1
Quality Control
◦ Six Sigma, Capability Score, Control Chart
◦ Impact of Defect on Flow
2
3
M&M’s is the flagship product of Mars, Incorporated.
Take a specific type of processes of this company for
example, where an output flow unit is a bag of M&M’s
chocolate that weighs 50g.
Due to the ever-changing real world operating environment, the exact weights of flow units may differ from
each other; i.e., there is variation in weights.
A flow unit that weighs either too little or too much is considered of insufficient quality and will lead to
customer dissatisfaction if sold.
This motivates the importance of quality control for Mars, Incorporated.
A prevailing tool of quality control is the so-called “Six Sigma” method.
Note:
There are two aspects of quality:
Quality of design = customer requirements vs. product specification of the organization;
Quality of conformance = product specification vs. the actual product delivered.
We focus on quality of conformance. 4
The following is the key measure for conformance quality of a process:
𝑈𝑆𝐿−𝐿𝑆𝐿
Capability Score 𝐶𝑝 = , where
6𝜎
◦ 𝑈𝑆𝐿 (resp., 𝐿𝑆𝐿) denotes the upper (resp., lower) specification limit for quality.
◦ 𝜎 is the standard deviation of the weights of all the bags output by the process.
In the example of M&M’s,
◦ suppose the company sets 𝑈𝑆𝐿 = 52𝑔, 𝐿𝑆𝐿 = 48𝑔.
◦ 𝜎 needs to be estimated by drawing a sample, e.g., random sampling 50 bags whose weights are as
follows:
50.00 50.20 48.20 51.10 50.7
49.30 50.30 47.50 49.60 49.6
50.60 50.00 49.70 53.00 48.9
50.80 50.20 49.10 49.9 49
48.60 49.60 50.50 52.2 50.6
50.30 49.40 51.40 50.1 48.8
50.50 49.20 48.40 49.5 49.8
49.40 49.70 52.70 50.8 49.9
49.10 51.30 52.00 50.8 49.4
49.30 50.30 49.80 48.8 50.6
Note:
1. A flow unit that goes beyond [LSL, USL] is considered a defect (non-tolerable by customers).
2. In quality control, we always use be a Normal Distribution to describe the distribution of data.
5
A Normal distribution, Normal(𝜇, 𝜎 2 ), has two parameters,
◦ 𝜇 indicates the center;
◦ 𝜎 2 indicates the shape, the larger 𝜎 2 , the more spread/uncertainty.
The following is a normal distribution with 𝜇 = 100.
6
A Normal distribution, Normal(𝜇, 𝜎 2 ), has two parameters,
◦ 𝜇 indicates the center;
◦ 𝜎 2 indicates the shape, the larger 𝜎 2 , the more spread/uncertainty.
The following is a series of normal distributions with 𝜇 = 0 and increasing values of 𝜎.
7
A Normal distribution, Normal(𝜇, 𝜎 2 ), has two parameters,
◦ 𝜇 indicates the center;
◦ 𝜎 2 indicates the shape/uncertainty, the larger 𝜎 2 , the more spread/uncertainty.
Excel function: 𝒏𝒐𝒓𝒎. 𝒅𝒊𝒔𝒕(𝒃, 𝝁, 𝝈, 𝟏).
◦ Meaning: For a sample, the estimated percentage of numbers that are ≤ 𝑏 is roughly
𝑛𝑜𝑟𝑚. 𝑑𝑖𝑠𝑡(𝑏, 𝜇, 𝜎, 1).
Note: To compute the percentage of numbers that are in between 𝑎 and 𝑏 (where a ≤ 𝑏), the formula
is 𝑛𝑜𝑟𝑚. 𝑑𝑖𝑠𝑡 𝑏, 𝜇, 𝜎, 1 − 𝑛𝑜𝑟𝑚. 𝑑𝑖𝑠𝑡(𝑎, 𝜇, 𝜎, 1), not 𝑛𝑜𝑟𝑚. 𝑑𝑖𝑠𝑡(𝑏 − 𝑎, 𝜇, 𝜎, 1).
How to estimate 𝜇 and 𝜎? (see Excel)
◦ For 𝜇, let 𝜇 = sample mean;
◦ For 𝜎 2 , let 𝜎 2 = sample variance.
