SlideShare a Scribd company logo
Improving productivity and yield by
using combination of Taguchi method
and Machine learning techniques
CONFIDENTIAL AND PROPRIETARY
Any use of this material without specific permission is strictly prohibited
1
Presentationby JagadishC.A.
jagadish.chandra@qmaxim.com
20th of Feb., 2015, v1.5
Illustrated in a case study related to manufacture of casting by sand casting
process
CASE STUDY PRESENTED HERE ILLUSTRATES A NEW
APPROACH FOR SOLVING THIS KIND OF PROBLEM.
KNOWLEDGE OF METALLURGY, PROCESS & CASTING IS
COMBINED WITH ADVANCED STATISTICAL OPTIMIZATION
TECHNIQUES OF MACHINE LEARNING AND TAGUCHI
METHOD IS USED. PROBLEMS CAN SOLVED MORE
QUICKLY, ECONOMICALLY & WITH GREATER CERTAINITY.
PROBLEMS LIKE CRACKING IN CASTINGS IS
TRADITIONALLY SOLVED BY TRIAL AND ERROR
METHOD USING KNOWLEDGE OF METALLURGY,
PROCESS & CASTING. THIS IS A HIT & MISS
PROCESS.
2
BY READING THIS PRESENTATION
ONE CAN GET SOME INTUITION
ABOUT THE APPROACH & HOW ONE
CAN APPLY IT IN THEIR OWN WORK
3
content
4
•Case study phase #2 – DOE by Taguchi Method & advanced
data analysis
•Appendix-1: Primer on Machine learning & Taguchi Method
•Case study phase #1 –exploratory analysis
Setting up context-Setting up context-
economicreality, challenges, opportunities
•Phase #2– confirmation
•Findings, Conclusions, summary
•Questions,background, Contacts
•Overview of new approach, case study background
•Appendix-2: Sand casting some resources
The manufacturing reality
manufacturing sector-many challenges
• Still low growth rates in manufacturing
• Customer requirements getting tougher
• Severe competition and pressure on margins
• Continuouspressure to reduce costs
• Increasing complexity, automation,increasing
need for new skills
• Younger inexperienced but ambitious workforce
• Continuouspressure for rapid & big
improvements
The manufacturing issues
data analysis & continuous improvement challenges
• Data deluge –vast amounts of data collected but not much
analysis. Many sources of data – SCADA, CRM,ERP, customer
data, handwritten notes, Excel., social networks....
• Data in electronic or paper form, many formats – quite often
not analysed
• Internet of things becoming reality – more data, more
communication- increasing complexity
• too many variables – how to reduce the number to do
analysis?
• variables are not just quantitative but categorical or ordered
categorical .
• Simple data analysis tools less and less useful e.g. Simple
Regression
• Continuous improvement based on Kaizen, 7 basic tools for
quality improvement, Shainin DOE are useful but not that
effective for big improvements
The manufacturing reality
many challenges but also opportunities...
• Classical vary one factor at time approach has severe
limitations
• Choosing operating conditions to balance productivity
& yield is not easy
• Traditional six sigma does not incorporate many of the
new tools and methodologies which are now available
• But, this is also an opportunity......
– Unheard of granularity & richness of data
nowadays
– Falling cost of computation, storage
– Powerful machine learning tools are available for
data analysis & optimization (many of them open
source & free)
8
content
8
•Case study phase #2 – DOE by Taguchi Method & advanced
data analysis
•Appendix-1: Primer on Machine learning & Taguchi Method
•Case study phase #1 –exploratory analysis
•Setting up context-
economicreality, challenges, opportunities
•Phase #2– confirmation
•Findings, Conclusions, summary
•Questions,background, Contacts
••Overview of new approach, case study background
•Appendix-2: Sand casting some resources
Outline of the new methodology is as follows
• New methodology
– Uses combination of machine learning & Taguchi
method
– Can find best operating conditions rapidly without
compromising productivity
• Illustrated with a case study related to sand
casting, but can be applied to any manufacturing
situation
(Why sand casting chosen?
Sand casting is very noisy process with many variables, uses
highly variable natural substances & is manual process – if it can
work in sand casting it can work in any manufacturing situation )
– see next 3 slides for overview of methodology
– case study in the subsequent slides
9
manufacturingimprovementstrategy outline
main objectives and major issues manufacturing faces
1. Improveyield by reducing
defectssuch as cracking, pin
holes, bad surface......
Mainobjectives of
manufacturing
2. Improveproductivity & cost
by reducing time taken by
operating at highest possible
speed and reducing inputs as
much as possible
How to set process parameters?
e.g. Temperature, degassing rate, speed – start
and steady, time- start, ramp up...
Make any major changes?
e.g. Equipment change, mould design, invest?
Majorissues in manufacturing
manufacturingimprovementstrategy outline
main objectives and major issues manufacturing faces
Classical approach to tackle this problem which most companies follow
1. Improveyield by reducing
defectssuch as cracking, pin
holes, bad surface......
Mainobjectives of
manufacturing
2. Improveproductivity & cost
by reducing time taken by
operating at highest possible
speed and reducing inputs as
much as possible
How to set process parameters?
e.g. Temperature, degassing rate, speed – start
and steady, time- start, ramp up...
Make any major changes?
e.g. Equipment change, mould design ..
at a time holding others constant
validity
Do trials by varying one factor
at a time holding others constant
Problem:
Many parameters – too many
trials. Results have no statistical
validity
Quite often investments are madeQuite often investments are made
without utilising potential of existing
equipment
Majorissues in manufacturing
ClassicalapproachClassicalapproach
manufacturing improvement strategy outline
main objectives and major issues manufacturing faces
Classical approach which most companies follow
new approach has three phases as outlined
1. Improveyield by reducing
defectssuch as cracking, pin
holes, bad surface......
Mainobjectives of
manufacturing
2. Improveproductivity & cost
by reducing time taken by
operating at highest possible
speed and reducing inputs as
much as possible
How to set process parameters?
e.g. Temperature, degassing rate, speed – start
and steady, time- start, ramp up...
Make any major changes?
e.g. Equipment change, mould design,
modernization ..
Phase #1Phase #1
Do exploratory data analysis existing data
using linear regression & machine
learning. Gain insights & short list
important variables
Do sophisticated data analysis and find best operating
Phase #2
1. Do trials by Statistical experimentation by making
changes to process parameters and system.
2. Do sophisticated data analysis and find best operating
conditions
Majorissues in manufacturing
New approachNew approach
manufacturing improvement strategy outline
main objectives and major issues manufacturing faces
Classical approach which most companies follow
new approach has three phases as outlined
1. Improveyield by reducing
defectssuch as cracking, pin
holes, bad surface......
Mainobjectives of
manufacturing
2. Improveproductivity & cost
by reducing time taken by
operating at highest possible
speed and reducing inputs as
much as possible
How to set process parameters?
e.g. Temperature, degassing rate, speed – start
and steady, time- start, ramp up...
Make any major changes?
e.g. Equipment change, mould design,
modernization ..
Phase #1Phase #1
Do exploratory data analysis existing data
using linear regression & machine
learning. Gain insights & short list
important variables
Do sophisticated data analysis and find best operating
Phase #2
1. Do trials by Statistical experimentation by making
changes to process parameters and system.
2. Do sophisticated data analysis and find best operating
conditions
Phase #3
Decide whether to make major investments
Majorissues in manufacturing
New approachNew approach
Phase #1,#2 covered
in in the following
slides
Outline of the new methodology, it has following steps
1. Do exploratory data analysis by using many
modelling algorithms - linear regression, cluster
analysis & LASSO (or other machine learning
algorithms) and gain insights and shortlist
variables of importance (Phase #1)
2. Use Taguchi Method (TM) to find important
factors & their best levels (DOE) by doing
planned manufacturing runs & analysing data
(Phase #2)
( what is machine learning & TM? see appendix-1)
Continued...
17
Outline of the new methodology, it has following steps
3. Confirm findings by doing actual
manufacturing runs using optimum
conditions found in #2 (Phase #2)
4. Use entire data and build models by using
linear regression & LASSO (or other machine
learning algorithms) – for further insights and
verify efficacy of findings (Phase #2)
5. Operate under optimum operating conditions
found in steps #1to #5 & monitor
performance on a long term basis
Note : these steps marked as: on the slides
18
1
Illustrative case study details
• Objective :
– Reduce cracking significantly
in a mild steel part of a
particular design
manufactured by sand
casting
– Also, significantly reduce
overallconsumption of
welding electrode for repair
(additional details next slide)
(what is sand casting? See Appendix -2)
19
0
Illustrative case study details
• Manufacturing Process : Sand casting
• Operating mode: completely manual
• Design : pipe joint, mild steel
• Complicated production process (2 box, cored)
• Project relates to cracking seen in a certain design
during 1st stage inspection after pouring, shake out,
fettling, heat treatment, shot blasting (see schematic of
process & some pictures in following slides)
• Magnetic particle inspection (MPI) required as crack is not
visible to naked eye. Defectiveportions have to be
repaired
• Almost all castings crack
• Cracking has severe cost implications –repair, welding
consumables,heat treatment, lost time....
20
0
Case study: Reducing cracking in pipe joint cast by sand casting
Schematic process flow rework by welding at 1st stage inspection,
Amplification effect- for each 1kg of casting got to cast 1.9 of metal
21
1st stage inspection,1st stage inspection,
defectdetection&
repair
0
Illustrative case study details –some pictures
Some Process steps –2 piece box bottom with chills
23
0
Bottomhalf of mould
with chillsseen
Core coated and ready
for insertion
Metalpoured into
mould
Illustrativecase study details
cracks almost always occur at approximately the same place
24
0
Cracks marked after inspection.
Cracks occur mostly at the same
location
Crack runs throughout
the section of pipe
25
content
25
•Case study phase #2 – DOE by Taguchi Method & advanced
data analysis
•Appendix-1: Primer on Machine learning & Taguchi Method
Case study phase #1Case study phase #1 – exploratory analysis
•Setting up context-
economicreality, challenges, opportunities
•Phase #2– confirmation
•Findings, Conclusions, summary
•Questions,background, Contacts
•Overview of new approach, case study background
•Appendix-2: Sand casting some resources
Phase #1 overview
varioussteps, sourceof knowledge
following slides havedetails
Furnacecontent, operating conditions, chemistry
records
14 months,100 furnaceloads, 4-6 designs / cast
=6500 records
Rework records
Technical
information
Cleanup, combine, createnew
variables, separate information
aboutcrack pronedesign of
interest
Exploratory graphs. Advanced
statistical analysis – regression,
machinelearning -model
building
Interpretations, Findings,
presentations
Get insights, shortlist
importantvariables & decide
on next steps
Metallurgy
casting
Machine
learning
1
Effect of Variables on response variable studied
Companycollects vast amountsof data
furnace content, schedule, chemistry, operating conditions....studied (~27 variables)
• PartName
• Fur No
• Furnace content
• Week day
• shift
• grade
• Pour Time Slot
• Tap temp
• Mould Hardness
• coating mould
• Core Hardness
• coating core
• Piece weight to total
Liquid ratio
• Ladle temp.
• C Mn Si S P
• Cr Ni Mo Cu V
• Carbon equivalent –
Conventional
• Carbon equivalent –
Japan
1
Some Exploratory graphs- histograms
tap temp, core hardness, mould hardness, ladle temperature have 2 distinct groups
1
Some Exploratory graphs of chemistry -histograms
Mostly near normal distribution, some large values seen 1
Exploratory graphs - histograms
Ni,Mo,Cu, skewed with some large values
1
Exploratory ECDF (empirical cumulative distribution function) graphs of variables
2 furnaces show different behaviours w.r.t. tap and ladle temperature this is probably the reason
for grouping seen
Furnace B
Furnace A
1
ExploratoryECDF graphs –mould/ core hardness
different mould coatings show highly different hardness- this is reason for groups seen
cotpol
Ceramol
930
1
Some of the Exploratorybox plot graphs of operatingconditions on cracking tendency
higher tap temperature, higher mould hardness, higher ladle temperature are better
1
cluster dendrogram done to group similar variables in entire dataset
Relationship ofcracking with various variablesmarked – zoomed view next slide 1
Cluster dendrogram , area of interest zoomed in - relationshipof cracking with various
variables
cracking tendency groups with shift, tap (melt) temperature,ladletemp, furnace No
1
Linear Regression analysis of data set between cracking built
ANOVA table created the table by introducing each of the terms in the model one at time
Variables of importance highlighted
Analysis of Variance Table
Response: CracksP
Df Sum Sq Mean Sq F value Pr(>F)
Fur_No 1 11.073 11.0733 18.0554 4.657e-05 ***
Paint_mld 1 3.529 3.5291 5.7544 0.0182110 *
Paint_core 3 2.266 0.7552 1.2314 0.3020912
Tap_temp 1 17.235 17.2349 28.1022 6.408e-07 ***
Mld_Hardness 1 0.525 0.5252 0.8564 0.3568672
Core_Hardness 1 0.126 0.1258 0.2052 0.6515131
Ladle_temp 1 0.949 0.9493 1.5479 0.2162225
C 1 8.978 8.9779 14.6388 0.0002214 ***
Mn 1 0.204 0.2045 0.3334 0.5649174
Si 1 0.122 0.1219 0.1987 0.6566789
S 1 2.260 2.2598 3.6846 0.0576309 .
P 1 1.143 1.1432 1.8640 0.1750845
Cr 1 0.254 0.2536 0.4135 0.5215845
Ni 1 3.314 3.3135 5.4028 0.0220296 *
Mo 1 0.005 0.0048 0.0078 0.9297360
Cu 1 4.506 4.5063 7.3477 0.0078453 **
V 1 0.116 0.1155 0.1883 0.6652000
P2_L_ratio 1 4.742 4.7424 7.7327 0.0064290 **
C:Ni 1 1.200 1.1995 1.9559 0.1648990
Cr:Ni 1 1.871 1.8714 3.0514 0.0835905 .
Residuals 105 64.396 0.6133
---
Signif. codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1
1
Model built using LASSO (generalized linear model via penalized maximum likelihood ) regression
output shown
The algorithm keeps coefficients of only important variables
(Intercept)Fur_NoB Paint_mldCotpol Tap_temp Ladle_temp C
-0.4368 .4353 0.3358 -0.4066 -0.2325 0.1821
S Ni Cu P2_L_ratio
0.1313 0.2184 -0.1775 -0.1882
> coef(cvfit, s = "lambda.min")
30 x 1 sparse Matrix of class "dgCMatrix"
1
(Intercept) 7.888494721
no_pcs -0.002647573
Tap_temp -0.003587278
Mld_Hardness .
Core_Hardness .
Ladle_temp -0.001463345
C 1.240515245
Mn .
Si .
S 1.225540417
P 1.746826872
Cr -0.100157651
Ni 0.595015164
Mo 0.270251629
Cu -1.480281443
V .
P2_L_ratio -0.129797323
CEa .
CEj .
Fur_NoB 0.051100821
Fur_NoC .
Paint_mldCotpol 0.002589163
Paint_mldHolcoat .
Paint_mldIsomol .
Paint_mldSNS .
Paint_coreCotpol 0.020449112
1
Summary of some of the findings from phase #1
distinct groupings seen, reason found
• distinct groups seen – tap temp, ladle temp, mould
hardness, core hardness, furnace modifiers ratio,
casting sequence
• tap temp, ladle temp depends on furnace used, casting
time slot, weekday.
• mould hardness, core hardness depends coating type
• Chemistry many elements skewed, contamination seen
• furnace modifiers ratio – large difference in ratios
between furnaces
1
Conclusions, insights, summary
operating conditions, chemistry are correlated with cracking
• Several algorithms used to shortlist variables of
importance and their effect on cracking
• Findings:
– almost similar conclusion can be drawn from output of all algorithms
– Some of the variable have significant effect on cracking e.g. Melt
temperature, ladle temperature , furnace No...
– Effectof variables on cracking is as follows:
Tap temp (+), ladle temp (+), mould hardness (+), core hardness (+),
furnaceA ( +), time slot (-), weekday (-), modifiers ratio (+/-), casting
sequence(mid better), piece to liquid ratio (+)
Chemistry – C(-),Mn(-),S(-),P(-),Cr(+),Ni(-),Mo(-),
Cu(+),Carbon Eq. (-)
( Coding :- + higher level better, - lower level better)
1
Conclusions, insights, summary
operating conditions, chemistry are correlated with cracking
• Significantinsights gained from phase #1 by
studying existing data without making any
process change
• Other aspects like methoding, raw material,
operating conditions-delay, ladle conditions,
pouringheight, ambient humidity etc may
contributeto cracking
• Detailed industrial study such as DOE needs to
be done to get even better insight to solve this
problem (to be done in phase #2)
1
41
content
41
•Case study phase #2 – DOE by Taguchi Method & advanced
data analysis
•Case study phase #2 – DOE by Taguchi Method & advanced
data analysis
•Appendix-1: Primer on Machine learning & Taguchi Method
•Case study phase #1 – exploratory analysis
•Setting up context-
economicreality, challenges, opportunities
•Phase #2– confirmation
•Findings, Conclusions, summary
•Questions,background, Contacts
•Overview of new approach, case study background
•Appendix-2: Sand casting some resources
Methodology followed (phase #2)
Overview of steps – following slides give details of each box
Make general observations about the problem –frequency, where, when, macro, …
Choose a DOE experimental plan based on phase #1 conclusions, new considerations & cost aspects -
