7Epsilon - World Foundry Congress 2014 presentation

Dr. HMd Roshan, Maynard Steel Casting, WI, USACinzia Giannetti, Swansea University, UKDr. Meghana R. Ransing, p-matrix Ltd.Dr. Rajesh S. Ransing, Swansea University, UK

A 7Epsilon Continual Process Improvement Case Study for Defect

Reduction and Quality Control(Official UK Exchange Paper)

71st World Foundry Congress. Bilbao 2014


7Epsilon 7Epsilon term coined by Dr. Patricia Caballero, Tecnalia Spain.

7Steps of 7Epsilon to ERADICATE defects – introduced by

Dr. Rajesh Ransing, Swansea University, UK

7Epsilon Consortium


Dr. Pedro Egizabal, Tecnalia, Spain

Dr. HMd Roshan, Maynard Steel, USA

Dr. Meghana RansingP-matrix Ltd, UK

Dr. Conny Gustavson, Swerea Swecast, Sweden

Dr. Salem SaffeidineSwerea Swecast, Sweden

Prof. Natalia SobczakFoundry Research Institute,

Poland

Mr. Sham ArjunwadkarInstitute of Indian Foundrymen, India


Acknowledgements

p-matrix LtdMaynard Steel Casting, WI, USA


Famous Quotes “If TI only knew what TI knows” – Jerry Junkins, the late

chairman, president and CEO of Texas Instruments

Lew Platt, chairman of Hewlett –Packard echoed with

“I wish we knew what we know at HP”

“If only my foundry knew what it knows …”

Challenges the global foundry industry faces today

Maynard Steel Casting Case Study using 7steps of 7Epsilon

Penalty Matrix Approach for rootcause analysis

Conclusion

Agenda


ISO 9001:2008 or similar quality accreditation

Problem solving & continuous improvement strategies in place

E.g. Physics based simulations, best practice principles, process

stability, in-process data capture

Most precision foundries have …






Assumption – Variability in ALL measurable factors is robust. It does not

influence process variation







Assumption – Variability in ALL measurable factors is robust. It does not

influence process variation

7Epsilon challenges this assumption

discovers ranges of factors within current tolerance limits that can be

associated with process response variations



It is a methodology of simultaneously tweaking single or multiple process

parameter settings to reduce the variation in response values.

Tolerance Limit Optimization


0.01

0.015

0.02

0.025

0.03

0.035

0.04

0.045

0.05

0.055

0 10 20 30 40 50 60 70

Observations No.

% Z

irco

niu

m

Top 50%

Bottom 50%

Mean / Median

7Epsilon Approach


Maynard Steel Casting Foundry, Wisconsin, USA

Low alloy steel foundry

Continual process improvement in melting sub-process

Discover product specific process knowledge

Find new tolerance limits for melting parameters

7Epsilon Case Study


Requirement 0% fractured surface area with conchoidal nature

Fracture tests failing in conchoidal fracture

Rock candy fracture / intergranular fracture

Chemistry within specification but considered to play a significant role in

incidence of conchoidal fracture

GOAL: achieve reduction of conchoidal fracture by optimizing chemistry

parameters in melting sub process

Problem Statement


1. Form a project team2. Acquire team members knowledge about

processes, their factors, responses and causal relationships

3. Gather process knowledge codified using

Process maps, SIPOC diagrams and cause and effect diagrams

Step1 – Establish Process Knowledge


Process knowledge is

The understanding that Y = f(Xs)

How variability in Xs affects variability in Ys

Foundries rely on experts for process knowledge

Generic knowledge comes from experience, published literature

Foundry knowledge needs to be systematically collected, recorded for

reuse

Systematic research on process factors and how they affect response with

written descriptions

Step2 - Refine Process Knowledge


Knowledge discovery

in-process data routinely collected as part of ISO 9001:2008

implementation

Perform rootcause analysis and discover correlations using penalty

matrix approach

Prioritise patterns using p-matrix software

Step3 – Analyse in-process data using penalty matrices


7Epsilon – Reuse in-process data


Response (Y)

fract-Surface

Carbon

Drop

Tap Temperatur

e,F

Pouring

Temperature,

F

Argon stir, mts

%C %Mn %S %P %Si %Ni %Cr %Mo %Cu %Al %TiMn/S Ratio

%Zr %Ca

%Ca/%Al

ratiox1000

0 57 3039 2840 8 0.21 1.00 0.008 0.013 0.42 1.70 1.04 0.40 0.160 0.025 0.0009 125 0.0096 0.0012 48

0 62 2965 2830 3 0.21 1.07 0.008 0.011 0.54 1.61 1.17 0.42 0.170 0.033 0.0018 134 0.0166 0.0027 82

0 73 2971 2850 8 0.21 0.96 0.01 0.012 0.54 1.68 1.11 0.42 0.149 0.04 0.0013 96 0.0224 0.0023 58

