queries for traceability data are made against the
implicit graph containing the inferred triples. Queries
on the process itself are made against the explicit
graph and returned to the API caller via the DTOs.
The business layer is comprised of a set of
classes that receive requests through a series of
REST Web service endpoints. The requests and
responses are encoded as JSON objects. Payloads
are translated, verified, business rules are enforced,
security applied and then forwarded as a data access
requests. Data extraction is done with SPARQL
SELECT and CONSTRUCT queries. Repository
modifications are done in transactions and are rolled
back in case of problem. Service endpoints that
manage process resources (objects) use the CRUD
paradigm (Create, Retrieve, Update, Delete) and
produce JSON-encoded messages. Data requested for
products is returned as a group of properties along
with the identifier of the individual (Listing 1).
The application server was built on the latest
version of Ruby on Rails and coordinates all requests
between the presentation layer components and the
back-end storage. It also enforces a certain number of
business rules, but its role is more to coordinate B2C
and B2B actions, unlike the back-end, which role is
to coordinate data requests. The B2B front-end is the
6
http://www.ontotext.com
7
http://www.openrdf.org
Listing 1: Product object encoded in JSON.
property_set: {
properties: [
{
property: {
name: " order_id " ,
label: {
" en ": " original order number "
} ,
value: " CA16576 "
}
} ,
{
property: {
name: " re f e r e n c e_id " ,
label: {
" en ": " purchase order number "
} ,
value: " 221 9 "
}
}
] ,
object_id: " order001 "
}
Web-based control panel of the system. It is intended
to be customizable to businesses willing to license
the Fairtrace ecosystem. It allows to dynamically
and graphically model the supply-chains, creating
the formulars and binding them to any step of the
associated process. It can monitor order progress by
showing various indicators and track any upstream
issue such dangerous chemical detection. Users can
then navigate vertically through traced data (Figure 2)
or horizontally (Figure 3) by querying the date inside
a specific time span.
The Operator front-end is aimed at supply-chain
partners working in coordination. It presents them
Fairtrace-ASemantic-webOrientedTraceabilitySolutionAppliedtotheTextileTraceability
41
Figure 2: Vertical navigation through data hierarchy.
Figure 3: Horizontal navigation through data.
formulars that were previously created in the B2B
front-end for them. It allows them to send data
related to their activities into the system. On
the mobile appliances, the frontend is an HTML5
container application that allows to download pending
formulars created on the B2B front-end, fill them,
update them and synchronize them back with the
application server. It provides support for an offine
mode, where data can be synchronized back later.
In the prototype, the detection of dangerous
substances was not codified in the model ontologies.
We decided to develop that functionality with a
chain a validators. Each validator is created as
a plug-in, which can analyze a subset of the
dataset and generate messages on certain conditions.
There are three types of messages: exception,
warning and informational. Validation can be trigged
manually from the dashboard or can be set to be
triggered automatically upon any modification of
the data. The validator for dangerous chemicals
validates each individual of class ft:Substance by
matching it against a database of known harmful
chemicals. One goal set for the next iteration of
Fairtrace is to use external sources directly integrated
with our models in order to be able to infer the
fto:DangerousSubstance class and take advantage
of the inferencing mechanisms.
4.2 Ontologies
Implementing the models was a bit delicate in
that a clear separation of concerns was necessary.
Having a multi-disciplinary team, we needed a
modularized design in order to allow parallel and
effective development. In order to create the
ontologies based on our models a RDF repository
to store ontology data was required. The choice of
BigOWLim as the semantic repository and reasoner
is the result of an objective test, where different
semantic repositories were compared based on a few
metrics (size, simplicity, ..). The BigOWLim 3.5
engine works as an extension of a SAIL repository
in Sesame. It is easy to setup, provides really
fast inferences on vast amounts of data, supports
geo-spatial and full-text search capabilities.
BigOWLim Triple Reasoning and Rule
Entailment Engine (BigTRREE) uses a total
materialization reasoning strategy that computes
inferences after insert/update/delete cycles. That
kind of strategy has the disadvantage of making
inferences a bit slower (especially on very large
datasets), because it computes the complete closure
(optimizations aside) on each transaction. Even
so, it does allow extremely fast queries, almost on
par with traditional DBMS systems. The Fairtrace
prototype heavily relies on queries, so BigOWLim
was certainly a righteous choice. An additional
feature that also helped in our choice was that it was
actually very easy to deploy and write entailment
rules for. We made an agreement with OntoText to
use their repository technology free of charge in our
research projects.
