KDAM Consortium
Knowledge
Discovery
and
Analysis
in
Manufacturing
Acknowledgement: Much of the content of this document has been
derived from the work of Professors Marcin Perzyk and Andrzej Kochański,
Institute of Materials Processing, Faculty
of Production Engineering, Warsaw
University of Technology.
For further
information on KDAM, please contact: Dr. Mark Polczynski, College of
Engineering, Marquette University, mark.polczynski@marquette.edu
Goal and Mission
Increasingly powerful and available computer and
communication capacity provides an ever-expanding sea of data that can be used
to improve the capabilities of a wide variety of processes. Furthermore and fortunately, this growing
computer/communication capacity also supports a growing base of technologies
for analyzing and utilizing this data. Typically
aggregated under terms such as data mining, machine learning, and knowledge
discovery in data (KDD), these approaches have made significant contributions
in business and social applications.
However, application of these technologies to manufacturing processes
has received less attention.
The goal of the Knowledge
Discovery and Application in Manufacturing (KDAM) Consortium is
to:
·
Provide a
forum for the free and frequent exchange of data mining and machine learning
technology…
·
And
technology application results and finding…
·
To rapidly
and effectively improve the capabilities and efficiencies of manufacturing
processes…
·
And the
quality and value of manufactured products.
Based on this ambitious goal, the mission of KDAM is to
create and maintain a non-competitive industry and academic community of
interest to:
- Identify and develop technologies
suitable for achieving the KDAM goal, including technology
reduction-to-practice in manufacturing environments;
- Identify and classify particular
manufacturing processes that have proven to benefit from or could benefit
from the application of specific KDAM-related technologies;
- Share technology developments and
communicate application results and findings among Consortium partners,
including application best/worst practice guidelines.
- Where possible, share databases among Consortium
members, thereby enabling development, verification, and comparison of
KDAM-related technologies.
Consortium Scope
The scope of KDAM activities conforms to the general scope
of data mining, machine learning, and related technologies and fields:
- Identification of patterns in data that
support predictions of future conditions and behaviors;
- Discovery of structures underlying the
data that reveal fundamental relationships among systems and system
elements.
Within the general field of engineering, there are a wide
range of applications that could benefit from work in these two areas. In order to provide a manageable scope for
KDAM work, Consortium applications focus on primary manufacturing operations,
e.g. molding, casting, forging, plating, painting, metal forming, etc. Of course, KDAM partners will pursue other
applications of the technologies included in KDAM activities. This application range is provided primarily
to stimulate specific areas of Consortium cooperation.
Consortium Focus
There are a wide range of improvement areas that can benefit
from KDAM-related technologies. In order
to provide additional focus to KDAM activities, KDAM efforts are directed
primarily at:
1.
Detection
of root causes of deteriorating product quality,
2.
Identification
of optimal and critical process parameters and prediction of results of
manufacturing process changes,
3.
Identification
of root causes and prediction of equipment breakdown.
Naturally, the research and application focus of KDAM
partners will extend beyond these areas of focus. The KDAM focus is provided primarily to
stimulate specific areas of Consortium cooperation.
Technology Scope
Referencing the three KDAM focus areas, the technologies
employed by KDAM provide three general functions:
- Regression: Defining functional
relationships between outputs of interest and multiple and possibly
dependent inputs. An example is
predicting the dimension of a plastic molded part given typical ranges of
molding process variables.
- Classification:
Grouping of objects (products) into classes given previously known
input/output (process/product) associations. An example is association of acceptable
and defective parts (two classifications) with the particular production
conditions under which the parts were manufactured.
- Clustering: Grouping of
objects (products) by characteristics where there exist no previously
known associations. An example is
discovering that a particular employee operating a particular machine
tends to produce parts with dimensions on the high side of the specified
value.
Technology Focus
A variety of specific technologies can be applied to
KDAM-related applications. Examples of
traditional technologies commonly used include linear and polynomial
regression, and analysis of variance (ANOVA).
The KDAM focuses on emerging technologies that have demonstrated
successful results in manufacturing applications. Technologies considered initially for KDAM
include:
It is anticipated that the scope of KDAM technologies will
become more limited as findings and results are generated and most suitable
technologies are identified. As with the
KDAM focus areas, technology research of KDAM partners will extend beyond the Consortium’s
technology focus. This KDAM technology
focus is provided primarily to stimulate specific areas of Consortium
cooperation.
