Anything We Want To Improve On Must Be Easily Measurable Parameters To Measure
Anything We Want To Improve On Must Be Easily Measurable Parameters To Measure
Anything We Want To Improve On Must Be Easily Measurable Parameters To Measure
Parameters to measure:
1) Make Sure you create a product which Creates Expectation/ Excitement in the prospective customers
2) Performance: product must do what it is expected to do. And do it without a fail. This was customer
develops a Perception
Therefore, Quality is equal to P/E i.e (Perception/ Expectation)…………….what I got / what I thought
Six sigma is a thing which has brought Quality to be linked with business. Earlier it was an operational issue. Now,
Management revers quality
Quality should translate into money.
People did make products but they were more of artist (quality was implicit/embedded in whatever they did/
thought. They always did quality work. People were specialized in whatever they did. They were perfectionist. High
degree of ownership and inheritance of vocations.)
With industrial revolution, volumes became high. Machines replaces humans. Therefore, that link of ownership was
lost. Reputation issue got lost. Numbers became more important.
World learnt this during second World War. They introspected. 30% of soldiers who died were due to malfunctioning
equipment. Not due to enemies.
1947- IOS (internation organization of standards was formed) , now known as ISO.
First time, standards were written for manufacturing. Now we have an industry standard to be complied with.
But all those standards we for END FINISHED PRODUCTS. No standards for PROCESSES.
All focus was on perfection in end product. This was done through comparison-----> INSPECTION. The role of
Inspector became most important in the industry. People made products, other group of people inspected.
In formal terms, his job was to find mistakes-----That was the only parameter to judge HIS performance. So, He was
responsible for maintaining the quality on the long run. Ideally, it should be the responsibility of the one who is
making the product.
Moreover, Inspection also costs money------on persons, process of inspection, time taken during inspection etc.
Y= X*5
Therefore, one round of correction costs 5 times the cost of first time production.
Eg. Assembly line of a TV-------------say 100 TVs throughput through assembly line in one hour
Repair of TVs possible, only 20 per hour.
Timeline:
1975- Moto was no. 1 in communication devices.
1980 – Moto faced survival probs with fierce Japanese competition.
Disgression: Any arising defect in a product during use gives a perception that the product with never work the
same again.
Imagine- jisko bannaate waqt hi sudharna padey vo kya kaam karega?
Americans focused on what, Japanese focused on how. If you want a product to performs without defect, make it
without defect. Manufacture it defect free, it will work defect free..
A recent eg:
Indian chairman said….we are a customer focused company……Japanese picked the phrase…..they were impressed
and asked how Indians managed it…(Indian boasted about the penetration of their service facilities)….Japanese
whispered…is your quality that bad?
Paradigm shift-
Dr.J.M.Juran-------Father of total quality management-------during 1950s tried to advise US industry that inspection
doesn’t mean quality. US industry ignored him, blinded by their own prevalent economic glory. He wrote a book
---handbook of quality---bible for quality professionals. Published in 1954 for first time.
Disgression: Japanes are does, they respect Deming more than Juran, because Deming was more of a shopfloor
person.
Done in 7 steps:
Eg:
Nike does only market and product design.
Everything else is outsourced------procurement to India, production to China, distribution to Dubai
2) PROCESS
Every process has Key Process Parameters (Input Variables of the Process)
KPIV- Key Process Input Variables
That input variable of the process that has a direct impact on the output that is delivered by the process
i.e what should be ensured so that process happens correctly------> output is good-----customer is satisfied.
Whole concept of six sigma is to control KPIVs, so that if they stay within operating standards, process can’t
go wrong, and thereby, outpur is always in desired form.
A disciplined, data -driven approach and methodology to help eliminate defects in a process…from manufacturing
to transactional and from products to service
Challenge:
Effort become higher as you go up the sigma. Cost increases disproportionately. That’s why most companies don’t go
for more that 4 to 5sigma.
But this is true for 80% industry only.
Medical, Aerospace, aviation etc are exceptions. There you need six sigma.
At whatever point customer is happy…that you neeed to achieve. That varies from industry to industry.
GE is most recognized and respected for Six Sigma. That too, only for 2 processes
The more who focus on internal customers, externa customer satisfaction just becomes the by-product.
Means…design KPIVs such that you satisfy the requirements of persons involved in manufacturing (survey satisfies
requirements of planning person efficiently, planning satisfied requirement of design personel and so on)
2. Boundaryless collaboration- All departments working together towards the common goal of customer
satisfaction.
Support each other
3. Proactive Management – anticipate and act rather than react before problem becomes a problem
Eg:
qty of plankton doubles everyday
Pond with fill in 50 days.
Frog needs 10 days to come up with action plan
4. Data Driven Decision - always base decisions on data. Not on qualitative experience.
Dr. Demings process.
“In gods I trust, everybody else get data”
“Change is not necessary, because survival is not mandatory”
6. Inspire perfection, tolerate failure- improvement always happens only through risks. Risks are taken when
there’s insurance against failures.
CHAMPION
DEMING on processes:
Don’t change employees. Change systems. Systems are cause of 85% defects.
