0% found this document useful (0 votes)
38 views

Manufacturing Process Optimization Using Data Mining Techniques

Manufacturing, being referred to, as an integrated management process has been radically transformed with the advent of latest technologies and advancements. Data generated by machines are continuously encapsulated and archived in storage systems for analysis and futuristic applications.Such storage necessitates enormous warehouse capacities thus vitalizing methods deployed for storage. Apart from vitality, they serve as critical parameters in improving efficiency, as well. With regard to engine

Uploaded by

IIR india
Copyright
© © All Rights Reserved
Available Formats
Download as DOCX, PDF, TXT or read online on Scribd
0% found this document useful (0 votes)
38 views

Manufacturing Process Optimization Using Data Mining Techniques

Manufacturing, being referred to, as an integrated management process has been radically transformed with the advent of latest technologies and advancements. Data generated by machines are continuously encapsulated and archived in storage systems for analysis and futuristic applications.Such storage necessitates enormous warehouse capacities thus vitalizing methods deployed for storage. Apart from vitality, they serve as critical parameters in improving efficiency, as well. With regard to engine

Uploaded by

IIR india
Copyright
© © All Rights Reserved
Available Formats
Download as DOCX, PDF, TXT or read online on Scribd
You are on page 1/ 4

Integrated Intelligent Research (IIR) International Journal of Data Mining Techniques and Applications

Volume: 04 Issue: 01 June 2015 Page No.15-18


ISSN: 2278-2419

Manufacturing Process Optimization using Data


Mining Techniques
R Shankar1, M Sundararajan2
Research Scholar, Computer Science, Bharathiar University, Coimbatore,Indai
1

2
Asst. Professor, Dept. of Computer science,Govt. Arts and Science College for Men,Nandanam, Chennai
Email:[email protected], [email protected]

Abstract - Manufacturing, being referred to, as an integrated which is the use of DM techniques in the process data generated
management process has been radically transformed with the in an automotive engine manufacturing system and the volume
advent of latest technologies and advancements. Data of real time data being generated and stored in data repositories.
generated by machines are continuously encapsulated and Section III explains the schema of an automotive engine
archived in storage systems for analysis and futuristic manufacturing system and the rough estimate of data that are
applications.Such storage necessitates enormous warehouse generated during the process of testing of the engines before
capacities thus vitalizing methods deployed for storage. Apart assembling them in to the vehicles. The details about the 2 steps
from vitality, they serve as critical parameters in improving involved during the analysis of the data usingclustering
efficiency, as well. With regard to engine assembly plants, technique has been explained in section IVand finally the
these data can comprehend subtle characteristics of the engine section IV concludes the analysis activity.
assembly and further their testing processes, as well.Simply
put, each engine assembly comprises a minimum of 50 major
II. LITERATURE SURVEY
assembly process parameters and 15 pivotal testing parameters
meant for observation and record. Of course, they are of great
Data mining for manufacturing and service applications is
significance to restore lineage and traceability.Competition
discussed by KUSIAK in his paper [5]. In this paper basic
demands the use of such legacy data repositories of automotive
concepts of machine learning and data mining are introduced.
engine manufacturing companies owing to market demand.
Machine learning algorithms supports yielding knowledge from
Hence trigger the need to deploy appropriate techniques to
diverse data bases that can be used to build decision-making
secure business insights for effective decision-making.Data
systems. For example, based on the operational engineering
Mining (DM) is one such subject that mandates a thorough and
data, equipment faults can be detected, the number of items to
detailed study. Here in this paper it is handled with the
be ordered can be predicted, optimal control parameters can be
clustering analysis of the subject DM with a sample data
determined.Same way, the use of DM in improving the
volume of five hundred engines’ performance test result files.
manufacturing Quality has widely been discussed by LIOR
ROKACH and ODED MAIMON in their paper. Here they
Keywords:Clustering Analysis, Manufacturing
emphasize the importance and the necessity of DM techniques
management,Automotive Engine testing Process.
towards the identification of the hidden patterns of the abundant
information and data which are produced during manufacturing
I. INTRODUCTION activity. They specifically focus on mining the quality related
data by the useofclassifier models towards the achievement of
Minim ax, there exist two requirements necessitating improvement in manufacturing quality [7].The need of DM
maintenance of manufacturing process data in an automotive techniques in identifying the inherent interdisciplinary character
engine manufacturing organization, via, of the automobile development and manufacturing process has
been apparently analyzed by RUDOLF KRUSE, MATHIAS
 Government regulations aligning with customerreservations; STEINBRECHER and CHRISTIAN MOEWES in their paper
 Demand for improvement of critical parameters that drive which emphasizes the use of DM methodologies in automotive
customer expectations! industries [8]. The authors reiterate the use of DM tools towards
getting a valuable means to manage the process which produces
The purpose of this paper is to justify the utility value of DM overwhelmingly large information during the manufacturing
models in consuming process parameter data, amassed for process in an automotive industry.
extrapolating events, for applying control measures, for
simulating trials of control range and for pursuing span of III. ASSEMBLY AND TESTING PROCESS OF AN
control.In the context of presenting this paper, the objective is AUTOMOTIVE ENGINE
to draw inference from cluster sets drawn against scatter graphs,
plotted of data sets from process parameter repositories, to Fig.1 represents the schema of Engine Manufacturing System.
bring out the hidden pattern of Specific Fuel Consumption Consequent to process time complexities, DM techniques
(SFC) of engines for differential torque valuesusing clustering surpassed other models meant for effective and efficient use of
analysis of DM from a selected volume of 500 datasets. The data collected over substantial periods.Performance starts with
remaining of this paper is organized as follows, Section II the mandate to conform to theoretical design specifications.
reveals the relevant literature inline with concept of this paper Though, this depends on various factorsas supplier parts,