factors to vary, their levels , interactions , number of replicates per run
Taguchi DOE chosen
Factors – chemistry (pure, impure), tap temp (low, high), ladle treatment( high, low), methoding -design (old, new)
Do casting runs (8x4) according to plan
Collect data
Identify each casting
Strip casting, section, HT, grit blast, prepare for MPI
Make Interpretations, findings, recommendations
Make presentation
Measure for each casting -crack length by MPI & DP and record
Data entry to specialized software, analyse data
and generate reports and
charts (mean , S & N )
Metallurgy
casting
Machine
learning
Build regression and machine learning models using all individual data
Do casting runs using optimum levels of factors
Measure crack length by MPI & DP and record
2
Phase #2
Steps in finding important factors & their best levels by doing TM casting runs.
Overview of experimental plan was as follows, details in following slides
44
1. methoding
2.chemistry
3. tap temp
4. Ladle treatment
Factors
chosen
As many as 150 variables.
Other factors heldconstant/monitored:
e.g. Sand properties, binder properties, mould
preparationmethod, mouldcoatings, furnace
content, furnace treatment, ambient temp. ,
relative humidity..........
Design
chosen
Experimental
design (L8), (27)
array, 8 runsin
duplicate
(2x8x2=32
castings)
Experiments
run
Response &
other data
collected
Data
analysed
Program
output:
response
table and
charts (mean
& S/N)
Conclusions
1. Factorsof
significance
2. Interactions
of
significance
3. Optimum
factor levels
Response : length of
crack measured by
MPI
Note:
1.Morethefactors/levels chosen larger will be the
number of experimental runsto be done, hence more
expensive it becomes.
2. Do production run as per the design chosen
2
Castingexperimental plan based on DOE
plan shown factorswith their levels, interactions, response variable
Chemical analysis,tap temp, ladletemp, fill time, start/end time,
preparationdelay,knockoutdelay, etc noted down
Response: Crack length
Interactionsalso estimated :
chemistry with melt temp, melt temp with ladle treatment
chem furnace melt temp
ladle
treatment
methoding
design DOE_Run_No
pure high (1625-1640) high old 1
pure high (1625-1640) low new 2
impure low (1590-1605) high old 3
impure low (1590-1605) low new 4
pure low (1590-1605) high new 5
pure low (1590-1605) low old 6
impure high (1625-1640) high new 7
impure high (1625-1640) low old 8
Factorstested:
Chemistry furnace, melt temp, methoding design*, ladle
treatment
* See next slide for details
2
Methoding–design changes made during phase #2
Old design shown in picture, changes made iteratively to some of the chills and a risers 2
DOE-means plots findings
Modified mould design is significantly better, higher levels of ladle treatment, higher level of melt temp
& impure chemistry better
2
DOE interactionplots
Interaction terms aresignificant 2
DOE findings - ANOVA of means output
Modified mould design is significantly better & the most importantamong factors.
Interactionsof chem. analysis with melt temp and melt temp with ladle treatment are also significant
Analysis of Variance for Means
Source DF Seq SS Adj SS Adj MS F P
chem furnace 1 35.596 35.596 35.596 41.33 0.098
melt temp 1 17.627 17.627 17.627 20.46 0.138
ladle treatment 1 104.221 104.221 104.221 121.00 0.058
mould design 1 397.268 397.268 397.268 461.23 0.030
chem furnace*melt temp 1 265.939 265.939 265.939 308.76 0.036
melt temp*ladle treatment 1 176.955 176.955 176.955 205.44 0.044
Residual Error 1 0.861 0.861 0.861
Total 7 998.467
Response Table for Means
chem ladle mould
Level furnace melt temp treatment design
1 36.94(P) 33.34 (H) 38.44(high) 41.88 (old)
2 32.72(IP) 36.31 (L) 31.22 (low) 27.78 (new)
Delta 4.22 2.97 7.22 14.09
Rank 3 4 2 1
2
General conclusions on cracking susceptibility as per DOE findings & best operating conditions are as
follows
• Best operating conditions:
– New methoding design is significantlybetter than old
design
– Higher melt temperature range is better
– Higher levels of ladle treatment is better
– Recommended chemical composition is impure, but
effectis small between pure and impure.
– Also, interaction between chem. Composition/ ladle
treatmentwith melt temperature is important
• Several casting runs were done under the
recommended conditions to verify findings
• Overall, 4 iterations of new designs also tried
3
Linear Regression was done of the complete dataset
ANOVA table output
Mould design, sequence of casting, S,P, ladle temperature significant
Analysis of Variance Table
Response: crack length
Df Sum Sq Mean Sq F value Pr(>F)
mould.design 4 4684.0 1170.99 12.1631 1.86e-05 ***
sequence 3 833.7 277.91 2.8867 0.057489 .
C 1 0.8 0.77 0.0080 0.929651
S 1 782.3 782.25 8.1253 0.009053 **
P 1 606.3 606.28 6.2974 0.019581 *
Ni 1 141.7 141.70 1.4719 0.237362
Cr 1 8.4 8.45 0.0878 0.769705
ladleTemp 1 432.5 432.53 4.4927 0.045046 *
Residuals 23 2214.3 96.27
---
Signif. codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1
4
Modelwas also built using LASSO (generalized linear model via penalized maximum likelihood ) regression of
completedata set
Outputis seen below
The algorithm drops variables which are not important, important variables shown in the table below
Coefficients generated by LASSO after cross-validation
3 x 1 sparse Matrix of class "dgCMatrix"
1
(Intercept) -472.51096012
C .
S 1594.72775370
P 3218.85844853
Ni -50.26465017
Cr .
Si .
Mn .
Cu .
Mo .
V .
Al .
Ti 8370.87631400
Fe.silinium .
tapTemp -0.05934132
ladleTemp 0.32709791
mould.designnew1 -15.17664341
mould.designnew2 -1.47706598
mould.designnew3 -41.25121292
mould.designnew4 -43.62759697
sequence2 -7.92797995
sequence3 -12.29255408
sequence4 -3.85168798
4
53
content
53
•Case study phase #2 – DOE by Taguchi Method & advanced
data analysis
•Appendix-1: Primer on Machine learning & Taguchi Method
•Case study phase #1 – exploratory analysis
•Setting up context-
economicreality, challenges, opportunities
••Phase #2– confirmation
•Findings, Conclusions, summary
•Questions,background, Contacts
•Overview of new approach, case study background
Severalcasting runs weremade under optimum operating conditions to verify findings
BoxPlots comparison of most important factor– methoding design shown
New designs are significantly better, Iteration new #4 is best of the lot under optimum operating conditions
Longer casting runs will be produced using the optimum conditions found to evaluate long term performance
4
5
55
content
55
•Case study phase #2 – DOE by Taguchi Method & advanced
data analysis
•Appendix-1: Primer on Machine learning & Taguchi Method
•Case study phase #1 – exploratory analysis
•Setting up context-
economicreality, challenges, opportunities
•Phase #2– confirmation
••Findings, Conclusions, summary
•Questions,background, Contacts
•Overview of new approach, case study background
•Appendix-2: Sand casting some resources
OVERALL CONCLUSIONS
56
• Approach on use of Taguchi method and
Machine learning combination for yield &
productivity improvement was explained. Some
background information was also presented
• Illustrative case study related to sand casting
manufacturing process was presented.
• Use of various algorithms to quickly gain
valuable insights and to find variables of
importance and their effect was illustrated
• Optimum operating conditions found from both
the phases were revalidated in actual
production run
Continued.....
OVERALL CONCLUSIONS
57
• This project required relatively limited effort,
was of low cost, was of short duration, did
not disrupt operations significantly
• By using this approach, it is possible to
efficiently find optimum operating
conditions even in highly noisy, completely
manual process with large number of
variables
• It is also possible to use the information
gained by this approach to decide on
prioritizing areas of investment for
modernization
58
content
58
•Case study phase #2 – DOE by Taguchi Method & advanced
data analysis
•Appendix-1: Primer on Machine learning & Taguchi Method
•Case study phase #1 – exploratory analysis
•Setting up context-
economicreality, challenges, opportunities
•Phase #2– confirmation
•Findings, Conclusions, summary
••Questions,background, Contacts
•Overview of new approach, case study background
THANKS!
QUESTIONS?
59
Jagadish C.A. (Rao)
jagadish.chandra@qmaxim.com
qmaxim001@gmail.com
+91 9900 606620
+91 9538 328704
blog: qmaxim.wordpress.com
Twitter: @JagSpeak
q-Maxim
Jagadish C.A. ‘s Profile
60LinkedIn profile: http://goo.gl/Lp3lWv
• Heads niche consultancy q-Maximfocused on adv. Optimization, quality, technology….
• B.Tech (MetallurgicalEngg., NIT-K,Surathkal,India)
• Done many graded online courses (4-15weeks) from leading US universities on
operationsmanagement, advanced data analysis, marketing, finance, accounting,
strategy,advancedcompetitive strategy,data scientist (5 courses),credit risk
management,game theory, logistics
• ASQ (American society for quality) certified Six sigma Black belt (since 2002)
• ASQ certified Manager of Org. Excellence/ Quality (since 1999)
• Certified EFQMassessor extensive experience assessing companies, fashioning
transformationalroadmap& implementing EFQMmodel
• JuranQI facilitator,Cert. adv. Industrial experimentation, Analytics
• ISO 9001:2008,14001,QS-9000(TS16949)lead auditor
• ~32years experience in Manufacturingandtransactionalfields, 7.5 years as
managementconsultant
• Richexperience in Quality, process management, R +D, Technology, Cost reduction,
Managementconsultancy.Extensive experience in using advanced methodologies for
problem solving and optimization
• Extensive exposure/ knowledge of heavy process oriented manufacturing – Aluminium
casthouse, Steel, Welding, casting, foundry, Metallurgy, etc
• Widely travelled – India, north America, Europe, M.E.
61
content
61
•Case study phase #2 – DOE by Taguchi Method & advanced
data analysis
••Appendix-1: Primer on Machine learning & Taguchi Method
•Case study phase #1 – exploratory analysis
•Setting up context-
economicreality, challenges, opportunities
•Phase #2– confirmation
•Findings, Conclusions, summary
•Questions,background, Contacts
•Overview of new approach, case study background
•Appendix-2: Sand casting some resources
Whatis Machine learning ?
Some definitions
• Definition : “A computer is able to learn by experience
without explicitly being programmed – & improves
performance as it learns”
• Based on field of artificial intelligence
• Examples :
– Mining data from large datasets website click trough data to improve
purchase conversion rate
– Autonomous self flying helicopter (Stanford University)
– Classify e-mail as spam or not spam (filtering spam in outlook.com)
– handwriting recognition (tablets)
– Computer Vision (reading car number plates & giving speeding tickets)
– Self driven cars (Google self driving car)
– Recommendersystems (Amazon recommending books)
• Not that common in manufacturing....
62
Whatis Machine learning ..
2 major types - Predictive & descriptivetypes
• Predictive learning (supervised learning)
there is outcome variable to guide the learning process
– Learning phase
learning by example. Tune the model until error is sufficiently
low
– Scoring phase
Use the model for making predictions (or score) in real time
e.g.: linear regression, polynomial regression, LASSO, random
forest, Neural network, Decision tree
• Descriptive (Unsupervised learning)
– No outcome variable
– Self-organization, no teacher
– Groups observations (variables) based on similarity, to uncover
hidden relationships
– e.g.: clustering
• Regression or classification problem based on type of outcome
63
Machine learning example –data in simple table form
Learning algorithm for predicting house prices- various parts labelled
64
Row
no
Area
[sq. Ft.]
Number
of rooms
Age of
flat
[years]
Gym
[Y/N]
Swim
ming
pool
[Y/N]
............... Other
features not
shown.............
Market
price
( lakh
Rupees)
1 1800 5 1.1 yes yes 68.6
2 900 3 4 no no 34.5
3 1720 5 8 yes no 47.7
4 560 2 .7 no no 25.4
.....
1000 2400 6 3 yes yes 91.8
CalledTarget
or outcome
or outputCalledPredictors or
inputsor features
Records
orrows
Machine learning– supervised learning -overview
example- predicting market price of house using simple linear learning algorithm
65
SampledTrainingdataset
Known
1. Area of house
2. Number of rooms
3. Age of house
4. Location
5. Gym [y/n]
6. ..... Etc, etc
Learning algorithm
predictive hypothesis
h(x)
Prediction
marketprice of
house
Calledtarget or
outcomeCalledfeatures
or predictors
h(x)is a linear equationof
the type:
hθ(x) = θ0+ θ1x1 + θ2x2 +....... Θnxn
Past data of
housing market
having features &
predictors
Learning
phase
scoring
phase
Unsupervised learning – K means cluster analysis and dendrogram
Groups observations (variables)based on similarity, to uncoverhidden
relationships
• K-means clustering- a type of unsupervised
learning
• Example:
Clustering of wine database from 1996, of wines grown
in the same region of Italy, but derived from three
different types ( cultivars )
Databasehas 78 observations, 13 chemical analysis
variables & 1 column of 3 cultivars
Source :University of California at Irving (UCI) Machine Learning
Depository
• We want the machine learning algorithm to cluster the
data & uncover the types without seeing the type (
cultivars ) & check the accuracy
Unsupervised learning – K means cluster analysis
It has classified 178 observations into 3 clusters using 13 chemical analysis
variables
Data: Universityof California at
Irving (UCI)Machine Learning
Depository
Unsupervised learning – K means cluster analysis- colored based on type (cultivar)
classificationaccuracy is pretty good as seen by coloring the cluster on type(cultivar) of wine
Supervised learning – decision tree a simple predictive learning algorithm
Decision tree to predict –’what possibility of a person surviving Titanic sinking given his/her
age,sex,sibling?’
69
A tree showing survivalof passengers on the Titanic ("sibsp" is the number of spouses or siblings aboard).
Thefigures under the leaves show the probability of survival and the percentage of observations in the leaf.
Source:WIKIPEDIA
Supervised learning – Random forest a predictivealgorithm based on decision tree
Exampleof random forest use for classifying land type using Landsat satelliteThermal
InfraredSensor image of 4 spectral bands
The algorithm can classify landmass automaticallyafter it has ‘learnt’ by seeing known
data
70Credit:internet resources
Supervised learning – Artificial Neural network predictive learning algorithm
Mimic working of brain and nervous system, Ideal for modelling complex relationships
Use of artificial neuralnetwork to do handwriting recognition.
71Credit:internet resources
Want to know more about data mining and machine
learning?
Read my presentation titled ‘Data mining and
Machine Learning - in jargon free, lucid
language’ .
Taguchi method (TM) is very popular optimization method. Some
details and advantages
• TaguchiMethod (TM) is very common optimization method
in science & engineering for over 40 years
• Has following components - System design, Tolerance design
and parameter design
• We are using parameter design
• TM is not a theoretical exercise, done in production
environment, justsomeadditional measurements
• TM has many advantages :
– Less effort, economicalthan T & E & full factorial
– much more insight
– Faster
– Canfind important factors, their best levels& interactions
– Also Signal to Noise to find best levelsfor minimizingvariation
– Ableto decide on parameterskeeping productivityin mind
– Findingshave Statisticalvalidity
73
74
• Factors :
– Control: best levelsfor desirable output : e.g. Pouring temperature
– Signal: can influence,can be controlled : e.g. volume knob position
– Noise : can’t be controlledOr are intentionallynotcontrollede.g. ambient
temperature
– Others: not changing during experimentation
• Response:
we are desirous of controllinge.g.- defect %age
• Factor levels:
– Range of values for doing experiments
• Interaction of factors
– when one factor has influenceon the effect of the other factorrespectively
DOE terminology
Taguchimethod (TM) steps involved are as follows
1. Define the problem and objective.
2. Choose response variable(s) for doing optimization.
3. List out all potential factors.
4. Choose important control factors and their range & number
of levels. Also, decide factors not to be varied and the
ranges in which they should be held constant.
5. Decide on experimental design
6. Conduct trials (called experimental runs) in actual
production environment & collect response variable(s) data.
7. Analyze the data, determine the importance & best levels of
controlfactors – conclude optimum operating conditions
8. Reconfirm findings by doing production using optimum
operating conditions
75
76
content
76
•Case study phase #2 – DOE by Taguchi Method & advanced
data analysis
•Appendix-1: Primer on Machine learning & Taguchi Method
•Case study phase #1 – exploratory analysis
•Setting up context-
economicreality, challenges, opportunities
•Phase #2– confirmation
•Findings, Conclusions, summary
•Questions,background, Contacts
•Overview of new approach, case study background
••Appendix-2: Sand casting some resources
Sand casting additional information
typical sequence of steps
Credit:Prof. Timothy Gutowski
Sand casting additional information
1. Sand casting page from Wikipedia gives a
good introduction. There are numerous
resources on the web
2. Simplified sand casting process video is
available here . There are numerous videos
on the internet
78

