Industrial Problem Solving
Presented by
Chris Butterworth Six Sigma Master Black Belt
Chris Butterworth ASQ Certified Six Sigma Master Black Belt 30+ years experience in Manufacturing Inspector - Quality Engineer – Six Sigma Manager – Freelance Consulting – Workshops – Workplace Training
Outline
• Need for problem solving skills
• Cost of Quality
• Six Sigma
• Pareto Distribution
• Chronic vs Sporadic Problems
• Seven Basic Quality Tools
• Human Error
• Mistake Proofing
• Using Data for Process Insight
• Binomial Distribution
• Variation Reduction
• Clue Generating Techniques • Multi Vari Studies
• Tukey End Count
From article in 20 / 20
Cost of Quality
How much does your organization spend on Quality?
3% of revenue?
5% ?
10% ?
20% ?
Cost of Quality
Many companies will find that their Cost of Quality, when properly evaluated, falls close to 15-20% of revenue.
In Paul Borawski’s blog A View from the Q, he writes that this number is 30% in Services, 70% in healthcare!
Cost of Quality
We spend money trying to make a quality product. Prevention – planning, SPC, training etc Appraisal – inspections, tests, measuring equipment, calibrations, audits etc.
We spend more when we fail to do so;
Internal failure – scrap, rework etc External failure – warranty service, complaints, recalls, reputation etc
Cost of Quality
0%
10%
20%
30%
40%
50%
60%
Prevention Appraisal InternalFailure
ExternalFailure
Six Sigma
Six Sigma was developed at Motorola in the 1980s. A structured approach to problem solving and process improvement.
Define Measure Analyze Improve Control
Outputs
High impact project selection Problem Statement Sponsors and Team defined Resources Allocated VOC Process Map
Key metrics identified Baseline of current process performance Defect levels and cycle times established
Evaluation of process stability and capability Sources of variation Identification of failure modes Root causes of defects Performance target for key factors
Significant input factors identified and optimum settings determined Implemented Solution Operating tolerances for new process Error proofing of new process
Control strategy implemented Estimate of new process stability and capability
DMAIC
Six Sigma
Organizations that adopt Six Sigma report incredible cost savings through the elimination of defects.
Pareto Distribution
The Pareto Law is empirical. It shows up often in the distribution of results vs activities.
Also called the 80 / 20 rule.
80% of a company’s revenues come from 20% of its SKUs.
80% of homes are sold by 20% of realtors.
80% of your failure costs come from 20% of your quality problems.
Pareto Distribution
Pareto Distribution
What this means for Manufacturers is that your top few problems really dominate your quality costs. The task is to identify your top few problems. Some are visible. Some require thinking and studying. Your top five problems are going to cost as much as the next 15.
Example 1
A food manufacturer had rejected product due to weight for several years. Products were made along a conveyor with different ingredients added at the various workstations. Products were large and some were diverted to a cutting line to be cut into five equal pieces. Just prior to packaging, all were weighed and many rejected due to over / under weight. What could be causing the rejects?
Example 1
After collecting much data on which pieces were being rejected, it was revealed that only end pieces failed. Center pieces were never rejected. This certainly is a valuable clue.
0.0%
2.0%
4.0%
6.0%
8.0%
10.0%
1st 2nd 3rd 4th 5th
Reject
Example 1
Process data provided insight that wasn’t known beforehand and changed the focus from the manufacturing process to the cutting process. It was suggested that the acceleration / deceleration rate of the conveyor may be the problem. Technician reprogrammed the controller to a lower speed and the reject rate went to zero!
Example 2
Manufacturer of chemical product was dealing with high failure rates at final test. Failures were seemingly random. Much brainstorming and several cause & effect diagrams showed that attempts were made, but unsuccessful. Needed to add structure to existing data. Started by identifying each item with unique ID.
Example 2
A few workstations upstream from final test was a conditioning oven. Oven had 12 shelves. All the failures at final test came from the bottom three shelves of the oven.
Chronic vs Sporadic Problems
Chronic problems • Have been around for along time • Don’t sound alarms • Cost a lot of money Sporadic Problems • A spike in performance • Triggers an alarm • Corrective action often includes adding inspection point
– increase cost
Sporadic Problem
Sporadic problems stand out. They’re new. Sometimes they’ve been resolved before you knew they existed.
Chronic Problem
Chronic problems don’t attract much attention – they are often caused by excessive variation in a process.
