We are involved in a company project to understand whether the samples we take are taken correctly (according to the Theory of Sampling). We produce animal feeds in our mills as separate batches (with each batch usually being very different to each other in terms of their analysis) and sampled at a point below the mixer.
We have been introduced to Variograms by a geologist and I am attempting to apply this technique to our quality sampling and analysis of our feeds and raw materials – to test the “spatial correlation” or complete lack of it – i.e. should we give up sampling at that point at all?
I have been trying to understand what the Variogram is attempting to measure and its practical application to our processes. I also understand that the Variogram is just the first step, the second (assuming I need to move to this) being using this information for use in the Kriging model.
Looking at some real life Variograms it appears that calculation of the Nugget, and Sill/Range is a subjective determination especially if the curve is wavy (periodicity or not?) and then whether it should be spherical or exponential etc.
So my first query is whether the Variogram model is at all valid for our process and if so does JMP have software tools for this and is further modelling i.e. Kriging of use? And if this isn’t applicable are there other JMP statistical models to answer the question “should we give up sampling at that point at all”. Hope all this makes sense. I have JMP Standard. Thank you.
I like your statement that you are "are trying to understand what the Variogram is attempting to measure and its practical application to our process". I have no experience with variograms (had to google it). This small book may be more applicable to your industry: ASA Siam Sampling Book, it covers basics on understanding taking random samples in manufacturing. I would take a step back and ask why are we sampling - presumably to monitor your product quality - so now link sampling to product quality (not always an easy thing to do). What are the key inputs that have the greater impact on the final quality product (this speaks to design of experiments and or data mining of existing data to then inform a better experiment). If and when you understand the input/output then you can work on where and how to sample...and what and how to test.
Looking forward to anyone who has input on variograms to better understand that concept and if it is useful in manufacturing.
Thank you Karen. Some good thoughts - not come across the book (reference) either. DoE is an interesting approach, and not one I thought of using in our attempts to see if our sampling points are fit-for-purpose. Another learning opportunity I guess (so many elements to bring together)! Thanks again.