More Related Content

What's hot

Design For Manufacture
Design For ManufactureDesign For Manufacture
Design For Manufacture
ahmad bassiouny
 
Productivity improvement in construction
Productivity  improvement in constructionProductivity  improvement in construction
Productivity improvement in construction
KabilanP1
 
International Journal of Engineering Research and Development (IJERD)
International Journal of Engineering Research and Development (IJERD)International Journal of Engineering Research and Development (IJERD)
International Journal of Engineering Research and Development (IJERD)
IJERD Editor
 
Minimising waste in construction by using lean six sigma principle
Minimising waste in construction by using lean six sigma principleMinimising waste in construction by using lean six sigma principle
Minimising waste in construction by using lean six sigma principle
IAEME Publication
 
Implementation of Single Minute Exchange of Die in Motor Manufacturing Unit
Implementation of Single Minute Exchange of Die in Motor Manufacturing UnitImplementation of Single Minute Exchange of Die in Motor Manufacturing Unit
Implementation of Single Minute Exchange of Die in Motor Manufacturing Unit
IRJET Journal
 
Five ways to improve productivity at the construction site
Five ways to improve productivity at the construction siteFive ways to improve productivity at the construction site
Five ways to improve productivity at the construction site
Vikaslal2006
 
A012210104
A012210104A012210104
A012210104
IOSR Journals
 
Productivity Improvement Study
Productivity Improvement StudyProductivity Improvement Study
Productivity Improvement Study
saranshshah
 
M1: Introduction to Design for Manufacture
M1: Introduction to Design for ManufactureM1: Introduction to Design for Manufacture
M1: Introduction to Design for Manufacture
taruian
 
The key role of business process analysis &
The key role of business process analysis &The key role of business process analysis &
The key role of business process analysis &
kinjal29
 
Lean Construction: From Theory to Implementation
Lean Construction: From Theory to ImplementationLean Construction: From Theory to Implementation
Lean Construction: From Theory to Implementation
Antonius Pompi Bramono
 
Eliminating the production bottlenecks
Eliminating the production bottlenecksEliminating the production bottlenecks
Eliminating the production bottlenecks
dutconsult
 
Computer aided quality control
Computer aided quality controlComputer aided quality control
Computer aided quality control
Ahmad Bajwa
 
