Documente Academic
Documente Profesional
Documente Cultură
discussions, stats, and author profiles for this publication at: https://www.researchgate.net/publication/228353886
CITATIONS READS
224 2,963
3 authors, including:
Some of the authors of this publication are also working on these related projects:
Build artificial intelligence for Wind Farm Health Management from Data View project
All content following this page was uploaded by A. Kusiak on 18 June 2015.
M. Shahbaz
e-mail: m.shahbaz@lboro.ac.uk
Data Mining in Manufacturing:
Srinivas
A Review
e-mail: s.srinivas@lboro.ac.uk
The paper reviews applications of data mining in manufacturing engineering, in particu-
Wolfson School of Mechanical and lar production processes, operations, fault detection, maintenance, decision support, and
Manufacturing Engineering, product quality improvement. Customer relationship management, information integra-
Loughborough University, tion aspects, and standardization are also briefly discussed. This review is focused on
Loughborough, Leicestershire LE2 4LA, UK demonstrating the relevancy of data mining to manufacturing industry, rather than dis-
cussing the data mining domain in general. The volume of general data mining literature
makes it difficult to gain a precise view of a target area such as manufacturing engineer-
A. Kusiak ing, which has its own particular needs and requirements for mining applications. This
Department of Mechanical and Industrial review reveals progressive applications in addition to existing gaps and less considered
Engineering, areas such as manufacturing planning and shop floor control. DOI: 10.1115/1.2194554
The University of Iowa,
Iowa City, IA 52242-1527
e-mail: andrew-kusiak@uiowa.edu
Journal of Manufacturing Science and Engineering NOVEMBER 2006, Vol. 128 / 969
Copyright 2006 by ASME
of data mining system integration. Finally, the conclusions and The CRISP-DM and SEMMA methodologies are most widely
future research directions outline the progress made by the ongo- used by the data mining community. CRISP-DM and SEMMA
ing research related to manufacturing control and quality improve- provide a step by step guide for data mining implementation.
ment. CRISP-DM is easier to use than SEMMA as it provides detailed
neutral guidelines that can be used by any novice in the data
mining field. SEMMA is developed as a set of functional tools for
SASs Enterprise Miner software. Therefore those who use this
2 Data Mining Models for Manufacturing Applica- specific software for their tasks are more likely to adopt this meth-
tions odology. Using SEMMA, results may be found quickly by mining
CRISP-DM TM Cross Industry Standard Process for Data Min- samples of data from the whole database, but if the discovered
ing, SEMMA, SolEuNet Data Mining and Decision Support for relationships do not follow in the whole database then new
Business Competitiveness: A European Virtual Enterprise, Kens- samples must be examined which means repeating the whole data
ington Enterprise Data Mining Imperial College, Department of mining process.
Computing, London, UK, and Data Mining Group (DMG) have The details of each step of CRISP-DM 10 make it a reliable
established methodologies and developed languages and software methodology that is easy to use and fast to implement. The de-
tools for the standardization of industrial applications of data min- tailed sub-stages are optional guidelines and can be skipped as
ing. However, most products focus on the implementation of data required. The CRISP-DM methodology can therefore be fully or
mining algorithms and application development rather than on the partially adopted depending on the problem and its requirements.
ease-of-use, integration, scalability, and portability. Most pub- This was the main reason that the CRISP-DM was used by the
lished research on data mining in manufacturing reports dedicated authors as a standard guide to implement the data mining research
applications or systems, tackling specific problem areas, such as that is reported in Ref. 11.
fault detection see Fig. 1. Only limited research has been done to Neaga and Harding also presented a framework for the integra-
address the integration of data mining with existing tion of complex enterprise applications including data mining sys-
manufacturing-based enterprise reference architectures, frame- tems 12,13. The presented approaches provide the definition and
works, middleware, and standards such as Common Object Re- development of a common knowledge enterprise model, which
quest Broker Architecture CORBA, Model-Driven Architecture represents a combination of previous projects on manufacturing
MDA, or Common Warehouse Metamodel CWM. Neaga 8 enterprise architectures and Object Management Group OMG
explained the neglect of these issues and highlighted their impor- models and standards related to data mining.