DMAIC vs DMADV
Import from study material (Smartart)
DMAIC
Define>Measure>Analyse>Improve>Control
Details in material
DEFINE: Identification of CTQ (Critical to quality)- That specific measurable attribute of the output
which is of maximum concern to the customer where we do not meet requirements.
Scope of project in order to improve that CTQ
Measure: Establish the gap in what customer wants and where we are.
Expected capability – current capability.
This gap is expressed in sigma level.
Eg. Our process is 1 sigma level. Customer expects 2.5 sigma.
Analyse: Identifying root causes due to which performance is impaired. Why we are not able to satisfy
customer expectations.
Eg: through Fish Bone, Pareto, YYY, Regression etc., correlation, scatter diagrams.
It’s the basic primary thing through which we draw any inference about a process.
i) a method statement:
5W 1H
Why How
What
When
Where
Who
Process Performance:
Mean is not a good parameter to judge performance. It averages out the variance.
Not that if you have achieved 3.4 defects per million, you have achieved 6 sigma
Eg. Hypothesis
Customer req. = 60hr + 6hours
LSL = 54 hours
USL = 66 hours
T = 60 hours
So, between the service window desired by customers, only 68% (corress to + 1sigma) values lie.
To fit more values, try to reduce sigma (std. deviation) somehow. This way you will be able to fit more values
between upper and lower specification
How many sigma you can fit within the customer desired window.
Eg. If you’re able to fit 1 std. deviation within customer , it will be 1sigma process.
DEFINE
Step 1 Determine the CTQs
Identify customers- list customers, define customer segments, narrow the list to most
relevant customers
Gather VOC- gather verbatim VOC and determine service quality issue
Surveys
Interviews
Be a customer
Focus group
Cust observation
Listening Posts
Competitive comparison
One to one meeting (most reliable since the data will be most reliable)
*this is not a customer satisfaction survey, it’s a customer dissatisfaction survey, which a customer best
articulates in a one to one meeting
*Ask standard questions- unanimously decided, common set of questions.
*Open Ended Questions, no Yes/No questions. Let the customer describe his problems. But problem
should be specific.
*Ask at a time where you are most likely to get a neutral response.
Organise VOC
Suppose,
You have got 50 negative statements (anyway only negative reviews are to be considered)
Prioritize the problems.
Debate on Unique meaning statements, eliminate similar meaning statements –
This is achieved through affinity diagrams. (Affinity diagram belongs to family of tree diagram. It is
not a statistical tool. It’s a management tool)
In affinity diagrams-All similar meaning statements are arranged vertically in a column. Different
meaning statement arranged in different columns.
Prioritise VOC
Generally, any DMAIC project will take only 16 weeks. In such a short span of time, you can address
hardly 2-3 CTQs at max. For each CTQs, you’ll have to make about 10-15 changes in the process.
So, draw an affinity diagram for one VOC, derive multiple CTQs, prioritize top CTQs to be addressed,
and then make changes to address those selected CTQs
Reduce the scope of the changes, but don’t increase the duration of DMAIC project. Because, if you
increase the time, the sustainability of interest goes away. Efforts become futile.
Translates VOC
KANO MODEL
While defining the project (i.e the addressing of CTQs), 5 questions need to be answered
Scope
Purpose- in terms of process improvement and financial gains
Roles and responsibilities
Timelines
Tool which we use for mapping the process is known as SIPOC- refer to page 57 of study material.
SIPOC- Supplier- Inputs-Process-Output-Customers
Read Page 57 to 67
DMAIC addresses the biggest problem- the resistance to change by involving people who are part of change making
process, to implement the change.
In the meeting, 5Ws and 1H must be decided for each of the five points under Charter.
Every decision must be unanimous. Each ‘NO’ from a panelist of the team means exclusion of one whole department
Separate Topic:
Organisation
Working for vertical growth
Eg: Revenue, Market Share etc.
Sales Planning Purchase Ops Logistics Customer
Goals: Resources cost down Optimize Cycle time
revenue up down resources reduction
Work across departments to satisfy the end customer
Problems arise at boundaries of two departments (blame game). Cross functional people are required to
douse that fire.
All stakeholders must be involved in problem solving. End to End representation is very important. Otherwise
solution for one dept. will become problem for another dept.
Ideal Six Sigma team should have a few stars (40%) and a lot of rats.(60%)
Rats are future stars!
Eg. Of Hospital (nurse, ward boy, pharmacy)
Value added activity - any activity in the process for which the customer is willing to pay for.
Eg. Procurement of raw material - value added
Storage of Raw material -non-value added
Logistics, warehousing -non value added
Maintenance downtime -non value added
Non value added activity-
Value Enabling Activities
Digression
Customer
Sales
Planning
Matrix flowchart is much more useful in visualization
Which activity is going on in which department. Where the work is getting transferred from one department to
other.
Generally misunderstanding/ bottlenecks happen at the interface of two departments (Internal customer
dissatisfaction)
In value stream mapping, we identify from the flowchart which activity is value added and which is non-value added.