15
Integrated Intelligent Research (IIR) International Journal of Data Mining Techniques and Applications
Volume: 04 Issue: 01 June 2015 Page No.15-18
ISSN: 2278-2419
handling of materials, equipment etc., and the most blatant fact contention that enterprises world-wide, use just 10% of data
remains that quality of the final product is gauged over real- compiled for analytics and decision-making. What is baffling is
time performance.Testing, though being performed in the end, it this insensitivity to corporate data assets that sail through time,
still endorses the product on its availability to the customer, uncared for, remaining as non-performing assets. This is where
promising on its delivery commitments. Testing qualifies the DM adds vitality.Most manufacturing processes running
engine for delivery and sale, thereof. The number of production process control systems deploy various types of sensors and
Units per Hour (UPH) / Jobs per Hour (JPH) in the assembly feedback devices to track parameters such as pressure, flow
lines span between15 - 60 UPH / JPH. Most certainly, this rate, temperature, variables and factors of efficiency needed to
conditions the manufacturer to maintain data volumes ranging exert control over processes [6]. When collected, these, as huge
between 1000 - 4000 Data Sets / Hour [1]. This obviates data repositories, are meant for proper and purposeful
collection of about 16,000 - 64,000 Data-sets / Day and expands utilization for better performance.What suits this scenario
to about 48, 00,000 - 1, 92, 00,000 Data-sets / Year. This is fittingly is DM clustering analysis.
simply huge.As the collection of data from the testing of
engines by real time basis during their manufacturing from the IV. THE CLUSTERING TECHNIQUE AND
production houses lies large, this paper encourages the ANALYSIS
DMpathway.Quoting SAP America Inc., [2] this paper brings to