Construction Productivity
Construction ProductivityConstruction Productivity
Construction Productivity
Sunil Manjeri
 
Design for manufacturing ppt anas lahrichi
Design for manufacturing ppt anas lahrichiDesign for manufacturing ppt anas lahrichi
Design for manufacturing ppt anas lahrichi
Anas Lahrichi
 
UNIT 3 - PRODUCTION PLANNING AND PROCESS PLANNING
UNIT 3 - PRODUCTION PLANNING AND PROCESS PLANNINGUNIT 3 - PRODUCTION PLANNING AND PROCESS PLANNING
UNIT 3 - PRODUCTION PLANNING AND PROCESS PLANNING
KIT-Kalaignar Karunanidhi Institute of Technology
 
180926 UT-FS - Identifying Business Cases for 3D Printing in Service Logistics
180926 UT-FS - Identifying Business Cases for 3D Printing in Service Logistics180926 UT-FS - Identifying Business Cases for 3D Printing in Service Logistics
180926 UT-FS - Identifying Business Cases for 3D Printing in Service Logistics
SINTAS
 
PROCESS CAPABILITY
PROCESS CAPABILITYPROCESS CAPABILITY
PROCESS CAPABILITY
Angelo Del Grosso
 
Case Study: Mass Production in Small Scale Industries
Case Study: Mass Production in Small Scale IndustriesCase Study: Mass Production in Small Scale Industries
Case Study: Mass Production in Small Scale Industries
paperpublications3
 

What's hot (19)

Design For Manufacture
Design For ManufactureDesign For Manufacture
Design For Manufacture
 
Productivity improvement in construction
Productivity  improvement in constructionProductivity  improvement in construction
Productivity improvement in construction
 
International Journal of Engineering Research and Development (IJERD)
International Journal of Engineering Research and Development (IJERD)International Journal of Engineering Research and Development (IJERD)
International Journal of Engineering Research and Development (IJERD)
 
Minimising waste in construction by using lean six sigma principle
Minimising waste in construction by using lean six sigma principleMinimising waste in construction by using lean six sigma principle
Minimising waste in construction by using lean six sigma principle
 
Implementation of Single Minute Exchange of Die in Motor Manufacturing Unit
Implementation of Single Minute Exchange of Die in Motor Manufacturing UnitImplementation of Single Minute Exchange of Die in Motor Manufacturing Unit
Implementation of Single Minute Exchange of Die in Motor Manufacturing Unit
 
Five ways to improve productivity at the construction site
Five ways to improve productivity at the construction siteFive ways to improve productivity at the construction site
Five ways to improve productivity at the construction site
 
A012210104
A012210104A012210104
A012210104
 
Productivity Improvement Study
Productivity Improvement StudyProductivity Improvement Study
Productivity Improvement Study
 
M1: Introduction to Design for Manufacture
M1: Introduction to Design for ManufactureM1: Introduction to Design for Manufacture
M1: Introduction to Design for Manufacture
 
The key role of business process analysis &
The key role of business process analysis &The key role of business process analysis &
The key role of business process analysis &
 
Lean Construction: From Theory to Implementation
Lean Construction: From Theory to ImplementationLean Construction: From Theory to Implementation
Lean Construction: From Theory to Implementation
 
Eliminating the production bottlenecks
Eliminating the production bottlenecksEliminating the production bottlenecks
Eliminating the production bottlenecks
 
Computer aided quality control
Computer aided quality controlComputer aided quality control
Computer aided quality control
 
Construction Productivity
Construction ProductivityConstruction Productivity
Construction Productivity
 
Design for manufacturing ppt anas lahrichi
Design for manufacturing ppt anas lahrichiDesign for manufacturing ppt anas lahrichi
Design for manufacturing ppt anas lahrichi
 
UNIT 3 - PRODUCTION PLANNING AND PROCESS PLANNING
UNIT 3 - PRODUCTION PLANNING AND PROCESS PLANNINGUNIT 3 - PRODUCTION PLANNING AND PROCESS PLANNING
UNIT 3 - PRODUCTION PLANNING AND PROCESS PLANNING
 
180926 UT-FS - Identifying Business Cases for 3D Printing in Service Logistics
180926 UT-FS - Identifying Business Cases for 3D Printing in Service Logistics180926 UT-FS - Identifying Business Cases for 3D Printing in Service Logistics
180926 UT-FS - Identifying Business Cases for 3D Printing in Service Logistics
 
PROCESS CAPABILITY
PROCESS CAPABILITYPROCESS CAPABILITY
PROCESS CAPABILITY
 
Case Study: Mass Production in Small Scale Industries
Case Study: Mass Production in Small Scale IndustriesCase Study: Mass Production in Small Scale Industries
Case Study: Mass Production in Small Scale Industries
 

Similar to Improvement strategy by using ML and TM. A case study for solving cracking in sand casting

q-Maxim’s approach to waste reduction in foundry application using Taguchi ...
q-Maxim’s approach to waste reduction in  foundry application  using Taguchi ...q-Maxim’s approach to waste reduction in  foundry application  using Taguchi ...
q-Maxim’s approach to waste reduction in foundry application using Taguchi ...
q-Maxim
 
IRJET - Invistigation and Implement of Six Sigma and Reduce Labour Cost a...
IRJET -  	  Invistigation and Implement of Six Sigma and Reduce Labour Cost a...IRJET -  	  Invistigation and Implement of Six Sigma and Reduce Labour Cost a...
IRJET - Invistigation and Implement of Six Sigma and Reduce Labour Cost a...
IRJET Journal
 
IRJET- Deburring Methods for Elimination of Chips in the Internal Tubes of Fr...
IRJET- Deburring Methods for Elimination of Chips in the Internal Tubes of Fr...IRJET- Deburring Methods for Elimination of Chips in the Internal Tubes of Fr...
IRJET- Deburring Methods for Elimination of Chips in the Internal Tubes of Fr...
IRJET Journal
 
STUDY ON THE DEFECTS AND TO IMPROVE THE PROCESS CAPABILITY OF TREAD RUBBER US...
STUDY ON THE DEFECTS AND TO IMPROVE THE PROCESS CAPABILITY OF TREAD RUBBER US...STUDY ON THE DEFECTS AND TO IMPROVE THE PROCESS CAPABILITY OF TREAD RUBBER US...
STUDY ON THE DEFECTS AND TO IMPROVE THE PROCESS CAPABILITY OF TREAD RUBBER US...
IRJET Journal
 
Process improvement-using-dmaic-approach-case-study-in-downtime-reduction-ije...
Process improvement-using-dmaic-approach-case-study-in-downtime-reduction-ije...Process improvement-using-dmaic-approach-case-study-in-downtime-reduction-ije...
Process improvement-using-dmaic-approach-case-study-in-downtime-reduction-ije...
AjitsinghDaud
 
A case study on productivity improvement of wearing insert and cutting ring
A case study on productivity improvement of wearing insert and cutting ringA case study on productivity improvement of wearing insert and cutting ring
A case study on productivity improvement of wearing insert and cutting ring
IJECSJournal
 
Optimization of sealing casting by identifying solidification defect and impr...
Optimization of sealing casting by identifying solidification defect and impr...Optimization of sealing casting by identifying solidification defect and impr...
Optimization of sealing casting by identifying solidification defect and impr...
IRJET Journal
 
Optimization of sealing casting by identifying solidification defect and impr...
Optimization of sealing casting by identifying solidification defect and impr...Optimization of sealing casting by identifying solidification defect and impr...
Optimization of sealing casting by identifying solidification defect and impr...
IRJET Journal
 
IRJET- Productivity Improvement by Implementing Lean Manufacturing Tools ...
IRJET-  	  Productivity Improvement by Implementing Lean Manufacturing Tools ...IRJET-  	  Productivity Improvement by Implementing Lean Manufacturing Tools ...
IRJET- Productivity Improvement by Implementing Lean Manufacturing Tools ...
IRJET Journal
 
Choosing your first AI project. How to get a quick ROI in process industries
Choosing your first AI project. How to get a quick ROI in process industriesChoosing your first AI project. How to get a quick ROI in process industries
Choosing your first AI project. How to get a quick ROI in process industries
Yandex Data Factory
 
IE7610_REPORT_GROUP_8
IE7610_REPORT_GROUP_8IE7610_REPORT_GROUP_8
IE7610_REPORT_GROUP_8
Parag Kapile
 
IRJET- Optimization of Plastic Injection Molding
IRJET- Optimization of Plastic Injection MoldingIRJET- Optimization of Plastic Injection Molding
IRJET- Optimization of Plastic Injection Molding
IRJET Journal
 
IRJET- Enhance the Capacity of Outer Tube Machining Cell
IRJET-  	  Enhance the Capacity of Outer Tube Machining CellIRJET-  	  Enhance the Capacity of Outer Tube Machining Cell
IRJET- Enhance the Capacity of Outer Tube Machining Cell
IRJET Journal
 
Implementation of Six Sigma Using DMAIC Methodology in Small Scale Industries...
Implementation of Six Sigma Using DMAIC Methodology in Small Scale Industries...Implementation of Six Sigma Using DMAIC Methodology in Small Scale Industries...
Implementation of Six Sigma Using DMAIC Methodology in Small Scale Industries...
IJMER
 
IRJET- Design of Spoon Mold using Flow Analysis and Higher End Design Software
IRJET- Design of Spoon Mold using Flow Analysis and Higher End Design SoftwareIRJET- Design of Spoon Mold using Flow Analysis and Higher End Design Software
IRJET- Design of Spoon Mold using Flow Analysis and Higher End Design Software
IRJET Journal
 
IRJET- A Case Study for Overall Equipment Effectiveness Improved in Manufactu...
IRJET- A Case Study for Overall Equipment Effectiveness Improved in Manufactu...IRJET- A Case Study for Overall Equipment Effectiveness Improved in Manufactu...
IRJET- A Case Study for Overall Equipment Effectiveness Improved in Manufactu...
IRJET Journal
 
Types of production system (production and operation management)
Types of production system (production and operation management)Types of production system (production and operation management)
Types of production system (production and operation management)
Yamini Kahaliya
 
How to Reduce Changeover Time and Increase Throughput
How to Reduce Changeover Time and Increase ThroughputHow to Reduce Changeover Time and Increase Throughput
How to Reduce Changeover Time and Increase Throughput
OH!Manufacturing
 
Suggestions on the Methodology of Parameters in Fused Deposition Modeling Pro...
Suggestions on the Methodology of Parameters in Fused Deposition Modeling Pro...Suggestions on the Methodology of Parameters in Fused Deposition Modeling Pro...
Suggestions on the Methodology of Parameters in Fused Deposition Modeling Pro...
IRJESJOURNAL
 
Joined document 24_5
Joined document 24_5Joined document 24_5
Joined document 24_5
JOSEPH FRANCIS
 

Similar to Improvement strategy by using ML and TM. A case study for solving cracking in sand casting (20)

q-Maxim’s approach to waste reduction in foundry application using Taguchi ...
q-Maxim’s approach to waste reduction in  foundry application  using Taguchi ...q-Maxim’s approach to waste reduction in  foundry application  using Taguchi ...
q-Maxim’s approach to waste reduction in foundry application using Taguchi ...
 
IRJET - Invistigation and Implement of Six Sigma and Reduce Labour Cost a...
IRJET -  	  Invistigation and Implement of Six Sigma and Reduce Labour Cost a...IRJET -  	  Invistigation and Implement of Six Sigma and Reduce Labour Cost a...
IRJET - Invistigation and Implement of Six Sigma and Reduce Labour Cost a...
 
IRJET- Deburring Methods for Elimination of Chips in the Internal Tubes of Fr...
IRJET- Deburring Methods for Elimination of Chips in the Internal Tubes of Fr...IRJET- Deburring Methods for Elimination of Chips in the Internal Tubes of Fr...
IRJET- Deburring Methods for Elimination of Chips in the Internal Tubes of Fr...
 
STUDY ON THE DEFECTS AND TO IMPROVE THE PROCESS CAPABILITY OF TREAD RUBBER US...
STUDY ON THE DEFECTS AND TO IMPROVE THE PROCESS CAPABILITY OF TREAD RUBBER US...STUDY ON THE DEFECTS AND TO IMPROVE THE PROCESS CAPABILITY OF TREAD RUBBER US...
STUDY ON THE DEFECTS AND TO IMPROVE THE PROCESS CAPABILITY OF TREAD RUBBER US...
 