Every decision activity (depicted by diamond in a flowchart) Is a non-value added. (They add delay to the process)
Start with an assumption that every activity is non-value added. Then try to disprove the assumption. If not able to
disprove, try to eliminate as much as possible.
Value Enabling activity: Which are non-value added, but they are essential. Eg.HR, Finance.
Any multiple reviews of docs, rework, holding, delay etc. are non-value adding.
Eg of improvements in industry: Mother godown models have been replaced by Hub and Spoke models.
Value added: make them more effective (do them right) Do not take any ad-hoc action
Value enabling: minimize their cost right now. Do in the Improve stage
No value added: minimize, if not eliminate of DMAIC
Efficiency = PT/LT
Ideal Value = 1
Generally, attribute data is weak for statistical analysis. It has very less predictive capability.
Exercise:
Identify the type of data
Amount of time taken to respond to call: continuous
No. of blemishes per sq yard attribute
Daily test of water acidity Continuous
Length of screws in a sample continuous
No. of employees who had accident attribute
Whenever collecting data, make a good data collection plan. Define meaning of each term eg…duration of a working
day, meaning of holiday etc.
Random (generally used when the population is homogenous, when there is no suspicion of any kind of bias-
location bias, time bias, operator bias)
Stratified (when there is heterogenous population. We convert heterogenous population into homogenous
strata and then do random sampling on each stratum)
Sequential (ideally, it is just a part of stratified sampling)- involves set of continuous pieces.
Eg, whenever there is a changeover in the assembly line, we do sequential sampling to check whether the
product has stabilized or not)
Sample size = n
(Thumb rule: every 10 out of thousand,
Also, if Zalpha/2 = 1.96, which is the critical value for 95% confidence, we calculate sample size by
n = [(Zα/2.sigma)/ delta]2
Mean = x
n = [zalpha/2/delta]2 x p(1-p)
but generally, these equations are seldom used. Mostly, we’ll decide sample size based on experience. As six
sigma professionals we do not challenge experience
(Eg. When a set of people rate a service, we use mode to guage the general perception, not the mean or media.
Mode gives the most popular rating)
Whenever using any survey…whenever using a rating….use less no. of nodes. Eg…while desiring results from 1 to
10….do not use 1, 2,3 4….10 as options…..Use only 1, 5, 9. Then the results will be more useful to analyse. It’s kind of
decisive)
How to know whether our data is normal or skewed, here minitab comes handy
If you don’t have minitab, calculate mean and median both. If median is withing 20% deviation from mean value (it is
close to a normal distribution), use mean, otherwise use median)
Test 1: Test for normality
Whether my data is normal
Eg.
Suppose for a process
Mean = 65
CI = 65+ - 0.59
If we were to actually improve the data, we’ll have to exceed this CI.
Digression
I will set the conveyor speed of my assembly line based on speed at which glass bottles are being made.
Say 65 pieces per hour
But 5% time, this will not be the case…may be outliers.
If the outliers are greater than 5%, then there is a problem.
Risk = 0.05
This helps calculating ‘z’…. which has a fixed value for corresponding value for confidence level.
Testing of assumptions
Hypothesis testing
Null hypothesis (Probability of achieving null hypothesis is 0.05)
What I am ‘claiming’.
Alternate hypothesis
That
We either reject the null or fail to reject the null (not ‘accept’ the null).
To test spread, we use histogram. Taller the histogram, lesser the spread. Wider the histogram, higher the
spread.
CASE:
Customer Specification
USL = 90
LSL = 60 75 +-15
Target = 75
Available data
X bar = 72
Sigma = 6
Therefore, on lower side, our process is 2 (4-5) sigma whereas on higher side, it is just 3 (6) sigma.
By observing the output of our process and the desired customer specification, we decide whether we need
to treat the mean or the standard deviation or both.
Test 3: Stability (whether all individual values are within the control limits i.e + - 3sigma limits)
BOX PLOT:
So formula evolved to
Means how many std. deviations we have between mean and upper service level. (that is nothing but the
sigma level of the process)
Defects (total no. of defects in a sample, where each object can have multiple defects also) vs Defective (how
many defective pieces are there in a sample, with an object with multiple defect counted only once)
Poisson (used for defects) vs Binomial (used for defective)
Repeatability - when a single operator is able to repeat his reading for the same part measured more than
once, he is repeatable. Otherwise, there is error of repeatability
Guage reproducibility: Same measuring instrument reporting different values for different observations.
When data type is defective, we take AAA (Attribute Agreement Analysis) (stats>quality>attribute agreement
analysis)
Understand the interpretation of kappa
Whenever there is low kappa score for an operator (lower than 0.9)
Find if his repeatability is low
If yes, train the operator
Otherwise
Find if his reproducibility is low,
If yes, make changes in the measurement system
Whenever the Overall kappa score is low (lower than 0.9), entire measurement system is to be scrapped)
Wherever human intervention is there in recording the measurement and humans are reporting a value, Gage R&R is
required.
Defectives:
P= (No. of Defectives)/No. of Samples Inspected
DPU = 20/200 = 0.1
DPMO (defects per million opportunities) = DPO x 10^6 = 0.02 x 10^6 = 20000
Sigma level = 3.5