8
The estimated percentage of flow units that meet the quality requirement is 𝑛𝑜𝑟𝑚. 𝑑𝑖𝑠𝑡 𝑈𝑆𝐿, 𝜇, 𝜎, 1 −
𝑛𝑜𝑟𝑚. 𝑑𝑖𝑠𝑡(𝐿𝑆𝐿, 𝜇, 𝜎, 1).
Those that fail the requirement will either be reprocessed or be thrown away.
𝑈𝑆𝐿−𝐿𝑆𝐿
Recall Capability Score 𝐶𝑝 = .
6𝜎
◦ For fixed 𝑈𝑆𝐿 and 𝐿𝑆𝐿, smaller 𝜎 ⟺ larger 𝐶𝑝 .
◦ For fixed 𝑈𝑆𝐿 and 𝐿𝑆𝐿, smaller 𝜎 ⟺ larger 𝑛𝑜𝑟𝑚. 𝑑𝑖𝑠𝑡 𝑈𝑆𝐿, 𝜇, 𝜎, 1 − 𝑛𝑜𝑟𝑚. 𝑑𝑖𝑠𝑡(𝐿𝑆𝐿, 𝜇, 𝜎, 1).
◦ Thus, larger 𝑪𝒑 ⟺ larger 𝒏𝒐𝒓𝒎. 𝒅𝒊𝒔𝒕 𝑼𝑺𝑳, 𝝁, 𝝈, 𝟏 − 𝒏𝒐𝒓𝒎. 𝒅𝒊𝒔𝒕(𝑳𝑺𝑳, 𝝁, 𝝈, 𝟏).
◦ For fixed 𝜎, larger 𝑈𝑆𝐿 and smaller 𝐿𝑆𝐿 ⟺ larger 𝐶𝑝 .
◦ For fixed 𝜎, larger 𝑈𝑆𝐿 and smaller 𝐿𝑆𝐿 ⟺ larger 𝑛𝑜𝑟𝑚. 𝑑𝑖𝑠𝑡 𝑈𝑆𝐿, 𝜇, 𝜎, 1 −
𝑛𝑜𝑟𝑚. 𝑑𝑖𝑠𝑡(𝐿𝑆𝐿, 𝜇, 𝜎, 1).
◦ Thus, larger 𝑪𝒑 ⟺ larger 𝒏𝒐𝒓𝒎. 𝒅𝒊𝒔𝒕 𝑼𝑺𝑳, 𝝁, 𝝈, 𝟏 − 𝒏𝒐𝒓𝒎. 𝒅𝒊𝒔𝒕(𝑳𝑺𝑳, 𝝁, 𝝈, 𝟏).
Note:
1. 𝐶𝑝 captures the quality of a process, the larger 𝐶𝑝 , the better the quality.
2. There are two levers to improve 𝐶𝑝 , either (i) making 𝑈𝑆𝐿 larger and 𝐿𝑆𝐿 smaller or (ii) reducing
𝜎, the variability.
9
A 6𝜎 process requires 𝜎 satisfying that 𝑈𝑆𝐿 is roughly 6𝜎 above 𝜇 and 𝐿𝑆𝐿 roughly 6𝜎 below 𝜇, i.e.,
𝑈𝑆𝐿 = 𝜇 + 6𝜎 and 𝐿𝑆𝐿 = 𝜇 − 6𝜎. In this way, the percentage of flow units that meet quality requirement
is
𝑛𝑜𝑟𝑚. 𝑑𝑖𝑠𝑡 𝜇 + 6𝜎, 𝜇, 𝜎, 1 − 𝑛𝑜𝑟𝑚. 𝑑𝑖𝑠𝑡 𝜇 − 6𝜎, 𝜇, 𝜎, 1 =
Why 6𝜎?
Defect per 10,000 3,000 1,000 100 10 1 0.002
million
𝐶𝑝 0.86 1 1.1 1.3 1.47 1.63 2
Howmany
How to estimate
𝜎 𝜇2.58
and 𝜎? (see 3Excel) 3.3 3.9 4.41 4.89 6
Note:
1. As 𝐶𝑝 increases, the percentage of defective flow units goes down.
2. 6𝜎 corresponds to almost no defect, and therefore, is extremely hard to achieve.