Almost every object used by the web and mobile
platforms is stored in the repository A-BOX as
an individual from either the process ontology or
the product ontologies. Each ontology is the
implementation of a conceptual model in RDF
Schema
8
.
The process ontology defines the building blocks
for modeling business processes. It is a straight
mapping from an object model to RDF. It does
not define any particular semantics, except the ones
explicitly given in the RDF Schema specification (?x
rdfs:type rdf:Resource for instance). It builds
around concepts like steps (activities), flows (links),
8
http://www.w3.org/TR/rdf-schema/
ICEIS2013-15thInternationalConferenceonEnterpriseInformationSystems
42
users, roles, authorizations, partners, collection
points and formulars. Formulars are assigned to
a particular collection point linked to a specific
step. Authorizations are defined for collection points
to state who and on what role can enter data
and download the formulars. The intent of these
primitives is not to fully map the BPMN specification,
but to present a synthetic view of the supply-chain.
Conditions, loops and other advanced primitives
are not yet supported. We did not reuse existing
ontologies, because we needed a simple and efficient
ontology.
The product ontology models the domain objects.
In contains definitions for textile industry objects,
such as fabrics, parts, dyes, yarns, mixing lots.
Concepts are linked together – in a critical path – so a
customer can eventually navigate through individuals
from the initial order down to the geographical place
where the cotton was produced. Every step and every
product of the supply-chain can thus be identified.
Documents and special objects, such as certificates
are also modeled and can be linked to binary data
(pictures). They can be used to prove that cotton is
really organic and fair and that all authorizations have
been obtained by a partner for instance. Importexa
can enforce that business partners insert them in the
system. In terms of semantics, the model uses mostly
subsumption hierarchies. We assert for example:
fto:Certificate rdfs:subClassOf
fto:Document .
fto:ProcessedFabric rdfs:subClassOf
fto:Fabric .
fto:hasCertificateName rdfs:subPropertyOf
rdfs:label .
Furthermore, to compensate for the lack of
property-chains (only available in OWL 2), we
defined custom rule-sets (see Section 3.3). These are
necessary in order to bridge the gaps in the model that
can occur due to the lack of sufficient data, causing
thus an impossibility to descend in the hierarchy.
4.3 Supply-chain Modelling
We also modeled a simple textile supply-chain
obtained from the analysis of the information taken
in the supply-chains in India. According to our
consultant, our process model, even simple matches
almost 90% of existing real chains. It models all
upstream steps such as Mixing, Spinning, Dyeing,
Processing, Transforming, Weaving, Sewing and
Packaging. Ginning was purposely left aside, since
information about the provenance of cotton was
readily available at the Mixing step.
4.4 Testing
The infrastructure was deployed and tested in India.
The objective was purely to assess its flexibility.
We did not acquire quantitative indicators such as
number of queries for each order, etc., because it was
materially (i.e. financially) not possible to launch
enough real orders to gather enough data for the
results to turn out meaningful. We concentrated
thus on a feasibility study following the project plan
methodology:
• We selected the supply-chain partners in India that
usually work with the textile company Importexa
in Switzerland ;
• A consultant from Importexa visited the partners
on site in order to thoroughly document the
supply-chains ;
• The consultant described formally the
supply-chains using a Business Process Modeling
notation ;
• On the basis of the consultant documentation
(data description, business process model, etc.),
we designed in our Web-based B2B Frontend
the processes and formulars acquiring information
inside those formulars ;
• Those formulars were instantaneously deployed
for data acquisition, both as HTLM5-enabled
mobile and Web formulars ;
• Importexa launched a real order for an organic fair
cotton tee-shirt, whose production data should be
acquired with our solutions ;
• A consultant was then sent to India in order to
teach working partners how to input data on the
forms and send it for treatment ;
• A final report described the results of the tests .
The tests ran for many days and a couple of
formulars had to be re-adjusted to better match the
partner’s requirements. We were reactive and could
almost immediately update the formulars and they
were able to use it straight away. As the tests were
made on a real Importexa order, we could monitor
data incoming from all steps including all chemicals
and certificates (GOTS, Fairtrade, GMO free, and so
on) in realtime.