tance and the investments already committed to the existing ef-
forts for enterprise integration. Neaga and Harding 8,9 presented
a holistic approach to a wide range of data mining applications 3 Data Mining Applications Relevant to Manufactur-
suitable for manufacturing enterprises. The areas of manufactur-
ing enterprise design, engineering and re-engineering, information ing
modeling and the suitability of applying data mining techniques to This section details the contributions of researchers and practi-
use previous knowledge and information about an enterprise are tioners in different areas of manufacturing from the late 1980s to
examined in Refs. 8,9. date. The literature was searched extensively in different journals,
3.1 Engineering Design. Engineering design is a multidisci- 3.2 Manufacturing Systems. Data collection in manufactur-
plinary, multidimensional, and non-linear decision-making pro- ing is common but its use tends to be limited to rather few appli-
cess where parameters, actions, and components are selected. This cations. Machine learning and computational intelligence tools
selection is often based on historical data, information, and provide excellent potential for better control of manufacturing
knowledge. It is therefore a prime area for data mining applica- systems see Fig. 3, especially in complex manufacturing envi-
tions and although as yet only a few papers have reported appli- ronments where detection of the causes of problems is difficult.
cations of data mining in engineering design see Fig. 2, this has Piatesky-Shapiro et al. 7 argued that the data mining industry is
been an area of increased research interests in recent years. These coming of age. However, this review of data mining in manufac-
recently published papers form an important part of this review turing shows that although there are several areas in manufactur-
paper due to the essential synergies between design and manufac- ing enterprises that have benefited from data-mining algorithms,
turing. The importance of considering how a product should be there are still numerous areas that could benefit further 24. In
manufactured during the design stage and the constraints imposed manufacturing environments the need and importance of data col-
on design by particular manufacturing processes and technologies lection is ever present for statistical process control purposes. Lee
have been accepted for many years. There is indeed great potential 5 discussed and suggested several principles leading to a
for data mined knowledge to integrate manufacturing, product knowledge-based factory environment utilizing the data collected
characteristics, and the engineering design processes. over several stages of the manufacturing-related processes. A
Sim and Chan 14 developed a knowledge-based system for comparative study of implicit and explicit methods to predict the
the selection of rolling bearings. They used heuristic knowledge non-linear behavior of the manufacturing process, using statistical
supported by a manufacturers catalogue to optimize design speci- and artificial intelligence tools, was discussed by Kim and Lee
fications by matching the temporal data of the new product against 25.
the knowledge base. Kusiak et al. 15 proposed a rough-set Semiconductor manufacturing is complex and faces several
theory approach to predict product cost. Ishino and Jin 16 used challenges relating to product quality, scheduling, work in pro-
data mining for knowledge acquisition in design from the data cess, cost reduction, and fault diagnosis. To overcome these prob-
obtained through observing design activities using a CAD system. lems several methods and systems have been developed, e.g.,
They developed a method called Extended Dynamic Program- Rule-Based Decision Support Systems RBDSS 26, CAQ 27,
ming to extract the knowledge. Romanowski and Nagi 17 pro- Knowledge Acquisition from Response Surface Methodology
posed a design system which supports the feedback of data mined KARSM, and GID3 6 or generalized ID3, a decision-tree al-
knowledge from the life cycle data to the initial stages of the gorithm for fault diagnostics and decision making have been de-
design process. Giess et al. 18,19 mined a manufacturing and veloped and used. Gardener and Bieker 28 showed a substantial
assembly database of gas turbine rotors to determine and quantify savings in the manufacture of semiconductors by applying
relationships between the various balance and vibration tests and decision-tree algorithms and neural networks to solve the yield
highlight critical areas. This knowledge could then be fed back to problem in the wafer manufacture. Sebzalli and Wang 29 applied
the designers to improve tolerance decisions in the future design principal component analysis and fuzzy c-means clustering to a
of components. They used a decision tree at the initial stage to refinery catalytic process to identify operational spaces and de-
determine appropriate areas of investigation and to identify prob- velop operational strategies for the manufacture of desired prod-
lems with the data. At the next stage, a neural network was used to ucts and to minimize the loss of product during system change-
model the data. Hamburg 20 applied data mining techniques to over. Four operational zones were discovered, with three for
support product development by analyzing global environment as- product grade and the fourth region giving high probability of
pects, market situation, strategy, philosophy, and culture of the producing off-specification product. Lee and Park 30 used self-
organizing maps and Last and Kandel 31 applied information
fuzzy networks for quality checks and extracted useful rules from
1
http://citeseer.ist.psu.edu/cs their model to check the quality of the products. Kusiak 32
Journal of Manufacturing Science and Engineering NOVEMBER 2006, Vol. 128 / 971
Journal of Manufacturing Science and Engineering NOVEMBER 2006, Vol. 128 / 973
Journal of Manufacturing Science and Engineering NOVEMBER 2006, Vol. 128 / 975
DownloadedViewFrom:
publicationhttp://manufacturingscience.asmedigitalcollection.asme.org/
stats on 06/18/2015 Terms of Use: http://asme.org/terms