Industrial Problem Solving
Industrial Problem Solving
Seven Basic Quality Tools
Process Map
Histogram
Cause and Effect diagram (Ishikawa)
Check Sheet (Tally Sheet)
Scatterplot
Control Chart
Pareto Chart
Process Mapping
Process Mapping
Start with flip chart paper or post-it notes Be flexible Finalize with software such as Visio or Dia or Diagramo
Histogram
23.80 25.54 22.90 28.03
19.89 23.54 27.70 26.87
25.98 23.69 23.47 28.50
30.11 23.52 28.03 27.38
29.79 30.37 19.22 19.51
31.93 24.66 21.61 20.54
16.27 24.26 18.91 27.78
24.06 22.95 23.55 26.29
29.38 32.89 24.87 21.24
20.65 28.46 25.11 24.04
22.24 34.50 23.71 25.53
18.24 22.38 33.78 27.23
17.61 31.65 18.03 25.55
21.09 18.55 22.05 21.36
21.91 27.16 14.69 32.54
16.53 28.61 30.79 26.95
22.73 32.68 19.88 25.29
23.38 24.66 22.39 28.32
Data needs to be graphed so that it can communicate important information to us. Given a set of data, always plot the points.
Histogram
Cause and Effect Diagram
Also called a Fishbone diagram or Ishikawa diagram. The cause-and-effect diagram is a tool used when solving a quality problem. It’s main benefit is that it enables many ideas to be sorted or categorized. The categories are usually the following: Manpower, Methods, Materials, Environment, Machines and Measurements.
Cause and Effect Diagram
Check Sheet
Check sheet is used to record data in real time. Provides quick estimate of the distribution of items.
X Y Plot
A strong correlation between C and I. As C increases, so does I.
Statistical Process Control
10
9
8
7
6
5
4
3
2
1
0
0 5 10 15 20
Unexpected Variation Region
Observation number
Ob
se
rva
tio
n v
alu
e
Expected Variation Region
Upper Control
Limit (UCL) + 3s
Mean
Lower Control
Limit (LCL) - 3s
Observation #10
Seven Quality Tools – Pareto Chart
Human Error
Human error is a dominant source of quality problems in many organizations.
Human Error
The necessity of training farm hands for first class farms in the fatherly handling of farm livestock is foremost in the minds of effective farm owners. Since the forefathers of the farm owners trained the farm hands for first class farms in the fatherly handling of farm livestock, the farm owners feel they should carry on with the former family tradition of training farmhands of first class farms in the effective fatherly handling of farm live stock, however futile, because of their belief that it forms the basis of effective farm management efforts.
Human Error
Mistake Proofing
Mistake Proofing
is a strategy for preventing errors in a process.
This is important because
ERRORS
DEFECTS
PRODUCT
FAILURE
Mistake Proofing
If we take a manufacturing and assembly line which has 15
workstations:
• 15 people
• 360 tasks every hour (do one thing every ten seconds)
• 7 hours per day
15 x 360 x 7 = 37,800 opportunities for error
It is unrealistic to expect error free performance.
Mistake Proofing
Also called: poka-yoke, fail–safing
Mistake proofing is the use of any automatic device or
method that either makes it impossible for an error to occur
or makes the error immediately obvious once it has
occurred.
www.asq.org
Mistake Proofing
Mistake Proofing
Data for Process Insight
A key way to increase process knowledge is to learn empirically - to learn by observation and experimentation.
It’s very important to distinguish between facts, opinions and beliefs.
Binomial Distribution
ABC Mfg Co experienced 14 failures out of 318 items submitted for pilot run. Big problem Many ideas generated and much money spent on design mods. Changes implemented. Sample of 25 tested. No failures. Much rejoicing. Next production run had more failures.
Binomial Distribution
With 14/318 = 4.4% failure rate, what is the probability of getting zero failures in a sample of 25?
Binomial Distribution
It would take a sample size of 67 so that zero failures occur less than 5% of the time.
Variation Reduction
Chronic problems are typically your most expensive and yet they get the least attention.
Chronic problems are often the result of excessive process variation.
The key to reducing variation in process output is to determine the dominant source and work on that.
There are several / many sources of variation in a process that affect the output. But they are not distributed evenly – they also follow the Pareto principle. There is a dominant source.
Variation Reduction
Variation Reduction
Example 3
Manufacturer was rejecting ~3% of final product due to over/under weight. Product was an assembly of 17 components.
σ2Total = σ2
a + σ2b + σ2
c + σ2d + …
Example 3
Example 3
Quotes
“If I had to reduce my message for management to just a few words, I’d say it all has to do with reducing variation.” Dr. W. Edwards Deming “Between 70 and 80 percent of quality issues can be resolved by reducing variation in the process as opposed to redesigning the product.” Don Mitchell, Director of Warranty Improvements, General Motors
Clue Generators
Solving quality problems is like solving a crime. You want to gather clues that steer you in the right direction. Like solving a crime, solving a quality problem is much more accurate when you acquire data and show evidence. Listening and buying into a theory isn’t successful enough to make it a strategic approach.