Process improvement-using-dmaic-approach-case-study-in-downtime-reduction-ije...
Process improvement-using-dmaic-approach-case-study-in-downtime-reduction-ije...Process improvement-using-dmaic-approach-case-study-in-downtime-reduction-ije...
Process improvement-using-dmaic-approach-case-study-in-downtime-reduction-ije...
 
A case study on productivity improvement of wearing insert and cutting ring
A case study on productivity improvement of wearing insert and cutting ringA case study on productivity improvement of wearing insert and cutting ring
A case study on productivity improvement of wearing insert and cutting ring
 
Optimization of sealing casting by identifying solidification defect and impr...
Optimization of sealing casting by identifying solidification defect and impr...Optimization of sealing casting by identifying solidification defect and impr...
Optimization of sealing casting by identifying solidification defect and impr...
 
Optimization of sealing casting by identifying solidification defect and impr...
Optimization of sealing casting by identifying solidification defect and impr...Optimization of sealing casting by identifying solidification defect and impr...
Optimization of sealing casting by identifying solidification defect and impr...
 
IRJET- Productivity Improvement by Implementing Lean Manufacturing Tools ...
IRJET-  	  Productivity Improvement by Implementing Lean Manufacturing Tools ...IRJET-  	  Productivity Improvement by Implementing Lean Manufacturing Tools ...
IRJET- Productivity Improvement by Implementing Lean Manufacturing Tools ...
 
Choosing your first AI project. How to get a quick ROI in process industries
Choosing your first AI project. How to get a quick ROI in process industriesChoosing your first AI project. How to get a quick ROI in process industries
Choosing your first AI project. How to get a quick ROI in process industries
 
IE7610_REPORT_GROUP_8
IE7610_REPORT_GROUP_8IE7610_REPORT_GROUP_8
IE7610_REPORT_GROUP_8
 
IRJET- Optimization of Plastic Injection Molding
IRJET- Optimization of Plastic Injection MoldingIRJET- Optimization of Plastic Injection Molding
IRJET- Optimization of Plastic Injection Molding
 
IRJET- Enhance the Capacity of Outer Tube Machining Cell
IRJET-  	  Enhance the Capacity of Outer Tube Machining CellIRJET-  	  Enhance the Capacity of Outer Tube Machining Cell
IRJET- Enhance the Capacity of Outer Tube Machining Cell
 
Implementation of Six Sigma Using DMAIC Methodology in Small Scale Industries...
Implementation of Six Sigma Using DMAIC Methodology in Small Scale Industries...Implementation of Six Sigma Using DMAIC Methodology in Small Scale Industries...
Implementation of Six Sigma Using DMAIC Methodology in Small Scale Industries...
 
IRJET- Design of Spoon Mold using Flow Analysis and Higher End Design Software
IRJET- Design of Spoon Mold using Flow Analysis and Higher End Design SoftwareIRJET- Design of Spoon Mold using Flow Analysis and Higher End Design Software
IRJET- Design of Spoon Mold using Flow Analysis and Higher End Design Software
 
IRJET- A Case Study for Overall Equipment Effectiveness Improved in Manufactu...
IRJET- A Case Study for Overall Equipment Effectiveness Improved in Manufactu...IRJET- A Case Study for Overall Equipment Effectiveness Improved in Manufactu...
IRJET- A Case Study for Overall Equipment Effectiveness Improved in Manufactu...
 
Types of production system (production and operation management)
Types of production system (production and operation management)Types of production system (production and operation management)
Types of production system (production and operation management)
 
How to Reduce Changeover Time and Increase Throughput
How to Reduce Changeover Time and Increase ThroughputHow to Reduce Changeover Time and Increase Throughput
How to Reduce Changeover Time and Increase Throughput
 
Suggestions on the Methodology of Parameters in Fused Deposition Modeling Pro...
Suggestions on the Methodology of Parameters in Fused Deposition Modeling Pro...Suggestions on the Methodology of Parameters in Fused Deposition Modeling Pro...
Suggestions on the Methodology of Parameters in Fused Deposition Modeling Pro...
 
Joined document 24_5
Joined document 24_5Joined document 24_5
Joined document 24_5
 

Recently uploaded

Income Tax exemption for Start up : Section 80 IAC
Income Tax  exemption for Start up : Section 80 IACIncome Tax  exemption for Start up : Section 80 IAC
Income Tax exemption for Start up : Section 80 IAC
CA Dr. Prithvi Ranjan Parhi
 
Top 10 Free Accounting and Bookkeeping Apps for Small Businesses
Top 10 Free Accounting and Bookkeeping Apps for Small BusinessesTop 10 Free Accounting and Bookkeeping Apps for Small Businesses
Top 10 Free Accounting and Bookkeeping Apps for Small Businesses
YourLegal Accounting
 
Innovation Management Frameworks: Your Guide to Creativity & Innovation
Innovation Management Frameworks: Your Guide to Creativity & InnovationInnovation Management Frameworks: Your Guide to Creativity & Innovation
Innovation Management Frameworks: Your Guide to Creativity & Innovation
Operational Excellence Consulting
 
Lundin Gold Corporate Presentation - June 2024
Lundin Gold Corporate Presentation - June 2024Lundin Gold Corporate Presentation - June 2024
Lundin Gold Corporate Presentation - June 2024
Adnet Communications
 
Call8328958814 satta matka Kalyan result satta guessing
Call8328958814 satta matka Kalyan result satta guessingCall8328958814 satta matka Kalyan result satta guessing
Call8328958814 satta matka Kalyan result satta guessing
➑➌➋➑➒➎➑➑➊➍
 
Pitch Deck Teardown: Kinnect's $250k Angel deck
Pitch Deck Teardown: Kinnect's $250k Angel deckPitch Deck Teardown: Kinnect's $250k Angel deck
Pitch Deck Teardown: Kinnect's $250k Angel deck
HajeJanKamps
 
Digital Transformation Frameworks: Driving Digital Excellence
Digital Transformation Frameworks: Driving Digital ExcellenceDigital Transformation Frameworks: Driving Digital Excellence
Digital Transformation Frameworks: Driving Digital Excellence
Operational Excellence Consulting
 
How are Lilac French Bulldogs Beauty Charming the World and Capturing Hearts....
How are Lilac French Bulldogs Beauty Charming the World and Capturing Hearts....How are Lilac French Bulldogs Beauty Charming the World and Capturing Hearts....
How are Lilac French Bulldogs Beauty Charming the World and Capturing Hearts....
Lacey Max
 
The Heart of Leadership_ How Emotional Intelligence Drives Business Success B...
The Heart of Leadership_ How Emotional Intelligence Drives Business Success B...The Heart of Leadership_ How Emotional Intelligence Drives Business Success B...
The Heart of Leadership_ How Emotional Intelligence Drives Business Success B...
Stephen Cashman
 
GKohler - Retail Scavenger Hunt Presentation
GKohler - Retail Scavenger Hunt PresentationGKohler - Retail Scavenger Hunt Presentation
GKohler - Retail Scavenger Hunt Presentation
GraceKohler1
 
Dpboss Matka Guessing Satta Matta Matka Kalyan Chart Satta Matka
Dpboss Matka Guessing Satta Matta Matka Kalyan Chart Satta MatkaDpboss Matka Guessing Satta Matta Matka Kalyan Chart Satta Matka
Dpboss Matka Guessing Satta Matta Matka Kalyan Chart Satta Matka
➒➌➎➏➑➐➋➑➐➐Dpboss Matka Guessing Satta Matka Kalyan Chart Indian Matka
 
Best Competitive Marble Pricing in Dubai - ☎ 9928909666
Best Competitive Marble Pricing in Dubai - ☎ 9928909666Best Competitive Marble Pricing in Dubai - ☎ 9928909666
Best Competitive Marble Pricing in Dubai - ☎ 9928909666
Stone Art Hub
 
list of states and organizations .pdf
list of  states  and  organizations .pdflist of  states  and  organizations .pdf
list of states and organizations .pdf
Rbc Rbcua
 
Chapter 7 Final business management sciences .ppt
Chapter 7 Final business management sciences .pptChapter 7 Final business management sciences .ppt
Chapter 7 Final business management sciences .ppt
ssuser567e2d
 
Best practices for project execution and delivery
Best practices for project execution and deliveryBest practices for project execution and delivery
Best practices for project execution and delivery
CLIVE MINCHIN
 
一比一原版新西兰奥塔哥大学毕业证(otago毕业证)如何办理
一比一原版新西兰奥塔哥大学毕业证(otago毕业证)如何办理一比一原版新西兰奥塔哥大学毕业证(otago毕业证)如何办理
一比一原版新西兰奥塔哥大学毕业证(otago毕业证)如何办理
taqyea
 
Brian Fitzsimmons on the Business Strategy and Content Flywheel of Barstool S...
Brian Fitzsimmons on the Business Strategy and Content Flywheel of Barstool S...Brian Fitzsimmons on the Business Strategy and Content Flywheel of Barstool S...
Brian Fitzsimmons on the Business Strategy and Content Flywheel of Barstool S...
Neil Horowitz
 
Dpboss Matka Guessing Satta Matta Matka Kalyan panel Chart Indian Matka Dpbos...
Dpboss Matka Guessing Satta Matta Matka Kalyan panel Chart Indian Matka Dpbos...Dpboss Matka Guessing Satta Matta Matka Kalyan panel Chart Indian Matka Dpbos...
Dpboss Matka Guessing Satta Matta Matka Kalyan panel Chart Indian Matka Dpbos...
➒➌➎➏➑➐➋➑➐➐Dpboss Matka Guessing Satta Matka Kalyan Chart Indian Matka
 
2024-6-01-IMPACTSilver-Corp-Presentation.pdf
2024-6-01-IMPACTSilver-Corp-Presentation.pdf2024-6-01-IMPACTSilver-Corp-Presentation.pdf
2024-6-01-IMPACTSilver-Corp-Presentation.pdf
hartfordclub1
 
Business storytelling: key ingredients to a story
Business storytelling: key ingredients to a storyBusiness storytelling: key ingredients to a story
Business storytelling: key ingredients to a story
Alexandra Fulford
 

Recently uploaded (20)

Income Tax exemption for Start up : Section 80 IAC
Income Tax  exemption for Start up : Section 80 IACIncome Tax  exemption for Start up : Section 80 IAC
Income Tax exemption for Start up : Section 80 IAC
 
Top 10 Free Accounting and Bookkeeping Apps for Small Businesses
Top 10 Free Accounting and Bookkeeping Apps for Small BusinessesTop 10 Free Accounting and Bookkeeping Apps for Small Businesses
Top 10 Free Accounting and Bookkeeping Apps for Small Businesses
 
Innovation Management Frameworks: Your Guide to Creativity & Innovation
Innovation Management Frameworks: Your Guide to Creativity & InnovationInnovation Management Frameworks: Your Guide to Creativity & Innovation
Innovation Management Frameworks: Your Guide to Creativity & Innovation
 
Lundin Gold Corporate Presentation - June 2024
Lundin Gold Corporate Presentation - June 2024Lundin Gold Corporate Presentation - June 2024
Lundin Gold Corporate Presentation - June 2024
 
Call8328958814 satta matka Kalyan result satta guessing
Call8328958814 satta matka Kalyan result satta guessingCall8328958814 satta matka Kalyan result satta guessing
Call8328958814 satta matka Kalyan result satta guessing
 
Pitch Deck Teardown: Kinnect's $250k Angel deck
Pitch Deck Teardown: Kinnect's $250k Angel deckPitch Deck Teardown: Kinnect's $250k Angel deck
Pitch Deck Teardown: Kinnect's $250k Angel deck
 
Digital Transformation Frameworks: Driving Digital Excellence
Digital Transformation Frameworks: Driving Digital ExcellenceDigital Transformation Frameworks: Driving Digital Excellence
Digital Transformation Frameworks: Driving Digital Excellence
 
How are Lilac French Bulldogs Beauty Charming the World and Capturing Hearts....
How are Lilac French Bulldogs Beauty Charming the World and Capturing Hearts....How are Lilac French Bulldogs Beauty Charming the World and Capturing Hearts....
How are Lilac French Bulldogs Beauty Charming the World and Capturing Hearts....
 