3 33 2980 2850 4 0.2 0.94 0.007 0.01 0.51 1.76 1.05 0.45 0.147 0.043 0.0015 134 0.0201 0.0024 56

0 60 2955 2820 2 0.24 1.12 0.01 0.013 0.4 1.72 1.09 0.43 0.153 0.032 0.0010 112 0.0129 0.0029 91

5 84 2905 2836 2 0.19 1 0.01 0.01 0.48 1.71 1.04 0.4 0.135 0.041 0.0153 100 0.0029 0.0030 73

5 35 3007 2846 4 0.2 0.96 0.009 0.01 0.43 1.69 1.07 0.42 0.133 0.013 0.0094 107 0.0021 0.0006 46

0 50 2988 2858 4 0.2 1.06 0.011 0.013 0.5 1.63 1.12 0.4 0.182 0.02 0.0075 96 0.0017 0.0002 10

0 61 2960 2850 4 0.19 0.9 0.009 0.01 0.37 1.64 1.02 0.41 0.146 0.022 0.0102 100 0.0025 0.0007 32

5 64 2950 2852 4 0.18 0.97 0.01 0.009 0.48 1.63 1.06 0.41 0.179 0.026 0.0136 97 0.0037 0.0035 135

15 31 2948 2850 4 0.2 1.07 0.011 0.014 0.4 1.63 1.14 0.4 0.173 0.032 0.0135 97 0.0029 0.0038 119

3 42 2983 2860 8 0.2 0.95 0.013 0.015 0.41 1.67 1.05 0.41 0.150 0.035 0.0011 73 0.0125 0.0125 357

10 51 2915 2840 3 0.23 1.07 0.01 0.013 0.38 1.65 1.11 0.4 0.139 0.028 0.0160 107 0.0027 0.0017 61

10 56 2942 2850 6 0.21 1.1 0.011 0.015 0.39 1.67 1.17 0.41 0.145 0.037 0.0134 100 0.0044 0.0024 65

5 48 2957 2860 2 0.23 0.97 0.01 0.014 0.42 1.63 1.09 0.41 0.117 0.033 0.0114 97 0.0030 0.0018 55

3 58 2990 2860 12 0.19 0.99 0.009 0.014 0.43 1.73 1.09 0.4 0.120 0.046 0.0159 110 0.0042 0.0054 117

0 20 2943 2818 4 0.24 0.95 0.008 0.011 0.6 1.64 1 0.41 0.136 0.03 0.0090 119 0.0029 0.0002 7

3 47 2966 2850 3 0.21 1.02 0.01 0.009 0.5 1.65 1.08 0.41 0.124 0.031 0.0110 102 0.0029 0.0014 45

20 45 2938 2850 4 0.21 1.01 0.013 0.013 0.5 1.63 1.08 0.41 0.119 0.049 0.0150 78 0.0036 0.0027 55

5 68 2994 2850 4 0.18 0.95 0.012 0.014 0.47 1.65 1.06 0.41 0.106 0.046 0.0149 79 0.0041 0.0022 48

0 53 2892 2832 2 0.21 0.95 0.011 0.011 0.52 1.63 0.98 0.42 0.140 0.046 0.0143 86 0.0046 0.0026 57

10 14 2978 2855 4 0.23 0.99 0.012 0.011 0.48 1.62 1.08 0.44 0.153 0.032 0.0121 83 0.0050 0.0015 47

0 72 2959 2832 4 0.19 0.97 0.009 0.009 0.35 1.64 0.97 0.43 0.171 0.041 0.0091 108 0.0022 0.0016 39

0 40 3019 2850 4 0.19 1.05 0.011 0.012 0.4 1.63 1.11 0.41 0.135 0.024 0.0055 95 0.0030 0.0007 29

5 38 2942 2861 4 0.18 1.01 0.011 0.011 0.46 1.68 1.04 0.4 0.144 0.033 0.0109 92 0.0033 0.0019 58

10 26 2925 2875 2 0.21 1 0.01 0.011 0.48 1.6 1.09 0.41 0.140 0.035 0.0134 100 0.0038 0.0036 103

15 47 3028 2865 2 0.18 0.93 0.011 0.015 0.44 1.64 1.06 0.4 0.160 0.027 0.0121 85 0.0029 0.0013 48

30 13 3034 2850 15 0.22 0.93 0.012 0.015 0.41 1.68 1.01 0.43 0.149 0.018 0.0058 78 0.0015 0.0008 44

5 40 2913 2850 2 0.17 0.93 0.009 0.015 0.42 1.66 1.01 0.41 0.145 0.039 0.0125 103 0.0032 0.0037 95

20 58 2940 2855 6 0.2 1.02 0.011 0.014 0.48 1.63 1.12 0.41 0.153 0.034 0.0142 93 0.0030 0.0023 68

