Formulating
Many years ago, when I was Director of the Smithsonian’s Furniture Conservation Training Program, I was meeting individually with each student in preparation for their fourth-year fellowship, during which they were required to undertake a project that would result in their Master’s Thesis. The students had great latitude in their projects, some were purely historical aesthetics, some were about historical technology or craft, and some were analytical.
After our conversations I would arrange for each student to have a “mentor” to help them with guidance along the way. For those projects reliant on analytical data I brought in a highly respected statistician with a specialty in experimental design and data analysis to advise the students in designing their project, gathering data, analyzing that data and formulating the conclusions. One of the great beauties of DC is the abundance of research institutions and their scholarly communities. (The stories the statistician told me sotto voce about “research” shenanigans made me distrustful of any “science” ever since, even before the anti-“science” of the past three years. One tale literally revolved around a “researcher” bringing in a box of lab notebooks and dropping them on the statistician’s desk the with the instructions, “My conclusion for this project is XYZ so you need to review this data and arrive at this conclusion.” As the experimental designer told me then, “science” is not a thing, science is a process, and if the process is corrupt then the outcome [data and conclusions] is worthless. From that perspective I can barely withhold laughter when public luminaries now tell us to “follow the science.”)
Back to my student. The topic proposed for the thesis project was an evaluation of shellac properties, the particulars are lost to me at the moment. I only remember that the number of variables combined with the number of identical samples required for a statistically valid set of results would have require formulating and preparing approximately seven million samples. Needless to say, the student changed their project rather fundamentally.
Which brings me to work I am undertaking in the studio right now and probably for weeks or months to come. I am not the trained experimental scientist in this household — that would be Mrs. Barn — but I am fairly able to harness and focus my curiosities from time to time. My demonstration of making my artificial tortoiseshell several weeks ago has re-lit that fire for me and I have been working on refining the formulation and process ever since. Although my paper from two decades still stands up well, it only gets about >95% of the way to a really good imitation vis-a-vie the physical properties of genuine tortoiseshell. >95% is not the same thing as ~99%, which is where I want to go.
Like my student’s those years ago every variable change requires a group of samples to be formulated and made. Unlike the putative shellac researcher I am NOT weighing each variable as random and equal. I am not making the number of samples that might be required if the variables were purely random; I am testing one variable first, then applying the second variable once the first spec is established, then the third once the first two are established, etc. Even so I am creating hundreds of samples to assess for their properties based on a menu of options I must consider.
What grade of collagen should I use?
What concentration?
Which plasticizer (if any)?
At what concentration?
Which protein reaction catalyst to use?
In what proportion?
In solution or infused ex poste?
How long to cure?
At what temperature?
My tools for this undertaking are fairly simple and non-specialized; an analytical digital scale, disposable pipettes, disposable cups, a old microwave, an ancient ebay stirring hotplate, a rice steamer (most samples curl when they dry and need to be steamed flat), drying screens, and a desiccating chamber, a/k/a a Gammo pet food storage unit filled with conditioned silica gel. By far the most time for the sample involves drying them to the lowest moisture content possible, at which time the most extreme properties become manifest. Yes indeed, most of the time is watching samples dry.
There is simply no point in ongoing daily blogging about that, or in changing formulations by a hundredth of a unit proportion, but I will report back when I have results that I find encouraging in directing my final minute adjustments. In the meantime when there is radio silence on the blog, you can assume I am either splitting firewood for next winter and beyond, or tinkering with formula minutiae.
One of my real headaches at the moment has to do with a German chemical component that has a) become unavailable, or b) become unaffordable. My current dwindling inventory was a 100g jar given to me by the president of a chemical company almost twenty years ago, and in the intervening years I have seen the price go from a couple hundred dollars a kilo to several hundred dollars per kilo to a few thousand dollars a kilo to a quote yesterday of almost $100k per kilo. I’m really wishing I’d bought a big bucket of it fifteen years ago. Obviously, I am doing my best to work around that headache.
Stay tuned.
Hi Don. I would encourage you to find another design of experiments expert for help. One factor at a time experimentation requires excessive numbers of trials compared to a well-designed multi-factor experiment. More importantly, one factor at a time can’t discern interactions between the factors, which are almost certainly present in your system. (i.e., optimizing one factor, holding that and optimizing the next factor will almost certainly not lead you to the global optimum.) I’ve long been intrigued by your tordonshell, but I’ve never found the paper that you mention. Could you post the reference some time?
Have you considered a “Design of Experiments” formulation with a fractional design?