The Heart of Leadership_ How Emotional Intelligence Drives Business Success B...
The Heart of Leadership_ How Emotional Intelligence Drives Business Success B...The Heart of Leadership_ How Emotional Intelligence Drives Business Success B...
The Heart of Leadership_ How Emotional Intelligence Drives Business Success B...
 
GKohler - Retail Scavenger Hunt Presentation
GKohler - Retail Scavenger Hunt PresentationGKohler - Retail Scavenger Hunt Presentation
GKohler - Retail Scavenger Hunt Presentation
 
Dpboss Matka Guessing Satta Matta Matka Kalyan Chart Satta Matka
Dpboss Matka Guessing Satta Matta Matka Kalyan Chart Satta MatkaDpboss Matka Guessing Satta Matta Matka Kalyan Chart Satta Matka
Dpboss Matka Guessing Satta Matta Matka Kalyan Chart Satta Matka
 
Best Competitive Marble Pricing in Dubai - ☎ 9928909666
Best Competitive Marble Pricing in Dubai - ☎ 9928909666Best Competitive Marble Pricing in Dubai - ☎ 9928909666
Best Competitive Marble Pricing in Dubai - ☎ 9928909666
 
list of states and organizations .pdf
list of  states  and  organizations .pdflist of  states  and  organizations .pdf
list of states and organizations .pdf
 
Chapter 7 Final business management sciences .ppt
Chapter 7 Final business management sciences .pptChapter 7 Final business management sciences .ppt
Chapter 7 Final business management sciences .ppt
 
Best practices for project execution and delivery
Best practices for project execution and deliveryBest practices for project execution and delivery
Best practices for project execution and delivery
 
一比一原版新西兰奥塔哥大学毕业证(otago毕业证)如何办理
一比一原版新西兰奥塔哥大学毕业证(otago毕业证)如何办理一比一原版新西兰奥塔哥大学毕业证(otago毕业证)如何办理
一比一原版新西兰奥塔哥大学毕业证(otago毕业证)如何办理
 
Brian Fitzsimmons on the Business Strategy and Content Flywheel of Barstool S...
Brian Fitzsimmons on the Business Strategy and Content Flywheel of Barstool S...Brian Fitzsimmons on the Business Strategy and Content Flywheel of Barstool S...
Brian Fitzsimmons on the Business Strategy and Content Flywheel of Barstool S...
 
Dpboss Matka Guessing Satta Matta Matka Kalyan panel Chart Indian Matka Dpbos...
Dpboss Matka Guessing Satta Matta Matka Kalyan panel Chart Indian Matka Dpbos...Dpboss Matka Guessing Satta Matta Matka Kalyan panel Chart Indian Matka Dpbos...
Dpboss Matka Guessing Satta Matta Matka Kalyan panel Chart Indian Matka Dpbos...
 
2024-6-01-IMPACTSilver-Corp-Presentation.pdf
2024-6-01-IMPACTSilver-Corp-Presentation.pdf2024-6-01-IMPACTSilver-Corp-Presentation.pdf
2024-6-01-IMPACTSilver-Corp-Presentation.pdf
 
Business storytelling: key ingredients to a story
Business storytelling: key ingredients to a storyBusiness storytelling: key ingredients to a story
Business storytelling: key ingredients to a story
 

Improvement strategy by using ML and TM. A case study for solving cracking in sand casting