5 DISCUSSION
Fairtrace is regarded as a success on management,
collaboration and technology support levels. The
project encountered high commercial concrete
Fairtrace-ASemantic-webOrientedTraceabilitySolutionAppliedtotheTextileTraceability
43
Figure 4: Test data acquisition in India.
interest when it was presented and demonstrated
in a technical session to potential customers. One
customer is ready to deploy the solution for his
supply-chains. The project plans to be integrated
in a start-up with the help of the CTI Start-up
9
, an
important Swiss coaching institution. Although we
had no quantitative metrics for the tests, we could
assess on its success in a real scenario. The testing
in India has faced many unexpected real-world
conditions that could not be foreseen. We were able
to respond to problems and modification requests
quickly (in a matter of minutes) and could attest of
the flexibility of the prototype.
Importexa is now heading to the creation of a
start-up that will commercialize the finished product.
Fairtrace is going be improved and adapted to market
segments other than the textile industry. A survey
on potential markets has already been done and first
discussions have already taken place with interested
brands and Swiss resellers, in order to adapt the tool to
brand new markets (clock industry, controlled cheese
production, cocoa, ...). Supply-chain traceability can
be a strong selling argument, because it implies that
companies master the entire production chain, can
react quickly to potential problems (such as to follow
the REACH regulation) and can prove their customers
that they exactly know what they produce.
In this project the use of Semantic Web
Technologies was a prerequisite, because it was a
good opportunity to explore its possibilities for future
projects. We were already a bit familiar with the
tools and techniques, but not sufficiently aware of its
benefits or complexities to apply it in future projects.
We had to work out many issues and carefully think
about our solutions. Still, Fairtrace proved that SWT
are safe for use in commercial projects, because they
9
http://www.ctistartup.ch
are now sufficiently mature. We kept things very
simple at the moment, because we had a real need
for pragmatism, so we only had a glimpse at the full
range of possibilities that Semantic Web has to offer.
Our model was not thought in terms of inferences it
could entail, but rather in terms of object orientation.
We needed to translate RDF data into Java classes and
thereby lost part of the expressiveness power of these
technologies. To compensate for this, the prototype
relied heavily on queries. The lack of SPARQL 1.1
UPDATE support has also somewhat hindered our
efforts. Programmatically inserting/updating/deleting
triples in the store was difficult, verbose and error
prone.
Although we did only use a small subset of
it, Semantic Web Technologies still provided a
real competitive advantage, because they allowed
a greater modeling flexibility in comparison to
other database technologies. Translating the logical
models to RDF Schema was simple and quick.
The expressiveness of the language coupled to
the rule-based OWLIM reasoning engine was good
enough to allow sufficiently complex constructs.
There is still a lot we can do to improve the
current state of things: we can link geographical
places to GeoNames
10
features for instance, redirect
specific concepts to DBpedia
11
definitions or even
automatically categorize items as dangerous by
making use of bridges to external datasets. In next
iteration of Fairtrace, we’ll definitely investigate on
integrating OWL features like (inverse) functional,
symmetric and invese properties, as well OWL2
property chain axioms vs. creating our own rules, also
on integrating geo-spatial extensions (for marking
places) to reason about distances and, of course, the
new SPARQL 1.1 constructs.
6 CONCLUSIONS
This paper presented a pragmatic semantic-based
research solution to the traceability problem in
supply-chains. The first prototype is finished
and has been successfully tested in a Swiss
and Indian supply-chain in the textile industry.
However, Fairtrace is still an ongoing work and
the roadmap contains many improvements. SWT
will continue playing a central role in the future
infrastructure. We are planning to use the
next-generation OWLIM semantic repositories and
take advantage on the new constructs offered by
10
http://www.geonames.org
11
http://dbpedia.org
ICEIS2013-15thInternationalConferenceonEnterpriseInformationSystems
44
OWL2 and SPARQL 1.1 UPDATE. We are also
planning important improvements on the creation
of formulars, on the customization of the system
and on the primitives for modeling supply-chains.
Finally, we consider intelligent agent techniques for
automatically queryingthe semantic repositories to
help the decision support of users.
ACKNOWLEDGEMENTS
We especially thank the Swiss Commission for
Technology and Innovation
12
who financed a big part
of this project under contract number PFES-ES No.
11141.1, and the company Ontotext who supported
us with a research license for BigOWLIM.
REFERENCES
Bechini, A., Cimino, M. G. C. A., Marcelloni, F.,
and Tomasi, A. (2008). Patterns and technologies
for enabling supply chain traceability through
collaborative e-business. Inf. Softw. Technol.,
50(4):342–359.
Bowersox, D., Closs, D., and Cooper, M. (2002). Supply
chain logistics management, volume 2. McGraw-Hill
New York.
