Tag Archives: manufacturing

The State of Chaos

When a process is out of control and it is producing nonconforming product it is in a state of chaos. The State of Chaos is one of the four states a process can be in as shown in “What State Is Your Process In?“. The manufacturer cannot predict how much nonconforming product his process will produce in any given hour or day. At times the process will produce nothing but conforming product. Then without warning it will produce nothing but nonconforming product. It might seem as if there were ghosts in the machine.

A process in such a state is affected by assignable causes that are easily identified through the use of process control charts. The effects of these assignable causes have to be eliminated one at a time. Patience and perseverance are necessary. It is essential that the process be brought under statistical control and made predictable. Once the process has achieved stability further improvement efforts can be made to reach the ideal state.

ChaosProcess

Note: I learned this material from reading Dr. Wheeler’s writings. My post is intended to reflect my current understanding. None of the ideas herein are original to me. Any errors are my failures alone.

References

Advertisements

The Brink of Chaos

Of the four states a process can be in (see “What State Is Your Process In?“) the most sinister state is the one where it is producing 100 percent conforming product but is operating in an unpredictable way. That is, the process is not under statistical control. Such a process is, in fact, on the brink of chaos. But, hold on. There is no nonconforming product, therefore there is no problem, right? It is easy to get lulled into complacency by this happy circumstance.

BrinkProcess

But because the process is not under statistical control it is impossible to predict what it will do in the next instance. Various assignable causes are affecting the process in an unpredictable fashion. The effect of these causes could very well be the production of nonconforming product without any warning. When that happens the process has moved into a state of chaos.

The only way to address a process on the brink of chaos is to use process control charts to identify assignable causes and eliminate their effects one-by-one and bring the process under statistical control. You can then start other improvement efforts like moving the process mean to the process aim and reducing the process variation by minimizing the influence of common causes affecting the process.

Note: I learned this material from reading Dr. Wheeler’s writings. My post is intended to reflect my current understanding. None of the ideas herein are original to me. Any errors are my failures alone.

References

The Threshold State

A process that is predictable or in a state of statistical control, but producing nonconforming product can be described as being in the Threshold State. This is one of the four possible states that a process can be in as noted in “What State Is Your Process In?“. But, what might such a process look like?

A process in the threshold state might be operating with its mean higher than the process aim,

ProcessMeanHigh

or it might be operating with its mean lower than the process aim,

ProcessMeanLow

or it might be operating with a process dispersion greater than the product specification window,

ProcessDispHigh

or it may be operating with some combination of a shift in its mean and breadth of its dispersion.

Nevertheless, the fact that such a process is in statistical control means that it will continue to produce consistent product so long as it stays in control. This in turn means that the producer can expect to produce a consistent amount of nonconforming product hour after hour day after day until a change is made in the process or a change is made to the specifications.

It is important to say here that exhorting your workers to work harder or to “Do it right the first time” or to show them the examples of nonconforming product from a process in the threshold state will not lead to improvements. They are not the cause for the failures. The causes for the nonconforming product are systemic and must be dealt with at the system level. Focusing on the worker will only serve to demoralize and frustrate them. It may lead to tampering with the process turning a bad situation worse.

You can always share your process data with your customer to demonstrate its stability and ask for a change in the product specifications. However, if specifications cannot be changed your only recourse is to modify your process to shift it from the threshold state into the ideal state. Adjusting the process mean to match the aim is usually relatively simple. In comparison, reducing the process variation requires an understanding of the common causes affecting the process and their respective effects – a much more involved activity.

While you are working on improving your process you are still producing nonconforming product. Until such time as you achieve the ideal state for your process, you must screen every unit or lot before shipping product to your customer – a 100 percent inspection and sort. This should be treated as a temporary stop-gap measure. You must recognize it as an imperfect quality control method and be mindful that defectives will escape.

Note: I learned this material from reading Dr. Wheeler’s writings. My post is intended to reflect my current understanding. None of the ideas herein are original to me. Any errors are my failures alone.

References

The Ideal State

In “What State Is Your Process In?” I noted that a process can be in one of four possible states. Here I write about the Ideal State wherein a process is predictable and is producing 100 percent conforming product.

A process that is predictable is one that is in a state of statistical control. The variability of the product from one unit to the next is randomly distributed about the average and bounded within statistically established limits – its natural limits (red solid lines in the figure below). So long as the process remains “in control”, it will continue to produce units within these limits.

Image

Complete product conformity comes about when the process’s natural limits fall within the product’s specification limits (blue solid lines in the figure below). This depicts the ideal state for a process.