  • 1. Improving productivity and yield by using combination of Taguchi method and Machine learning techniques CONFIDENTIAL AND PROPRIETARY Any use of this material without specific permission is strictly prohibited 1 Presentationby JagadishC.A. jagadish.chandra@qmaxim.com 20th of Feb., 2015, v1.5 Illustrated in a case study related to manufacture of casting by sand casting process
  • 2. CASE STUDY PRESENTED HERE ILLUSTRATES A NEW APPROACH FOR SOLVING THIS KIND OF PROBLEM. KNOWLEDGE OF METALLURGY, PROCESS & CASTING IS COMBINED WITH ADVANCED STATISTICAL OPTIMIZATION TECHNIQUES OF MACHINE LEARNING AND TAGUCHI METHOD IS USED. PROBLEMS CAN SOLVED MORE QUICKLY, ECONOMICALLY & WITH GREATER CERTAINITY. PROBLEMS LIKE CRACKING IN CASTINGS IS TRADITIONALLY SOLVED BY TRIAL AND ERROR METHOD USING KNOWLEDGE OF METALLURGY, PROCESS & CASTING. THIS IS A HIT & MISS PROCESS. 2
  • 3. BY READING THIS PRESENTATION ONE CAN GET SOME INTUITION ABOUT THE APPROACH & HOW ONE CAN APPLY IT IN THEIR OWN WORK 3
  • 4. content 4 •Case study phase #2 – DOE by Taguchi Method & advanced data analysis •Appendix-1: Primer on Machine learning & Taguchi Method •Case study phase #1 –exploratory analysis Setting up context-Setting up context- economicreality, challenges, opportunities •Phase #2– confirmation •Findings, Conclusions, summary •Questions,background, Contacts •Overview of new approach, case study background •Appendix-2: Sand casting some resources
  • 5. The manufacturing reality manufacturing sector-many challenges • Still low growth rates in manufacturing • Customer requirements getting tougher • Severe competition and pressure on margins • Continuouspressure to reduce costs • Increasing complexity, automation,increasing need for new skills • Younger inexperienced but ambitious workforce • Continuouspressure for rapid & big improvements
  • 6. The manufacturing issues data analysis & continuous improvement challenges • Data deluge –vast amounts of data collected but not much analysis. Many sources of data – SCADA, CRM,ERP, customer data, handwritten notes, Excel., social networks.... • Data in electronic or paper form, many formats – quite often not analysed • Internet of things becoming reality – more data, more communication- increasing complexity • too many variables – how to reduce the number to do analysis? • variables are not just quantitative but categorical or ordered categorical . • Simple data analysis tools less and less useful e.g. Simple Regression • Continuous improvement based on Kaizen, 7 basic tools for quality improvement, Shainin DOE are useful but not that effective for big improvements
  • 7. The manufacturing reality many challenges but also opportunities... • Classical vary one factor at time approach has severe limitations • Choosing operating conditions to balance productivity & yield is not easy • Traditional six sigma does not incorporate many of the new tools and methodologies which are now available • But, this is also an opportunity...... – Unheard of granularity & richness of data nowadays – Falling cost of computation, storage – Powerful machine learning tools are available for data analysis & optimization (many of them open source & free)
  • 8. 8 content 8 •Case study phase #2 – DOE by Taguchi Method & advanced data analysis •Appendix-1: Primer on Machine learning & Taguchi Method •Case study phase #1 –exploratory analysis •Setting up context- economicreality, challenges, opportunities •Phase #2– confirmation •Findings, Conclusions, summary •Questions,background, Contacts ••Overview of new approach, case study background •Appendix-2: Sand casting some resources
  • 9. Outline of the new methodology is as follows • New methodology – Uses combination of machine learning & Taguchi method – Can find best operating conditions rapidly without compromising productivity • Illustrated with a case study related to sand casting, but can be applied to any manufacturing situation (Why sand casting chosen? Sand casting is very noisy process with many variables, uses highly variable natural substances & is manual process – if it can work in sand casting it can work in any manufacturing situation ) – see next 3 slides for overview of methodology – case study in the subsequent slides 9
  • 10. manufacturingimprovementstrategy outline main objectives and major issues manufacturing faces 1. Improveyield by reducing defectssuch as cracking, pin holes, bad surface...... Mainobjectives of manufacturing 2. Improveproductivity & cost by reducing time taken by operating at highest possible speed and reducing inputs as much as possible How to set process parameters? e.g. Temperature, degassing rate, speed – start and steady, time- start, ramp up... Make any major changes? e.g. Equipment change, mould design, invest? Majorissues in manufacturing
  • 11. manufacturingimprovementstrategy outline main objectives and major issues manufacturing faces Classical approach to tackle this problem which most companies follow 1. Improveyield by reducing defectssuch as cracking, pin holes, bad surface...... Mainobjectives of manufacturing 2. Improveproductivity & cost by reducing time taken by operating at highest possible speed and reducing inputs as much as possible How to set process parameters? e.g. Temperature, degassing rate, speed – start and steady, time- start, ramp up... Make any major changes? e.g. Equipment change, mould design .. at a time holding others constant validity Do trials by varying one factor at a time holding others constant Problem: Many parameters – too many trials. Results have no statistical validity Quite often investments are madeQuite often investments are made without utilising potential of existing equipment Majorissues in manufacturing ClassicalapproachClassicalapproach
  • 12. manufacturing improvement strategy outline main objectives and major issues manufacturing faces Classical approach which most companies follow new approach has three phases as outlined 1. Improveyield by reducing defectssuch as cracking, pin holes, bad surface...... Mainobjectives of manufacturing 2. Improveproductivity & cost by reducing time taken by operating at highest possible speed and reducing inputs as much as possible How to set process parameters? e.g. Temperature, degassing rate, speed – start and steady, time- start, ramp up... Make any major changes? e.g. Equipment change, mould design, modernization .. Phase #1Phase #1 Do exploratory data analysis existing data using linear regression & machine learning. Gain insights & short list important variables Do sophisticated data analysis and find best operating Phase #2 1. Do trials by Statistical experimentation by making changes to process parameters and system. 2. Do sophisticated data analysis and find best operating conditions Majorissues in manufacturing New approachNew approach
  • 13. manufacturing improvement strategy outline main objectives and major issues manufacturing faces Classical approach which most companies follow new approach has three phases as outlined 1. Improveyield by reducing defectssuch as cracking, pin holes, bad surface...... Mainobjectives of manufacturing 2. Improveproductivity & cost by reducing time taken by operating at highest possible speed and reducing inputs as much as possible How to set process parameters? e.g. Temperature, degassing rate, speed – start and steady, time- start, ramp up... Make any major changes? e.g. Equipment change, mould design, modernization .. Phase #1Phase #1 Do exploratory data analysis existing data using linear regression & machine learning. Gain insights & short list important variables Do sophisticated data analysis and find best operating Phase #2 1. Do trials by Statistical experimentation by making changes to process parameters and system. 2. Do sophisticated data analysis and find best operating conditions Phase #3 Decide whether to make major investments Majorissues in manufacturing New approachNew approach Phase #1,#2 covered in in the following slides
  • 14. Outline of the new methodology, it has following steps 1. Do exploratory data analysis by using many modelling algorithms - linear regression, cluster analysis & LASSO (or other machine learning algorithms) and gain insights and shortlist variables of importance (Phase #1) 2. Use Taguchi Method (TM) to find important factors & their best levels (DOE) by doing planned manufacturing runs & analysing data (Phase #2) ( what is machine learning & TM? see appendix-1) Continued... 17
  • 15. Outline of the new methodology, it has following steps 3. Confirm findings by doing actual manufacturing runs using optimum conditions found in #2 (Phase #2) 4. Use entire data and build models by using linear regression & LASSO (or other machine learning algorithms) – for further insights and verify efficacy of findings (Phase #2) 5. Operate under optimum operating conditions found in steps #1to #5 & monitor performance on a long term basis Note : these steps marked as: on the slides 18 1
  • 16. Illustrative case study details • Objective : – Reduce cracking significantly in a mild steel part of a particular design manufactured by sand casting – Also, significantly reduce overallconsumption of welding electrode for repair (additional details next slide) (what is sand casting? See Appendix -2) 19 0
  • 17. Illustrative case study details • Manufacturing Process : Sand casting • Operating mode: completely manual • Design : pipe joint, mild steel • Complicated production process (2 box, cored) • Project relates to cracking seen in a certain design during 1st stage inspection after pouring, shake out, fettling, heat treatment, shot blasting (see schematic of process & some pictures in following slides) • Magnetic particle inspection (MPI) required as crack is not visible to naked eye. Defectiveportions have to be repaired • Almost all castings crack • Cracking has severe cost implications –repair, welding consumables,heat treatment, lost time.... 20 0
  • 18. Case study: Reducing cracking in pipe joint cast by sand casting Schematic process flow rework by welding at 1st stage inspection, Amplification effect- for each 1kg of casting got to cast 1.9 of metal 21 1st stage inspection,1st stage inspection, defectdetection& repair 0
  • 19. Illustrative case study details –some pictures Some Process steps –2 piece box bottom with chills 23 0 Bottomhalf of mould with chillsseen Core coated and ready for insertion Metalpoured into mould
  • 20. Illustrativecase study details cracks almost always occur at approximately the same place 24 0 Cracks marked after inspection. Cracks occur mostly at the same location Crack runs throughout the section of pipe
  • 21. 25 content 25 •Case study phase #2 – DOE by Taguchi Method & advanced data analysis •Appendix-1: Primer on Machine learning & Taguchi Method Case study phase #1Case study phase #1 – exploratory analysis •Setting up context- economicreality, challenges, opportunities •Phase #2– confirmation •Findings, Conclusions, summary •Questions,background, Contacts •Overview of new approach, case study background •Appendix-2: Sand casting some resources
  • 22. Phase #1 overview varioussteps, sourceof knowledge following slides havedetails Furnacecontent, operating conditions, chemistry records 14 months,100 furnaceloads, 4-6 designs / cast =6500 records Rework records Technical information Cleanup, combine, createnew variables, separate information aboutcrack pronedesign of interest Exploratory graphs. Advanced statistical analysis – regression, machinelearning -model building Interpretations, Findings, presentations Get insights, shortlist importantvariables & decide on next steps Metallurgy casting Machine learning 1
  • 23. Effect of Variables on response variable studied Companycollects vast amountsof data furnace content, schedule, chemistry, operating conditions....studied (~27 variables) • PartName • Fur No • Furnace content • Week day • shift • grade • Pour Time Slot • Tap temp • Mould Hardness • coating mould • Core Hardness • coating core • Piece weight to total Liquid ratio • Ladle temp. • C Mn Si S P • Cr Ni Mo Cu V • Carbon equivalent – Conventional • Carbon equivalent – Japan 1
  • 24. Some Exploratory graphs- histograms tap temp, core hardness, mould hardness, ladle temperature have 2 distinct groups 1
  • 25. Some Exploratory graphs of chemistry -histograms Mostly near normal distribution, some large values seen 1
  • 26. Exploratory graphs - histograms Ni,Mo,Cu, skewed with some large values 1
  • 27. Exploratory ECDF (empirical cumulative distribution function) graphs of variables 2 furnaces show different behaviours w.r.t. tap and ladle temperature this is probably the reason for grouping seen Furnace B Furnace A 1
  • 28. ExploratoryECDF graphs –mould/ core hardness different mould coatings show highly different hardness- this is reason for groups seen cotpol Ceramol 930 1
  • 29. Some of the Exploratorybox plot graphs of operatingconditions on cracking tendency higher tap temperature, higher mould hardness, higher ladle temperature are better 1
  • 30. cluster dendrogram done to group similar variables in entire dataset Relationship ofcracking with various variablesmarked – zoomed view next slide 1
  • 31. Cluster dendrogram , area of interest zoomed in - relationshipof cracking with various variables cracking tendency groups with shift, tap (melt) temperature,ladletemp, furnace No 1
  • 32. Linear Regression analysis of data set between cracking built ANOVA table created the table by introducing each of the terms in the model one at time Variables of importance highlighted Analysis of Variance Table Response: CracksP Df Sum Sq Mean Sq F value Pr(>F) Fur_No 1 11.073 11.0733 18.0554 4.657e-05 *** Paint_mld 1 3.529 3.5291 5.7544 0.0182110 * Paint_core 3 2.266 0.7552 1.2314 0.3020912 Tap_temp 1 17.235 17.2349 28.1022 6.408e-07 *** Mld_Hardness 1 0.525 0.5252 0.8564 0.3568672 Core_Hardness 1 0.126 0.1258 0.2052 0.6515131 Ladle_temp 1 0.949 0.9493 1.5479 0.2162225 C 1 8.978 8.9779 14.6388 0.0002214 *** Mn 1 0.204 0.2045 0.3334 0.5649174 Si 1 0.122 0.1219 0.1987 0.6566789 S 1 2.260 2.2598 3.6846 0.0576309 . P 1 1.143 1.1432 1.8640 0.1750845 Cr 1 0.254 0.2536 0.4135 0.5215845 Ni 1 3.314 3.3135 5.4028 0.0220296 * Mo 1 0.005 0.0048 0.0078 0.9297360 Cu 1 4.506 4.5063 7.3477 0.0078453 ** V 1 0.116 0.1155 0.1883 0.6652000 P2_L_ratio 1 4.742 4.7424 7.7327 0.0064290 ** C:Ni 1 1.200 1.1995 1.9559 0.1648990 Cr:Ni 1 1.871 1.8714 3.0514 0.0835905 . Residuals 105 64.396 0.6133 --- Signif. codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1 1
  • 33. Model built using LASSO (generalized linear model via penalized maximum likelihood ) regression output shown The algorithm keeps coefficients of only important variables (Intercept)Fur_NoB Paint_mldCotpol Tap_temp Ladle_temp C -0.4368 .4353 0.3358 -0.4066 -0.2325 0.1821 S Ni Cu P2_L_ratio 0.1313 0.2184 -0.1775 -0.1882 > coef(cvfit, s = "lambda.min") 30 x 1 sparse Matrix of class "dgCMatrix" 1 (Intercept) 7.888494721 no_pcs -0.002647573 Tap_temp -0.003587278 Mld_Hardness . Core_Hardness . Ladle_temp -0.001463345 C 1.240515245 Mn . Si . S 1.225540417 P 1.746826872 Cr -0.100157651 Ni 0.595015164 Mo 0.270251629 Cu -1.480281443 V . P2_L_ratio -0.129797323 CEa . CEj . Fur_NoB 0.051100821 Fur_NoC . Paint_mldCotpol 0.002589163 Paint_mldHolcoat . Paint_mldIsomol . Paint_mldSNS . Paint_coreCotpol 0.020449112 1
  • 34. Summary of some of the findings from phase #1 distinct groupings seen, reason found • distinct groups seen – tap temp, ladle temp, mould hardness, core hardness, furnace modifiers ratio, casting sequence • tap temp, ladle temp depends on furnace used, casting time slot, weekday. • mould hardness, core hardness depends coating type • Chemistry many elements skewed, contamination seen • furnace modifiers ratio – large difference in ratios between furnaces 1
  • 35. Conclusions, insights, summary operating conditions, chemistry are correlated with cracking • Several algorithms used to shortlist variables of importance and their effect on cracking • Findings: – almost similar conclusion can be drawn from output of all algorithms – Some of the variable have significant effect on cracking e.g. Melt temperature, ladle temperature , furnace No... – Effectof variables on cracking is as follows: Tap temp (+), ladle temp (+), mould hardness (+), core hardness (+), furnaceA ( +), time slot (-), weekday (-), modifiers ratio (+/-), casting sequence(mid better), piece to liquid ratio (+) Chemistry – C(-),Mn(-),S(-),P(-),Cr(+),Ni(-),Mo(-), Cu(+),Carbon Eq. (-) ( Coding :- + higher level better, - lower level better) 1
  • 36. Conclusions, insights, summary operating conditions, chemistry are correlated with cracking • Significantinsights gained from phase #1 by studying existing data without making any process change • Other aspects like methoding, raw material, operating conditions-delay, ladle conditions, pouringheight, ambient humidity etc may contributeto cracking • Detailed industrial study such as DOE needs to be done to get even better insight to solve this problem (to be done in phase #2) 1
  • 37. 41 content 41 •Case study phase #2 – DOE by Taguchi Method & advanced data analysis •Case study phase #2 – DOE by Taguchi Method & advanced data analysis •Appendix-1: Primer on Machine learning & Taguchi Method •Case study phase #1 – exploratory analysis •Setting up context- economicreality, challenges, opportunities •Phase #2– confirmation •Findings, Conclusions, summary •Questions,background, Contacts •Overview of new approach, case study background •Appendix-2: Sand casting some resources
