application of data mining for prospective assembly time … · 2019-03-05 · ids 2017 industrial...
TRANSCRIPT
IDS 2017
Industrial Data Science Conference
Application of Data Mining for Prospective
Assembly Time Determination
Dortmund / 05.09.2017, Dr. Olga Erohin and Ralf Kretschmer
Miele Group and Business Unit Professional
Research Project „Pro Mondi“ and time data management
Knowledge discovery for prospective assembly time prediction
Assembly time prediction in the product development phase
Conclusion
2
Agenda
3
Scope and goals of the research project Pro Mondi
Assembly oriented data model
(product and process view)
Identification and
representation
of process-relevant
product properties
Detection and description
of product-characteristic
process pattern
Mapping of product
and process structure
Functional BOM
Product planning
Engineering BOM
Product
development
Manufacturing BOM Manufacturing BOM
Production
Product emergence
Manufacturing
Cross-domain knowledge
BOM=Bill of Materials
Process planning
Research project (2012-2015):
„Prospective determination of
assembly work content in digital
factory (Pro Mondi)“
Source: Research project “Pro Mondi”, http://www.pro-mondi.de/
15-70% of production time is assembly time
Manual assembly is a wide-spread assembly method
for multi-variant products
4
Assembly time is the basis for various processes
Time
data
Benchmarking
Work system
design
Production
control
Performance
assessment
Capacity
planning
Cost
calculation
Human resource
planningScheduling
Invest
planning
Value stream
design
Sources: B. Lotter: Einführung. In: B. Lotter, H.-P. Wiendahl (Hrsg.): Montage in der industriellen Produktion – Ein Handbuch für die Praxis. Springer Verlag 2012.
J. Deuse, F. Busch: Zeitwirtschaft in der Montage. In: B. Lotter, H.-P. Wiendahl (Hrsg.): Montage in der industriellen Produktion – Ein Handbuch für die Praxis. Springer Verlag 2012
Miele & Cie. KG.
Time-related data are applied in manifold areas
Time data management (e.g. time studies) is an
essential task field of Industrial Engineering
5
Two results of the research project Pro Mondi
5
Time agreement
Predetermined motion time systemsSimulation (calculation)
Registering by devices Inquiry
Self-recordingTime and motion study
Work sampling study
Standard data building blocks
Comparative estimating
MTM-AnalysisMTM-ProKon
ProductionProduction planning
Product development
ProWiZei Knowledge
discovery for
prospective
assembly
time
prediction
ATPAssembly
time
prediction in
the product
development
phase
Sources: O. Erohin, J. Schallow, J. Deuse, R. Klinkenberg: Application of data mining to predict assembly time in early phases of product emergence. CIE43 Proceedings, 2013.
Miele & Cie. KG
6
KDID in context of time data management
Milestones Iterative steps TDM: Time Data Management
KDID: Knowledge Discovery in Industrial Databases
Preparation
Data Mining
Realization
Define the goals of
knowledge discovery in time
data management
Explore the processes
of time data management
Explore relevant IT systems
and TDM-data connections
1.1
1.2
1.3
Create an IT prototype
Integrate KDID into
planning processes and
IT systems
3.1
3.2
Visualize and interpret
the results
Create and apply data
mining models
Descriptive and
explorative analysis of
data
Build a raw data
matrix of TDM-data
M2
2.4
2.1
2.2
2.5
Preprocessing of raw
data matrix of TDM-data
2.3
M1
Source: O. Erohin: Wissensgewinnung durch Datenanalyse zur prospektiven Zeitermittlung. Shaker Verlag 2017.
Pro
WiZ
ei
Morp
holo
gy Determination Pre-Processing Application
Data management
Rapid
Min
er
7
Realization of KDID and some prediction results
Data Mining method RapidMiner operatorRelative
error
Linear regression Linear Regression 18,2%
Local polynomial
regression
Local Polynomial
Regression
33,1%
Regression model tree W-M5P 33,4%
Regression tree W-RepTree 29,5%
Support vector
regression
LibSVM
Support Vector Machine
17,1%
Source: O. Erohin: Wissensgewinnung durch Datenanalyse zur prospektiven Zeitermittlung. Shaker Verlag 2017.
8
Continuity of time data determination along the product
emergence process (PEP)
design phasedesign
freezedevelopment phase
desig
ner
industr
ialengin
eer,
assem
bly
pla
nner
process planningdevelopment
iteration loop iteration loop iteration loop
PEP
time
type
process
step
software
Data
class
approximate time
production-oriented
design (assessment of
assembly fairness)
target time
analysis of
predetermined motion
time systems
prediction time
prospective assembly
time prediction
product data process data
Source: Personal research work by R. Kretschmer
9
Mapping of product and process data
B-building-block(s)
(joining processes)
V-building-block(s)
(connection processes)
+
basic assembly time for component A
+
+
pro
du
ct-
orie
nte
dtim
e-d
ata
str
uctu
re
…
hie
rarc
hic
alp
rod
uctstr
uctu
re
product
component
group 1
component A
component B
component C
component
group 2
component D
component E
…
process data(industrial engineering, assembly planning, …)
product data(CAD, PDM, …)
assignment via ID process dataproduct data
Source: Personal research work by R. Kretschmer
10
Concept for assembly time prediction
current data (model use)data based on the past
(model creation based on instances group)
process steps process dataproduct data
pro
du
ctd
ata
pro
ce
ss
da
tacomponents 1-n new component
assembly times 1-n
n
n
B-building blocks
(joining processes)
n
n
V-building blocks
(connection processes)
Connection process
type(s) + number
identification of k-
nearest-neighbours
predicted assembly time for a new
component
n n
prediction
B-building block(s)
(joining processes)
prediction
V-building block(s)
(connection processes)
+
mapping
via ID
Source: Personal research work by R. Kretschmer
Assembly-time prediction compared to system of predetermined time (MTM-TiCon)
and other established time-determination systems
Conclusion:
Best results for further development within the component family
Widely varying quality of results for complete new design and/or missing component
family (=> small data amount for comparable components related to the past) 11
Evaluation: Validation and assessment of results
Deviation to system of predetermined time (relative error)
Source: Personal research work by R. Kretschmer
0,0% 5,0% 10,0% 15,0% 20,0% 25,0% 30,0%
front panel A
front panel B
front panel C
front panel D approximate time according toProKondigital
assembly time prediction (ATP) for achanged computer-aided design (k=4)
assembly time prediction (ATP) for a newcomputer-aided design (k=4)
actual time recording according to REFA
Integration of data mining for prospective determination of assembly time leads to
essential added value for planning and decision-making
… and supports the idea of simultaneous engineering to reduce the product
emergence time.
Current portfolio of methods for time determination can be successfully extended
by new data mining methods.
Fundamental factors of success are
Integration of specific know-how of the application area (especially at the beginning of
knowledge discovery).
Overcoming the challenges of “historically evolved” IT infrastructures.
12
Conclusion
13
Ralf Kretschmer
Director Segment Professional
Laundry Technology Lehrte
Miele & Cie. KG
Industriestraße 3, 31275 Lehrte
http://www.miele.com
Dr. Olga Erohin
Director Corporate Development
Professional Technology
Miele & Cie. KG
Mielestraße 2, 33611 Bielefeld
http://www.miele.com
Thank you very much for your kind attention!
For further information please visit:
www.pro-mondi.de