Chandra, C. and Kumar, S. (2001). Enterprise architectural
framework for supply-chain integration. Industrial
Management and Data Systems, 101(6):290–304.
Christopher, M. (2005). Logistics and supply chain
management: creating value-added networks.
Pearson education.
Datta, S., Lyu, J., and Ping-Shun, C. (2007). Decision
support and systems interoperability in global
business management. International Journal of
Electronic Business Management.
Fox, M., Barbuceanu, M., and Teigen, R. (2000).
Agent-oriented supply-chain management.
International Journal of Flexible Manufacturing
Systems, 12(2):165–188.
Grubic, T. and Fan, I.-S. (2010). Supply chain ontology:
Review, analysis and synthesis. Comput. Ind.,
61(8):776–786.
Haller, A., Oren, E., and Kotinurmi, P. (2006). m3po:
An ontology to relate choreographies to workflow
models. In Services Computing, 2006. SCC’06. IEEE
International Conference on, pages 19–27. IEEE.
K
¨
arkk
¨
ainen, M. (2003). Increasing efficiency in the
supply chain for short shelf life goods using
rfid tagging. International Journal of Retail &
Distribution Management, 31(10):529–536.
Kelepouris, T., Pramatari, K., and Doukidis, G. (2007).
Rfid-enabled traceability in the food supply
12
http://www.kti-cti.ch
chain. Industrial Management & Data Systems,
107(2):183–200.
Kim, J. W. and Jain, R. (2005). Web services composition
with traceability centered on dependency. In
Proceedings of the Proceedings of the 38th Annual
Hawaii International Conference on System Sciences
(HICSS’05) - Track 3 - Volume 03, HICSS ’05, pages
89–, Washington, DC, USA. IEEE Computer Society.
Kwok, S. K. and Wu, K. K. W. (2009). Rfid-based
intra-supply chain in textile industry. Industrial
Management and Data Systems, 109(9):1166–1178.
Lambert, D. and Cooper, M. (2000). Issues in supply chain
management. Industrial marketing management,
29(1):65–83.
Lambert, D., Cooper, M., and Pagh, J. (1998). Supply chain
management: implementation issues and research
opportunities. International Journal of Logistics
Management, The, 9(2):1–20.
Mentzer, J., DeWitt, W., Keebler, J., Min, S., Nix, N.,
Smith, C., and Zacharia, Z. (2001). Defining supply
chain management. Journal of Business logistics,
22(2):1–25.
Mocan, A., Moran, M., Cimpian, E., and Zaremba, M.
(2006). Filling the gap - extending service oriented
architectures with semantics. In Proceedings of the
IEEE International Conference on e-Business
Engineering, ICEBE ’06, pages 594–601,
Washington, DC, USA. IEEE Computer Society.
Opara, L. (2003). Traceability in agriculture and
food supply chain: a review of basic concepts,
technological implications, and future prospects.
Journal of Food Agriculture and Environment,
1:101–106.
Paolucci, M., Kawamura, T., Payne, T., and Sycara,
K. (2002). Semantic matching of web services
capabilities. The Semantic WebISWC 2002, pages
333–347.
Schumacher, M., Helin, H., and Schuldt, H. (2008).
CASCOM: Intelligent Service Coordination in the
Semantic Web. Birkhäuser Basel, 1st edition.
Virgilio, R., Sciascio, E., Ruta, M., Scioscia, F.,
and Torlone, R. (2010). Semantic-based rfid
data management. In Ranasinghe, D. C. C.,
Sheng, Q. Z. Z., and Zeadally, S., editors,
Unique Radio Innovation for the 21st Century,
pages 111–141. Springer Berlin Heidelberg.
10.1007/978-3-642-03462-6˙6.
Werlen, D., Alves, B., Cherix, G., Gapany, C., and
Schumacher, M. (2012). Proc
´
ed
´
e d’
´
edition de
formulaires pour la saisie de donn
´
ees en diff
´
erents
points de collecte. Swiss patent application
CH00104/12 filed on Jan. 25, 2012.
Ye, Y., Yang, D., Jiang, Z., and Tong, L. (2008a).
An ontology-based architecture for implementing
semantic integration of supply chain management. Int.
J. Comput. Integr. Manuf., 21(1):1–18.
Ye, Y., Yang, D., Jiang, Z., and Tong, L. (2008b).
Ontology-based semantic models for supply chain
management. The International Journal of Advanced
Manufacturing Technology, 37(11):1250–1260.
Fairtrace-ASemantic-webOrientedTraceabilitySolutionAppliedtotheTextileTraceability
45