Process Model
A primary goal of KDAM is sharing of results and
findings. This goal is supported by the
use of a standard process model for planning and executing KDAM-related
projects. Where appropriate,
KDAM-related work will utilize the CRISP-DM “Cross Industry Standard Process
for Data Mining” (www.crisp-dm.org) to
plan, conduct, and report project results.
CRISP-DM is industry, tool, and application neutral, is non-proprietary
and freely available, and is internationally recognized.
KDAM also encourages the sharing of databases where
possible. To enable this sharing, KDAM
partners will be encouraged to use ARFF
Attribute-Relation File Format for data exchange where feasible.
Consortium Partners
The Consortium consists of industry partners directly
involved in manufacturing activities that benefit from KDAM technologies, and
academic partners developing and applying new solutions to industry needs. The basic KDAM model is for academic partners
to link directly with industry partners, and also to link directly with other
academic partners. The academic partners
form the hubs of industry partner clusters, with the academic partners forming
the “glue” for an extended KDAM industry “community of interest”.
An objective of the Consortium is to provide close geographical
coupling between universities and their industry partners, thus allowing direct
hands-on investigation of problems and application of solutions. In order to address the global nature of the
problems and solutions pursued, international partnerships among academic
partners is emphasized. These academic partnerships
enable global resource sharing and support international channels for
communicating findings and results among partners.
KDAM Funding Model
Funding of KDAM-related activities is the responsibility of
the partners directly involved in activities.
KDAM will neither solicit nor provide funds to Consortium partners. There are currently no membership fees
associated with participation in KDAM.
In some cases, it may be beneficial for Consortium partners
to share costs for certain resources (e.g., software license). On a case-by-case basis, the Consortium can
serve as a vehicle for enabling cost-sharing activities among partners, but all
matters of this nature are the responsibility of the partners, not the
Consortium.
A wide variety of funding agencies support the type of
research to be pursued by KDAM.
Membership in KDAM can strengthen partner funding proposals to these
agencies. KDAM partners are strongly
encouraged to pursue joint funding proposals.
Confidentiality and Intellectual Property
In order to support free sharing of data, technology
developments, and application results and findings among Consortium partners,
academic partners will be responsible for selecting industry partners which are
non-competitors.
In some cases, academic partners will desire to publish
results of projects conducted with industry partners. In these cases, industry partners will
determine what sensitive information is not appropriate for publication by
academic partners.
In some cases, KDAM activities may result in the development
of proprietary technology. It is
suggested that KDAM industry and academic partners establish an understanding
of intellectual property rights before significant project work is
performed. This understanding will
include intellectual property ownership, ownership and publication of data
shared and/or generated, and application and publication of proprietary
technology details.
Outreach
While the goal of KDAM is to create a closely-coupled
community of interest, there are clear benefits to sharing non-confidential
information outside of the Consortium beyond academic publications. KDAM will maintain a simple website to
provide a portal to relevant KDAM-related information. The web site (www.technologyforge.net/KDAM)
will be maintained by the Marquette University consortium partner.
Consortium Partnership Benefits
The preceding outline presents an overview of what the KDAM
Consortium is and how it operates. Based
on this description, the following specific benefits to industry and academic
partnership accrue:
Industry:
- Development
of real solutions to real manufacturing problems,
- Access to
new and potentially superior solutions,
- Access to
a network of non-competing industrial partners seeking and developing
solutions to similar problems,
- Creation
of massive (sanitized) databases suitable for cross-industry data mining.
Academic:
- Access to
potentially useful technology developed by Consortium partners,
- Access to
experts in similar and complementary technologies,
- Shared
databases (sanitized) for developing and testing new technologies and
solutions,
- Opportunity
for stronger funding proposals.
Current Status
At this point, potential academic Consortium partners are
being contacted to determine feasibility of the concepts provided here and to
provide inputs which are being integrated into this document. Regarding initial KDAM activities,
three areas are being investigated:
- Create a
bibliography of references that describe actual successful applications of
the technologies cited in the Technology
Focus section, and cross-referenced with the areas cited in the KDAM Focus and Technology Scope sections.
This bibliography would also be cross-referenced according to the
specific manufacturing process applications. For example, a particular reference
might be cross-referenced as: ANN, classification, product quality,
foundry casting.
- Determine
high-priority focus areas for industry partners and correlate with the
bibliography cited in the previous point.
- Determine
what data analysis software/tools are currently being used by partners,
and investigate the potential to share software.
Contact
the Webmaster