Monthly Archives: August, 2011

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.

Note: A full factorial experiment is not necessary. I will have more to say about this in another post.

References

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

Advertisements