Clue Generators
• Multi Vari Analysis • Tukey End Count
Multi Vari Analysis
Which factors contribute the most to overall variation? The purpose of the study is to narrow down the large list of potential causes to a small number which includes the dominant cause.
Multi Vari Analysis
Product Quality Measure
Positional Cyclical Temporal Stream to
Stream
Measurement System
Measurement System Study
There are many variables affecting the quality of a finished item. Among those variables, and one of the most neglected, is the measurement system.
Measurement System Study
Measuring system A is not very repeatable when measuring the same item multiple times.
Multi Vari Analysis
Positional - variations within the unit such as several measures of diameter on a ball bearing. Cyclical - one after another. Differences between consecutive units drawn from a process. Could also be batch-to-batch. Temporal - time. Variations from hour-to-hour, day-to-day, etc. Stream-to-Stream - cavity-to-cavity variation. Machine-to-machine, operator-to-operator.
Multi Vari Analysis
Sample 1 Sample 2 Sample 3 Sample 4
9:00 11:00 1:00
Monday
9:00 11:00 1:00
Tuesday Wednesday
Sampling plan for a multi vari study.
Multi Vari Analysis
Example 6
Speakers were failing final test at a high rate ~8%.
Example 6
Example 6
Problem Solving
Solutions are hard to find but clues are plentiful.
Instead of moving from effect to cause, we are far more successful when we look for clues.
Start at the point where the problem is revealed and work upstream one step at a time.
Hypothesis Testing
You may recall this topic from your college Statistics course.
If we want to know whether or not a new process is better than the current process then we need to use data.
Hypothesis Testing
The null hypothesis is the default position. That there is no relationship (or significant difference) between two groups of data.
If our data yields a test statistic that is improbable, under the conditions of the null hypothesis, then we reject the null in favour of the alternate hypothesis.
Hypothesis Testing
Here we will build a depiction of a particular null hypothesis.
We have 16 marbles. Eight are green and eight are red. The marbles are all from the same batch and there is no difference between a red and a green marble other than colour.
Hypothesis Testing
We mix them up and then throw the marbles into a funnel.
The arrangement of red and green is merely random.
Hypothesis Testing
Now we perform an end count.
Counting from the top, how many marbles in a row before we see a colour change?
3
From the bottom, how many?
2
End count = 3 + 2 = 5
Hypothesis Testing
The end count of five was a random occurrence.
Repeat this same test 10,000 times to see the distribution of end counts for this sample.
Hypothesis Testing
End counts of 10 or more happen less than 1% of the time.
Hypothesis Testing
How can we use this information in problem solving?
The eight reds represent eight failed units and the greens are eight units that did not fail.
Hypothesis Testing
With all sixteen items on the lab bench, we need to find other features. These could be measures of the actual units (e.g. length, diameter, weight, etc).
Hypothesis Testing
This table lists each unit along with other variables (e.g. length, height, dia. etc).
Sort the table in increasing order of each variable.
Product Var 1 Var 2 Var 3
green 26.12 5.469 0.55
green 19.53 6.279 0.41
green 29.39 4.111 0.54
green 25.53 4.905 0.37
red 21.4 3.325 0.54
red 26.09 3.957 0.43
red 24.12 3.871 0.36
green 18.82 4.527 0.51
red 24.96 3.919 0.45
green 20.44 4.101 0.35
green 27.63 3.966 0.48
red 23.28 3.152 0.3
red 26.9 3.297 0.34
green 27.68 4.463 0.32
red 22.59 3.546 0.39
red 26.62 4.075 0.35
Hypothesis Testing
The end count when sorted by Var 2 is 14.
This is insightful.
All the failures were low in Var 2 and all the good units were high in Var 2.
Product Var 1 Var 2 Var 3
red 23.28 3.152 0.3
red 26.9 3.297 0.34
red 21.4 3.325 0.54
red 22.59 3.546 0.39
red 24.12 3.871 0.36
red 24.96 3.919 0.45
red 26.09 3.957 0.43
green 27.63 3.966 0.48
red 26.62 4.075 0.35
green 20.44 4.101 0.35
green 29.39 4.111 0.54
green 27.68 4.463 0.32
green 18.82 4.527 0.51
green 25.53 4.905 0.37
green 26.12 5.469 0.55
green 19.53 6.279 0.41
Industrial Problem Solving
Skills you need on your team.
linear regression statistical process control
multiple linear regression industrial problem solving
binary logistic regression data visualization
design of experiments hypothesis testing
gauge r & r process capability analysis
Industrial Problem Solving
Industrial Problem Solving Workshop
Nov 13 to 14
Register at www.belfield.ca