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.