  • 38. Methodology followed (phase #2) Overview of steps – following slides give details of each box Make general observations about the problem –frequency, where, when, macro, … Choose a DOE experimental plan based on phase #1 conclusions, new considerations & cost aspects - factors to vary, their levels , interactions , number of replicates per run Taguchi DOE chosen Factors – chemistry (pure, impure), tap temp (low, high), ladle treatment( high, low), methoding -design (old, new) Do casting runs (8x4) according to plan Collect data Identify each casting Strip casting, section, HT, grit blast, prepare for MPI Make Interpretations, findings, recommendations Make presentation Measure for each casting -crack length by MPI & DP and record Data entry to specialized software, analyse data and generate reports and charts (mean , S & N ) Metallurgy casting Machine learning Build regression and machine learning models using all individual data Do casting runs using optimum levels of factors Measure crack length by MPI & DP and record 2
  • 39. Phase #2 Steps in finding important factors & their best levels by doing TM casting runs. Overview of experimental plan was as follows, details in following slides 44 1. methoding 2.chemistry 3. tap temp 4. Ladle treatment Factors chosen As many as 150 variables. Other factors heldconstant/monitored: e.g. Sand properties, binder properties, mould preparationmethod, mouldcoatings, furnace content, furnace treatment, ambient temp. , relative humidity.......... Design chosen Experimental design (L8), (27) array, 8 runsin duplicate (2x8x2=32 castings) Experiments run Response & other data collected Data analysed Program output: response table and charts (mean & S/N) Conclusions 1. Factorsof significance 2. Interactions of significance 3. Optimum factor levels Response : length of crack measured by MPI Note: 1.Morethefactors/levels chosen larger will be the number of experimental runsto be done, hence more expensive it becomes. 2. Do production run as per the design chosen 2
  • 40. Castingexperimental plan based on DOE plan shown factorswith their levels, interactions, response variable Chemical analysis,tap temp, ladletemp, fill time, start/end time, preparationdelay,knockoutdelay, etc noted down Response: Crack length Interactionsalso estimated : chemistry with melt temp, melt temp with ladle treatment chem furnace melt temp ladle treatment methoding design DOE_Run_No pure high (1625-1640) high old 1 pure high (1625-1640) low new 2 impure low (1590-1605) high old 3 impure low (1590-1605) low new 4 pure low (1590-1605) high new 5 pure low (1590-1605) low old 6 impure high (1625-1640) high new 7 impure high (1625-1640) low old 8 Factorstested: Chemistry furnace, melt temp, methoding design*, ladle treatment * See next slide for details 2
  • 41. Methoding–design changes made during phase #2 Old design shown in picture, changes made iteratively to some of the chills and a risers 2
  • 42. DOE-means plots findings Modified mould design is significantly better, higher levels of ladle treatment, higher level of melt temp & impure chemistry better 2
  • 44. DOE findings - ANOVA of means output Modified mould design is significantly better & the most importantamong factors. Interactionsof chem. analysis with melt temp and melt temp with ladle treatment are also significant Analysis of Variance for Means Source DF Seq SS Adj SS Adj MS F P chem furnace 1 35.596 35.596 35.596 41.33 0.098 melt temp 1 17.627 17.627 17.627 20.46 0.138 ladle treatment 1 104.221 104.221 104.221 121.00 0.058 mould design 1 397.268 397.268 397.268 461.23 0.030 chem furnace*melt temp 1 265.939 265.939 265.939 308.76 0.036 melt temp*ladle treatment 1 176.955 176.955 176.955 205.44 0.044 Residual Error 1 0.861 0.861 0.861 Total 7 998.467 Response Table for Means chem ladle mould Level furnace melt temp treatment design 1 36.94(P) 33.34 (H) 38.44(high) 41.88 (old) 2 32.72(IP) 36.31 (L) 31.22 (low) 27.78 (new) Delta 4.22 2.97 7.22 14.09 Rank 3 4 2 1 2
  • 45. General conclusions on cracking susceptibility as per DOE findings & best operating conditions are as follows • Best operating conditions: – New methoding design is significantlybetter than old design – Higher melt temperature range is better – Higher levels of ladle treatment is better – Recommended chemical composition is impure, but effectis small between pure and impure. – Also, interaction between chem. Composition/ ladle treatmentwith melt temperature is important • Several casting runs were done under the recommended conditions to verify findings • Overall, 4 iterations of new designs also tried 3
  • 46. Linear Regression was done of the complete dataset ANOVA table output Mould design, sequence of casting, S,P, ladle temperature significant Analysis of Variance Table Response: crack length Df Sum Sq Mean Sq F value Pr(>F) mould.design 4 4684.0 1170.99 12.1631 1.86e-05 *** sequence 3 833.7 277.91 2.8867 0.057489 . C 1 0.8 0.77 0.0080 0.929651 S 1 782.3 782.25 8.1253 0.009053 ** P 1 606.3 606.28 6.2974 0.019581 * Ni 1 141.7 141.70 1.4719 0.237362 Cr 1 8.4 8.45 0.0878 0.769705 ladleTemp 1 432.5 432.53 4.4927 0.045046 * Residuals 23 2214.3 96.27 --- Signif. codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1 4
  • 47. Modelwas also built using LASSO (generalized linear model via penalized maximum likelihood ) regression of completedata set Outputis seen below The algorithm drops variables which are not important, important variables shown in the table below Coefficients generated by LASSO after cross-validation 3 x 1 sparse Matrix of class "dgCMatrix" 1 (Intercept) -472.51096012 C . S 1594.72775370 P 3218.85844853 Ni -50.26465017 Cr . Si . Mn . Cu . Mo . V . Al . Ti 8370.87631400 Fe.silinium . tapTemp -0.05934132 ladleTemp 0.32709791 mould.designnew1 -15.17664341 mould.designnew2 -1.47706598 mould.designnew3 -41.25121292 mould.designnew4 -43.62759697 sequence2 -7.92797995 sequence3 -12.29255408 sequence4 -3.85168798 4
  • 48. 53 content 53 •Case study phase #2 – DOE by Taguchi Method & advanced data analysis •Appendix-1: Primer on Machine learning & Taguchi Method •Case study phase #1 – exploratory analysis •Setting up context- economicreality, challenges, opportunities ••Phase #2– confirmation •Findings, Conclusions, summary •Questions,background, Contacts •Overview of new approach, case study background
  • 49. Severalcasting runs weremade under optimum operating conditions to verify findings BoxPlots comparison of most important factor– methoding design shown New designs are significantly better, Iteration new #4 is best of the lot under optimum operating conditions Longer casting runs will be produced using the optimum conditions found to evaluate long term performance 4 5
  • 50. 55 content 55 •Case study phase #2 – DOE by Taguchi Method & advanced data analysis •Appendix-1: Primer on Machine learning & Taguchi Method •Case study phase #1 – exploratory analysis •Setting up context- economicreality, challenges, opportunities •Phase #2– confirmation ••Findings, Conclusions, summary •Questions,background, Contacts •Overview of new approach, case study background •Appendix-2: Sand casting some resources
  • 51. OVERALL CONCLUSIONS 56 • Approach on use of Taguchi method and Machine learning combination for yield & productivity improvement was explained. Some background information was also presented • Illustrative case study related to sand casting manufacturing process was presented. • Use of various algorithms to quickly gain valuable insights and to find variables of importance and their effect was illustrated • Optimum operating conditions found from both the phases were revalidated in actual production run Continued.....
  • 52. OVERALL CONCLUSIONS 57 • This project required relatively limited effort, was of low cost, was of short duration, did not disrupt operations significantly • By using this approach, it is possible to efficiently find optimum operating conditions even in highly noisy, completely manual process with large number of variables • It is also possible to use the information gained by this approach to decide on prioritizing areas of investment for modernization
  • 53. 58 content 58 •Case study phase #2 – DOE by Taguchi Method & advanced data analysis •Appendix-1: Primer on Machine learning & Taguchi Method •Case study phase #1 – exploratory analysis •Setting up context- economicreality, challenges, opportunities •Phase #2– confirmation •Findings, Conclusions, summary ••Questions,background, Contacts •Overview of new approach, case study background
  • 54. THANKS! QUESTIONS? 59 Jagadish C.A. (Rao) jagadish.chandra@qmaxim.com qmaxim001@gmail.com +91 9900 606620 +91 9538 328704 blog: qmaxim.wordpress.com Twitter: @JagSpeak q-Maxim
  • 55. Jagadish C.A. ‘s Profile 60LinkedIn profile: http://goo.gl/Lp3lWv • Heads niche consultancy q-Maximfocused on adv. Optimization, quality, technology…. • B.Tech (MetallurgicalEngg., NIT-K,Surathkal,India) • Done many graded online courses (4-15weeks) from leading US universities on operationsmanagement, advanced data analysis, marketing, finance, accounting, strategy,advancedcompetitive strategy,data scientist (5 courses),credit risk management,game theory, logistics • ASQ (American society for quality) certified Six sigma Black belt (since 2002) • ASQ certified Manager of Org. Excellence/ Quality (since 1999) • Certified EFQMassessor extensive experience assessing companies, fashioning transformationalroadmap& implementing EFQMmodel • JuranQI facilitator,Cert. adv. Industrial experimentation, Analytics • ISO 9001:2008,14001,QS-9000(TS16949)lead auditor • ~32years experience in Manufacturingandtransactionalfields, 7.5 years as managementconsultant • Richexperience in Quality, process management, R +D, Technology, Cost reduction, Managementconsultancy.Extensive experience in using advanced methodologies for problem solving and optimization • Extensive exposure/ knowledge of heavy process oriented manufacturing – Aluminium casthouse, Steel, Welding, casting, foundry, Metallurgy, etc • Widely travelled – India, north America, Europe, M.E.
  • 56. 61 content 61 •Case study phase #2 – DOE by Taguchi Method & advanced data analysis ••Appendix-1: Primer on Machine learning & Taguchi Method •Case study phase #1 – exploratory analysis •Setting up context- economicreality, challenges, opportunities •Phase #2– confirmation •Findings, Conclusions, summary •Questions,background, Contacts •Overview of new approach, case study background •Appendix-2: Sand casting some resources
  • 57. Whatis Machine learning ? Some definitions • Definition : “A computer is able to learn by experience without explicitly being programmed – & improves performance as it learns” • Based on field of artificial intelligence • Examples : – Mining data from large datasets website click trough data to improve purchase conversion rate – Autonomous self flying helicopter (Stanford University) – Classify e-mail as spam or not spam (filtering spam in outlook.com) – handwriting recognition (tablets) – Computer Vision (reading car number plates & giving speeding tickets) – Self driven cars (Google self driving car) – Recommendersystems (Amazon recommending books) • Not that common in manufacturing.... 62
  • 58. Whatis Machine learning .. 2 major types - Predictive & descriptivetypes • Predictive learning (supervised learning) there is outcome variable to guide the learning process – Learning phase learning by example. Tune the model until error is sufficiently low – Scoring phase Use the model for making predictions (or score) in real time e.g.: linear regression, polynomial regression, LASSO, random forest, Neural network, Decision tree • Descriptive (Unsupervised learning) – No outcome variable – Self-organization, no teacher – Groups observations (variables) based on similarity, to uncover hidden relationships – e.g.: clustering • Regression or classification problem based on type of outcome 63
  • 59. Machine learning example –data in simple table form Learning algorithm for predicting house prices- various parts labelled 64 Row no Area [sq. Ft.] Number of rooms Age of flat [years] Gym [Y/N] Swim ming pool [Y/N] ............... Other features not shown............. Market price ( lakh Rupees) 1 1800 5 1.1 yes yes 68.6 2 900 3 4 no no 34.5 3 1720 5 8 yes no 47.7 4 560 2 .7 no no 25.4 ..... 1000 2400 6 3 yes yes 91.8 CalledTarget or outcome or outputCalledPredictors or inputsor features Records orrows
  • 60. Machine learning– supervised learning -overview example- predicting market price of house using simple linear learning algorithm 65 SampledTrainingdataset Known 1. Area of house 2. Number of rooms 3. Age of house 4. Location 5. Gym [y/n] 6. ..... Etc, etc Learning algorithm predictive hypothesis h(x) Prediction marketprice of house Calledtarget or outcomeCalledfeatures or predictors h(x)is a linear equationof the type: hθ(x) = θ0+ θ1x1 + θ2x2 +....... Θnxn Past data of housing market having features & predictors Learning phase scoring phase
  • 61. Unsupervised learning – K means cluster analysis and dendrogram Groups observations (variables)based on similarity, to uncoverhidden relationships • K-means clustering- a type of unsupervised learning • Example: Clustering of wine database from 1996, of wines grown in the same region of Italy, but derived from three different types ( cultivars ) Databasehas 78 observations, 13 chemical analysis variables & 1 column of 3 cultivars Source :University of California at Irving (UCI) Machine Learning Depository • We want the machine learning algorithm to cluster the data & uncover the types without seeing the type ( cultivars ) & check the accuracy
  • 62. Unsupervised learning – K means cluster analysis It has classified 178 observations into 3 clusters using 13 chemical analysis variables Data: Universityof California at Irving (UCI)Machine Learning Depository
  • 63. Unsupervised learning – K means cluster analysis- colored based on type (cultivar) classificationaccuracy is pretty good as seen by coloring the cluster on type(cultivar) of wine
  • 64. Supervised learning – decision tree a simple predictive learning algorithm Decision tree to predict –’what possibility of a person surviving Titanic sinking given his/her age,sex,sibling?’ 69 A tree showing survivalof passengers on the Titanic ("sibsp" is the number of spouses or siblings aboard). Thefigures under the leaves show the probability of survival and the percentage of observations in the leaf. Source:WIKIPEDIA
  • 65. Supervised learning – Random forest a predictivealgorithm based on decision tree Exampleof random forest use for classifying land type using Landsat satelliteThermal InfraredSensor image of 4 spectral bands The algorithm can classify landmass automaticallyafter it has ‘learnt’ by seeing known data 70Credit:internet resources
  • 66. Supervised learning – Artificial Neural network predictive learning algorithm Mimic working of brain and nervous system, Ideal for modelling complex relationships Use of artificial neuralnetwork to do handwriting recognition. 71Credit:internet resources
  • 67. Want to know more about data mining and machine learning? Read my presentation titled ‘Data mining and Machine Learning - in jargon free, lucid language’ .
  • 68. Taguchi method (TM) is very popular optimization method. Some details and advantages • TaguchiMethod (TM) is very common optimization method in science & engineering for over 40 years • Has following components - System design, Tolerance design and parameter design • We are using parameter design • TM is not a theoretical exercise, done in production environment, justsomeadditional measurements • TM has many advantages : – Less effort, economicalthan T & E & full factorial – much more insight – Faster – Canfind important factors, their best levels& interactions – Also Signal to Noise to find best levelsfor minimizingvariation – Ableto decide on parameterskeeping productivityin mind – Findingshave Statisticalvalidity 73
  • 69. 74 • Factors : – Control: best levelsfor desirable output : e.g. Pouring temperature – Signal: can influence,can be controlled : e.g. volume knob position – Noise : can’t be controlledOr are intentionallynotcontrollede.g. ambient temperature – Others: not changing during experimentation • Response: we are desirous of controllinge.g.- defect %age • Factor levels: – Range of values for doing experiments • Interaction of factors – when one factor has influenceon the effect of the other factorrespectively DOE terminology
  • 70. Taguchimethod (TM) steps involved are as follows 1. Define the problem and objective. 2. Choose response variable(s) for doing optimization. 3. List out all potential factors. 4. Choose important control factors and their range & number of levels. Also, decide factors not to be varied and the ranges in which they should be held constant. 5. Decide on experimental design 6. Conduct trials (called experimental runs) in actual production environment & collect response variable(s) data. 7. Analyze the data, determine the importance & best levels of controlfactors – conclude optimum operating conditions 8. Reconfirm findings by doing production using optimum operating conditions 75
  • 71. 76 content 76 •Case study phase #2 – DOE by Taguchi Method & advanced data analysis •Appendix-1: Primer on Machine learning & Taguchi Method •Case study phase #1 – exploratory analysis •Setting up context- economicreality, challenges, opportunities •Phase #2– confirmation •Findings, Conclusions, summary •Questions,background, Contacts •Overview of new approach, case study background ••Appendix-2: Sand casting some resources
  • 72. Sand casting additional information typical sequence of steps Credit:Prof. Timothy Gutowski
  • 73. Sand casting additional information 1. Sand casting page from Wikipedia gives a good introduction. There are numerous resources on the web 2. Simplified sand casting process video is available here . There are numerous videos on the internet 78