The process of partitioning sets of data or objects into engines have been corroborated. Applying the
sets of meaningful sub-classes begets clusters [3]. clustering technique of DM, data-sets thus obtained
These throw light on the nature of grouping and the during Engine Performance Testing for every single
resultant structure in data-sets.This paper examines, a engine on Power, Torque, SFC, Temperature etc., have
sample volume “N”, being test result data-sets of 500 been tabulated as shown in the Table: 1.This is for
engines which are ASCII files generated as test results inference and improvement.Factors, for this paper, for
from the Engines Dynamometer testing equipment. the study of deployment of clustering are SFC and
N  500 Torque. Results obtained as multitudes of specific SFC
Dynamometer testing equipment is special testing values against particular Torque ratings over clusters
equipment in automotive testing process of engines are comprehended. Even visual inspection reveals the
which applies load to the engines during its running intensity range of clusters where SFC scores are in
conditions as specified in Pass-off standards given in maximum level against certain Torque numbers, on
the engine design specifications. During the process of clustering.This enables to identify correct Torque
applying load to the engines the resultant values for which SFC remains high. This extends in
characteristics are collected as data-sets against the establishing appropriate process design for better SFC
engine serial number and stored in repositories mainly production.
for genealogy and traceability purposes.Real-time
data-sets (ASCII files) gathered of manufactured

Table 1: Data table view

Why Torque? Torque, for commercial vehicle, is the process design be made robust.Using tabulated values as shown
determining factor for serving horse power to engine. More in Table: 1.A sample cluster has been obtained using DM which
horse power implies better load-bearing capacity, speed. The is explained below in steps 1 and 2.Step- 1-Exploring the data
experiment hence determines the range over which SFC reaches from Table: 1.Graphically plotting, SFC values along Y axis
its maximum.By modifying the assembly process settings, it is and the respective engine numbers in X axis, we have
possible to manufacture engines by delimiting performance Fig:2.thusarrived.Interpretation of results obtained during
variances over Torque and SFC values. What is fitting for SFC testing as process parameter data-sets:-The green line on the
and Torque can be extrapolated to befit other critical graph represents the safe limit of SFC which is 233 g / KWh.If
parameters, as well for more improvements. In doing so, such SFC value is  233 g / kWh at a fixed engine speed of 2200
deviations, limitations and variances can be leveraged and rpm, then the process is conforming to engine process design
16
Integrated Intelligent Research (IIR) International Journal of Data Mining Techniques and Applications
Volume: 04 Issue: 01 June 2015 Page No.15-18
ISSN: 2278-2419
specifications; this is traced with the help of green colour arrow
in the graph.The overall trend reveals that almost 99.4% of
engines are within specified limits of SFC with regard to engine
process design specification [8].However, the customer survey
conducted as post sales feedback revealed that 30% of
customers are not happy with the performance with the product
in terms of fuel economy of the vehicle which is contrary to
the obtained test result shown in Fig: 2.This is where this paper
C4 Engines with Nominal SFC and Nominal Torque
builds logic and substance for improvement towards the data
C5 Engines with Nominal SFC and Good Torque
analysis methodology.

Figure 1

To isolate areas of concern from the rest and to distinguish each


such collected set of data based on performance, in most
manufacturing problems, it is necessary to project the
summarized data in descriptive terms along with overall
picture.However, due to limited linkage with real-time data, the
uses of tools for all purposeful decision-making processes have
been minimal, in practice [7]. This has been whipping and Figure 3:Clustered SFC - Torque relation Data
constraint-bound, in the past.DM has weeded out such
inadequacies and has made deep inroads. Putting DM to From the worst to the best these clusters may be sequenced into
practice, business decisions are implemented [4]. These have C4, C5, C3, C1 and C2Which is for further analysis as
proven to be quite informed, transparent and autonomous.Step - mentioned in Fig: 4. Firstly, this paper warms up DM into
2 -Compounding relevance of attributes, suffice to their Engine Design Process Parameter Settings and extends into
differences, clustering has been computed on the design process deploying clustering for study. Secondly, the paper infers the
parameter data-set tabulated above with different sets of Torque contour, over which influences of Torque and SFC are realized.
values ending with different volumes of SFCs, but at the same As a third step, this paper, enables conditions for optimality
rpm.In other words, the experiment has been repeated for with regard to fuel consumption, isolating constraints.As a
different Torque values at the same rpm for all 500 engines and result, a comprehensive and conclusive analysis towards the
results captured. Fig.3 represents SFC - Torque relation graph final solution can be taken up for further study. The authors are
with identified clusters.From the identified clusters, it has been working on the same data-sets, applying Association Rule
inferred that results mainly fell into 5 major categories, Mining Technique, one more of DM Application to analyze the
discarding anomalies that ended up beyond the design associated parts of the engine to move closer to arrive at the
specifications of torque and SFC conditions of optimality [5]. As a result, a comprehensive and
Specifications: conclusive analysis towards the final solution can be taken up
Torque ≥ 150 Nm for further study. The authors are working on the same data-
SFC ≤ 233 g / kWh sets, applying Association Rule Mining Technique, one more of
DM Application to analyze the associated parts of the engine to
move closer to arrive at the conditions of optimalitywith regard
to process design parameter settings for optimal consumption of
fuel at specific values of control parameters.