3. Higher multiples of 𝜎 corresponds to higher cost. How many 𝜎 to employ depends on the
industry. E.g.,
(a) Hospitals strive for 6𝜎 as a defect, which corresponds to wrong prescriptions or other life-
threatening act, is extremely expensive.
(b) For M&M’s, probably 4𝜎 is good enough as the consequence of a defect is not heavy.
10
Control chart is a graphical way of monitoring the quality of flow units.
The procedure is as follows: (refer to Excel for an example)
◦ Set an upper control limit (UCL) and a lower control limit (LCL).
◦ Each of the following is considered abnormal:
(i) If the mean of a sample goes beyond the range [𝐿𝐶𝐿, 𝑈𝐶𝐿].
(ii) Several (e.g., 7) consecutive sample means, though within [𝐿𝐶𝐿, 𝑈𝐶𝐿], are all near 𝐿𝐶𝐿 or near
𝑈𝐶𝐿.
◦ If one or both of the above abnormality happen, then the process is immediately paused and we will do
investigation to find the cause and resolve the issue.
Note:
1. Typically, 𝑈𝐶𝐿 < 𝑈𝑆𝐿 and 𝐿𝐶𝐿 > 𝐿𝑆𝐿, i.e., the control limits are tighter than product
specification limits. (to act in time before too late)
2. There is no formal rule to decide the exact values of 𝑈𝐶𝐿 and 𝐿𝐶𝐿. Yet, one should take the
following trade-off into consideration: The tighter the range [𝐿𝐶𝐿, 𝑈𝐶𝐿], the better the quality
control, whereas more often the wrong alarm, i.e., the process will pause when there is in fact no
abnormality.
3. There is no formal rule to decide how often to draw a sample. There is a trade-off: The more
often a sample is drawn, the more in-time the quality control, whereas the higher cost.
11
Capability Analysis Conformance Analysis
(sampling, calculate 𝐶𝑝 ) (UCL, LCL, Control Chart)
Investigate on abnormalities,
Eliminate Causes
brainstorming
Note: The above procedure is never ending.
12
An automotive company is evaluating the capability of one of its key suppliers. The supplier delivers a part of
the dashboard, and it is critical for the quality of the vehicle that the part measures between 110.45cm and
110.55cm. Every part that is beyond this range is labeled defective.
A random sample of 100 parts that were inspected over the course of the last year revealed that the average
measurement was 110.5cm. The standard deviation was 0.05cm.
Q1: What is the capability score of the supplier?
13
Q2: What percentage of the deliveries were likely to be outside the specification limits [110.45, 110.55]?
Q3: To what level would the supplier have to reduce the standard deviation of the operation if his goal were to
obtain a capability score of 1.67 (i.e., get 0.6 defects per million)?
14
15
Example 1:
Consider the following process with 3 activities, A1, A2, and A3, and 3 resources, R1, R2, and R3.
There is 1 unit of each resource.
For simplicity, assume
◦ A1 and A3 do not produce defect.
◦ 40% of flow units after A2 are defective and need to go through A1 and A2 again.
◦ Those reworked flow units are not defective, i.e., there is no need to go through A1 and A2 a 3rd time.
40% defect after A2 and need rework
A1 A2 A3
R1; 5 mins R2; 4 mins R3; 6 mins
Q: What is the capacity of this process?
16
17
Example 2:
Consider the following process with 3 activities, A1, A2, and A3, and 3 resources, R1, R2, and R3.
There is 1 unit of each resource.
For simplicity, assume
◦ A1 and A3 do not produce defect.
◦ 30% of flow units after A2 are defective and need to go through A2 again.
◦ Those reworked flow units are not defective, i.e., there is no need to go through A2 a 3rd time.
30% defect after A2 and need rework
A1 A2 A3
R1; 5 mins R2; 4 mins R3; 2 mins
Q: What is the capacity of this process?
18
19
Example 3:
Consider the following process with 10 activities, A1, A2, …, A10.
Each activity has 1% chance of producing defect.
The defective flow units have to be thrown away and are wasted.
A1 A2 A10
Q: What is the defective rate of this process?
Note: For processes with multiple serial activities, defect may blow up. This shows the importance of
maintaining extremely high quality at each activity for such processes.
20
Wharton Business School, “Introduction to Operations Management”, Coursera, 2018.
21