Image

In order for a process to achieve this ideal state

  • The process must be inherently stable over time. This means that in the absence of external disturbances – what Dr. Shewhart referred to as assignable causes – the process’s natural variability does not change over time. (Note: There are processes that are inherently chaotic. An excellent reference to such processes is “Nonlinear Dynamics And Chaos” by Professor Steven Strogatz)
  • The process must be operated in a stable and consistent manner. The operating conditions cannot be selected or changed arbitrarily. Often machine parameters are tweaked in response to natural fluctuations in the process’s output. These actions add to the process’s natural variation disrupting its stability. Dr. Deming demonstrated the effects of such tampering through the “Nelson Funnel Experiment”. (Bill Scherkenbach has an excellent discussion of it in “Deming’s Road to Continual Improvement”.)
  • The process average must be set and maintained at an appropriate level. If you refer to the charts above, you can imagine the consequence of moving the process average up or down from its aim. The result is the production of nonconforming product either on the high or low side.
  • The natural tolerance of the process must be less than the specified tolerance for the product. This is obvious upon a first glance at the second chart above.

A process that satisfies these four conditions will be in the ideal state and the manufacturer can be confident that he is shipping only conforming product. In order to maintain the process in the ideal state he must use process control charts. He must act on the signals they provide to promptly identify assignable causes and eliminate their effects.

Note: I learned this material from reading Dr. Wheeler’s writings. My post is intended to reflect my current understanding. None of the ideas herein are original to me. Any errors are my failures alone.

References

What State Is Your Process In?

You can take one of two approaches to controlling the quality of your product. Once manufactured, you can compare it against its specifications and sort it as either conforming or nonconforming. This, however, will guarantee the greatest degree of variability between units as anything within specifications is considered acceptable. Alternatively, you can work to run the manufacturing process as consistently as possible to produce units that vary as little as possible from one another.

There is no path to improving performance with the first approach where you sort manufactured product as conforming or nonconforming. However, there is a clear path to improving performance through the use of process control techniques introduced by Dr. Walter Shewhart. But, how do you gauge improvement? One measure is achieving a state of statistical control for the process so that its behavior is predictable. Another measure is the manufacture of 100 percent conforming product by the process.

When these two measures are taken together, there are four possible states a process can be in (see figure below): The process is

ProcessState

Note: I learned this material from reading Dr. Wheeler’s writings. My post is intended to reflect my current understanding. None of the ideas herein are original to me. Any errors are my failures alone.

References

A Very Cool History of the Toyota Production System

The Toyota Production System was born out of necessity. These two videos show its evolution.

A Simple Process Validation Example

Consider the bread making process shown in the figure below. Someone developed it for making bread. The bread made using this process has various characteristics that consumers find desirable: look, feel, taste, etc. Each of these characteristic can be measured and will have some target value (based on consumer research) such that if a loaf is made with all its characteristics on target, there is a high probability that it will meet the consumer’s expectation and the consumer will enjoy eating it. The question that process validation seeks to answer is will this process consistently produce bread loaves of the specified quality.

A fundamental assumption in manufacturing is that if the inputs to the process remain constant e.g. you use exactly 6 cups of bread flour each time, and the process itself is constant e.g. the oven generates 350 F of heat every time, then each output of the process will be the same as the previous with no discernable difference. However, nothing is constant: there is natural variability in the quantity of the flour used; sometimes you might use as little as 5.5 cups; other times you might use as much as 6.5 cups. Even the oven periodically turns its heating mechanism on and off to provide a mean temperature of 350 F, but the actual temperature at any given instant is more likely than not to be above or below the mean. So in the physical world each output of the process i.e. each loaf of bread will be different from the previous.

The question then becomes is the loaf to loaf variability in the output, the result of the variability in the inputs and the process, noticeable by the consumer? Each characteristic of the output not only has a target value but also a range about the target that is considered acceptable. The bread may be okay if its crust is slightly more or less brown, but rejected if it is significantly dark (suggesting burnt) or light (suggesting underdone). What exactly are the limits of acceptability for each characteristic? That is decided through consumer research. Assuming, for our purposes that these limits are already specified, then if the measured value of a particular characteristic for a given loaf of bread falls within its upper specification limit and lower specification limit, it is considered acceptable.

During process validation the process is kept constant i.e. step sequence, parameter settings, etc. are fixed, while its inputs are varied between their extreme possible conditions. The thought is if the output of the process subjected to such extreme conditions of its inputs is within acceptable limits, then the output of the process with normal conditions of inputs will also be acceptable. The intent of this exercise is to demonstrate the robustness of the process to the natural variations in its inputs.

The design of experiments provides an efficient way to simultaneously vary every input between its extremes. For the bread making process in this example, there are 6 inputs: amount of bread flour, salt, vegetable oil, active dry yeast, white sugar and water. If we assume that each of these inputs will vary from their specified quantity as shown in the table below, then we can construct a two level six factor experiment for the process validation study.