0 30 3020 2850 6 0.19 1.04 0.01 0.013 0.46 1.6 1.04 0.41 0.146 0.037 0.0135 104 0.0044 0.0023 62

10 39 2993 2850 5 0.22 0.96 0.01 0.013 0.41 1.67 1.02 0.41 0.117 0.039 0.0153 96 0.0039 0.0045 115

5 49 2944 2850 3 0.18 1.03 0.013 0.013 0.39 1.69 1.01 0.42 0.164 0.026 0.0090 79 0.0029 0.0003 12

3 37 3045 2870 12 0.23 1.18 0.013 0.015 0.52 1.64 1.13 0.42 0.127 0.029 0.0097 91 0.0032 0.0032 110

5 46 2933 2850 6 0.19 0.92 0.012 0.011 0.46 1.68 1.08 0.42 0.110 0.033 0.0109 77 0.0034 0.0024 73

Factors (X)

Penalise variability in one or more process response values

0% given 0 penalty

10% and above given 100 penalty

Linear scaling for intermediate values

How does it work?


100 Penalty Values

0 Penalty Values

Response Scatter Diagram Response Bubble Diagram

How does it work?


Main Effects Bubble DiagramMain Effects Scatter Diagram

Bottom 50%

Bubbles with smaller diameter correspond to 0 penalty (optimal)

Bubbles with bigger diameter correspond to 100 penalty (avoid)

How does it work?


Penalty Matrix

Main Effects Bubble Diagram


Interactions Bubble Diagram

How does it work?

Penalty Matrix

Main Effects Bubble Diagram

Advantages of p-matrix


p-matrix analyses hundreds and thousands of penalty matrices among

factors

Classifies factor settings as Optimal, Avoid or No Effect

Ranks them in order of importance

At end of analysis, a process engineer would get a list of top 15-20

matrices he/she needs to look at

Analyses discrete and continuous parameters together in one analysis

Up to 200 factors and 40 responses can be analysed at once.

Caution


Findings need to be used as pointers for discussion

p-matrix discovers correlations based on data collected

Correlation does not mean Causation

Domain knowledge necessary to interpret results

Hidden causes may be found which require further investigation

Step 4: Develop hypotheses for new product specific ‘process knowledge’


Knowledge Discovery and Reuse

Analyse p-matrix reports

Hypotheses on causation are established using knowledge acquired

in Step 2

Hypotheses are potential solutions

New tolerance limits proposed and

corrective action plan is outlined or collect more in-process data

or conduct one or more design of experiments

Step 5: Innovate using rootcause analysis and conducting confirmation trials


Confirmation trials are carried out to validate the hypotheses and create

new product specific process knowledge

Optimal ranges for all the process variables (X) are determined

New product specific process knowledge is created in the form of

list of values with their new specification ranges

Product specific process knowledge


Sub-ProcessProcess

Variable (CTQ)New tolerance

limitsFrequency of

data collection

Melting and Pouring

Carbon Drop (X1)

47-84 Every Heat

Melting and Pouring

% Sulfur (X7) 0.007-0.009 Every Heat

Melting and Pouring

%Titanium (X15) 0.0009-0.011 Every Heat

Melting and Pouring

Mn/S Ratio (X16) 104-134 Every Heat

Melting and Pouring

%Ca/%Al Ratiox1000 (X19)

6.67-57.5 Every Heat

Step 6: Corrective actions and update process knowledge


New knowledge obtained stored in knowledge repository in tabular

form

This is specific for a given part and process

The new knowledge acquired contributes to devise preventive and

corrective action plans to achieve desired response as required by

ISO 9001:2008 standard

Step 7: Building Aspiring Teams and Environments by monitoring performance


Continually monitor performance so that continual improvement on the

processes can be made to meet the requirement of ISO 9001:2008.

The foundry specific 7Epsilon process knowledge repository can also be

used to train operators and process engineers.

7 Epsilon Knowledge repository

Conclusion


7 Epsilon steps

Tolerance limit optimisation to reduce defects

Penalty matrix visualization to verify evidence in data

Establish causation using domain knowledge

Design corrective actions plan

Validate by implementing on shop floor

Update process knowledge, retain and reuse

7Epsilon Training Courses


Visit the 7Epsilon website at www.7epsilon.org

One day training courses hosted by

Institute of Cast Metals Engineers, ICME, UK on 16th September

2014

On Demand Online course with American Foundrymen Society, AFS - by

Dr. HMd Roshan, course instructor

Join Webinar with Dr. Meghana Ransing, p-matrix Ltd.

Write to [email protected] to book your next course

7Epsilon - World Foundry Congress 2014 presentation

Data & Analytics

world foundry congress

process factors

process stability

process variation

process maps

maynard steel casting

process response variations

melting sub process