Figure 2: Graph of SFC results of tested engines


The five major clusters are classified as,
C1 Engines with Very good Torque and Good SFC
C2 Engines with Very Good SFC and Good Torque Figure 4:Clustered SFC - Torque relation Data
C3 Engines with Good Torque and Good SFC
V. CONCLUSION

Some trivial states of production results revealed especially


when it is dealt with real-time reports of“engine performance
process test result analytics” where the Process capability
constant(Cpk) establishment becomes very tough and
critical.Therefore, theoretical events may be simulated towards
achieving the same by using Data-Mining techniques. This
17
Integrated Intelligent Research (IIR) International Journal of Data Mining Techniques and Applications
Volume: 04 Issue: 01 June 2015 Page No.15-18
ISSN: 2278-2419
attempt is a significant step towards expressing the relationship
between SFC and Torque standards since the work has been
consistent and within specified design limits.This attempt has
revealed many interesting patterns and brought out facts about
test result anomalies. This is where, manufacturers shall have
to pay due attention, obtain insight into business and end up
improving quality standards of their products to increase their
market shares.

REFERENCES
[1] Data Mining Technologies Inc.Web. 14 Nov. 2014. < http://www.data-
mine.com/white_papers/Manufacturing >.
[2] SAP Inc. Web. 20 Dec 2014. < http://www.sap.com/asset/detail.2014-04-
apr.what-is-predictive-analytics-data-mining-mp4.html >.
[3] Jiawei Han, Michelin Kamber, JianPei. “Data Mining Concepts and
Techniques” -3rd Edition,pp. 443-445.
[4] Waldemar Wójcik and Konrad Gromaszek (2011). “Data Mining
Industrial Applications” , “Knowledge-Oriented Applications in Data
Mining”, Prof. Kimito Funatsu (Ed.), ISBN: 978-953-307-154-1, InTech,
Available from http://www.intechopen.com/books/knowledge-oriented-
applications-in-data-mining/data-mining-industrialapplications, pp. 435
– 440.
[5] Kusiak, A. “Data mining: manufacturing and service applications”,
International Journal of Production Research, Vol. 44 (2006), pp. 18-19,
ISSN: 4175-4191.
[6] Salim Charaniya, Huong Le, Huzefa Rangwala, Keri Mills, Kevin
Johnson, George Karypis, Wei-Shou Hu, “Mining manufacturing data for
discovery of high productivity process charecteristics” Journal of
Biotechnology 147 (2010), pp. 186–197.
[7] Lior Rokach, Oded Maimon, “Data mining for improving the quality of
manufacturing: a featureset decomposition approach” pp. 26-27.
Rudolf Kruse, Matthias Steinbrecher, Christian Moewes, “Data Mining
Applications in the Automotive Industry” proceedings of 4th
International Workshop on Reliable Engineering Computing (REC
2010), pp. 25-26.

18

You might also like