 

 

Low (-)

High (+)

A

Bread flour (cups)

5.75

6.25

B

Salt (teaspoon)

1.25

1.75

C

Vegetable oil (cups)

3/16

5/16

D

Active dry yeast (tablespoon)

1.25

1.75

E

White sugar

5/9

7/9

F

100F warm water

1.75

2.25

Such an experiment is referred to as a full factorial experiment i.e. one where every combination of high and low values of every factor is made. Each combination will then be run through the process in randomized order. And each resulting loaf of bread will have various quality characteristics measured e.g. look (I), feel (II), and taste (III). These measured values will be plotted on separate run charts with their respective specification limits drawn in. The expectation is that the actual values will all fall within the spec limits. If that is the case, we can state with confidence that as long as the input variables remain within the upper and lower limits of their respective specifications, the quality characteristics of the resulting output will also be within their respective specification limits. And, thus we can conclude that the process is validated… for the set of inputs specifications defined.

Links
[1] Guidance for Industry — Process Validation: General Principles and Practices. U.S. Department of Health and Human Services, Food and Drug Administration, CDER/CBER/CVM. 2011. Web.

Appendix – Full factorial experiment design (order not randomized)

A

B

C

D

E

F

I

II

III

1

+

+

+

+

+

+

2

+

+

+

+

+

3

+

+

+

+

+

4

+

+

+

+

5

+

+

+

+

+

6

+

+

+

+

7

+

+

+

+

8

+

+

+

9

+

+

+

+

+

10

+

+

+

+

11

+

+

+

+

12

+

+

+

13

+

+

+

+

14

+

+

+

15

+

+

+

16

+

+

17

+

+

+

+

+

18

+

+

+

+

19

+

+

+

+

20

+

+

+

21

+

+

+

+

22

+

+

+

23

+

+

+

24

+

+

25

+

+

+

+

26

+

+

+

27

+

+

+

28

+

+

29

+

+

+

30

+

+

31

+

+

32

+

33

+

+

+

+

+

34

+

+

+

+

35

+

+

+

+

36

+

+

+

37

+

+

+

+

38

+

+

+

39

+

+

+

40

+

+

41

+

+

+

+

42

+

+

+

43

+

+

+

44

+

+

45

+

+

+

46

+

+

47

+

+

48

+

49

+

+

+

+

50

+

+

+

51

+

+

+

52

+

+

53

+

+

+

54

+

+

55

+

+

56

+

57

+

+

+

58

+

+

59

+

+

60

+

61

+

+

62

+

63

+

64

On Validation and Verification

The intent underlying validation or verification activities is to answer the question “How do you know?”. How do you know the product you designed meets the requirements for its intended use? How do you know a given unit you manufactured, based on that design, will perform as expected in the field? Objective evidence is needed to demonstrate that requirements of a given product that define its fitness for a specific purpose will be consistently fulfilled. For industries whose products can have a harmful impact on life, the requirement to perform validation and verification activities is codified in regulations.

For the purposes of this discussion, let’s assume that the product’s design fulfills the requirements of its intended use. This will allow us to focus on just its manufacturing process. How can you show that an output of the manufacturing process meets the requirements that define its fitness for use? One mechanism is to inspect and test every manufactured unit. And, so long as such activities do not destroy the manufactured unit in the process, it is a perfectly acceptable method to show its fitness for use. This sort of check is referred to as product verification.

However, inspection and testing of certain performance characteristics of a manufactured unit do destroy the unit in the process. Verifying such characteristics of each manufactured unit would not leave any units for use or sale. To address this issue, we have to look at data from samples of manufactured units, viewed through the lens of statistical theory, to draw conclusions about their overall population. This is the basis of process validation.

As long as the distribution of data points describing a particular performance characteristic of the product, collected from samples of manufactured units, falls within the limits that define the performance requirements for that particular product characteristic, we can be confident that the rest of the population of manufactured product meets those performance requirements as well. The theory of statistics provides us with a mechanism by which to quantitatively express the degree of our confidence that untested units will perform as expected.

An implicit point made in my assertion is that the manufacturing process is subjected to the range of variability present in its inputs. Each input to the manufacturing process has a distribution that describes its center and variation. These inputs interact with the manufacturing process and with each other as they are transformed into the output with its own distribution. However, the shape, location and spread of the output distribution is only revealed after significant data has been collected over time. And, it is the boundaries of this distribution when compared against requirements that demonstrates whether the manufacturing process can consistently produce units that are fit for use.

Companies do not have unlimited time or money to collect such data by conducting large numbers of manufacturing runs. And, such a large data set isn’t necessary either. Subjecting the manufacturing process to extreme values of each input will yield output values that represent the boundaries of the manufactured product. It is reasonable to expect that if the inputs are within their extreme values, the output will be within its boundary values. If the boundary values of the output are within the limits that define performance requirements, we can rest assured that the manufacturing process will produce units that are fit for the intended purpose. And, this manufacturing process can then be said to have been validated to produce the particular part.

A final thought: manufacturing processes have several inputs. It is not efficient to vary them one at a time. In fact, varying them one at a time doesn’t give you the complete picture of how they interact with the manufacturing process or with each other. Running controlled experiments that are properly designed can paint a more full picture. The science of the design of experiments should be the tool of choice when validating a process.