Statistical relevent sampling plan for powders, 1000-2000 g

[#1]

My company recently purchased licenses for JMP and currently running 17.2.0. I have been tasked to write a PQ on a couple of our powder manufacturing processes. I have been told to use a statistically relevant sampling plan and one of the engineers pointed me to the JMP sampling plan add-in but I don't think its the right add-in to use for powders. It seems like a statistical plan for sampling items, e.g. bolts. I wanted to use a similar add-in for about 1000-2000g of powders. The JMP sampling add-in for variable sampling when used recommends 3 samples per lot and we double to 6 samples for assurance which may be good enough but the statistical justification isn't there. We usually tray dry in 1-6 trays depending on process and amount. Our current sampling practice for lot release testing is pulling one 3 g sample but this doesn't seem like a statistically relevant sample for uniformity. We don't want to change the procedure to add a mixing step and then pull a sample, and would rather prove that the powder is homogenous by pulling several random samples. i just don't know how many is relevant. 

[#2 · Accepted solution]

What is the purpose of your sampling?
Is it to prove that the powder is acceptable?
Or is the powder uniform?
Or both?
If you want to prove that the powder is homogeneous, then you need to define the homogeneity difference, as well as determine the sampling point location, which statistics can't tell you.

[#3 · Accepted solution]

Similar to the previous response, there is not enough information (what is known about the powder? mean, variation, consistency).  I believe the question is: how to get a representative sample?  You must think about the different components of variation.  For example, measurement system, within tray, between tray, within lot, between lot, etc.  What are the possible sources of variation (hypotheses about factors that affect the powder)?  How do you ensure these vary in your sample?  These are not statistical questions.  You need SME to help decide what a representative sample is.  It is not a question of sample size, but how the sample is acquired.

[#4]

Thanks @lehaofeng and @statman. Makes sense. I asked the same questions early on before I started writing the protocol but was told that we didn't want to spend the extra money for now. I let the experts decide how we want to sample and the sampling plan direction they were ok with I mentioned in my OP. Might be from their past experience. Looks like I will have to accept it for now and hopefully be able to convince them later when the processes are optimized and revalidated. Thanks again!