I got a really wild idea this morning, one too wild to keep to myself.
Estimating mash pH, an endeavour that receives more than a little attention here, is actually a simple task if one uses a model based on the chemistry of acids and bases. But, with few exceptions, the people who like to publish pH predictors do not, will not, cannot or do not choose to understand that model. Now there is a special set of techniques for controlling systems or estimating the parameters of systems whose models we cannot, do not or do not choose to understand and those use so called fuzzy variables rather than the "crisp" one we would use in estimating pH from the chemistry model.
Those familiar with fuzzy control (used in the inexpensive "PID" controllers so many of us use in brewing) will know what I am talking about. For the rest the general idea is that the "crisp" quantities of grain, the amount of acid, the alkalinity of the water etc. are "fuzzified" into "lingusitic" variables that have values like "added acid is moderate" or "caramel malt content is low". Then "rules" using these variable such as "if alkalinity is high AND caramel malt content is low AND black malt content is low AND.... then pH is medium. These rules are formed based on what "experts" know about mashing and go into a "knowledge base". An inference engine combines (using very simple math) the rules that apply (and there will be several) to determine "membership" in output sets corresponding to "mash pH is midrange", "mash pH is a bit low", "mash pH is quite low" etc. and a "defuzzifier" turns those membership values into a crisp pH estimate.
The algorithm is really quite simple (but I said that about the chemical model too, didn't I), simple enough that the Wikipedia description ought to be able enough to get an interested investigator started.
The big advantage is that as knowledge is gained in can be incorporated into the knowledge base or the knowledge base can be modified if observations show that the current one is is faulty in some regard. This should be of tremendous appeal to those that basically favor the empirical approach to the deterministic one for whatever reason.
Caveat/Disclosure: I don't really know much about this. I've been aware of fuzzy control for years but never really used it in a project.
Estimating mash pH, an endeavour that receives more than a little attention here, is actually a simple task if one uses a model based on the chemistry of acids and bases. But, with few exceptions, the people who like to publish pH predictors do not, will not, cannot or do not choose to understand that model. Now there is a special set of techniques for controlling systems or estimating the parameters of systems whose models we cannot, do not or do not choose to understand and those use so called fuzzy variables rather than the "crisp" one we would use in estimating pH from the chemistry model.
Those familiar with fuzzy control (used in the inexpensive "PID" controllers so many of us use in brewing) will know what I am talking about. For the rest the general idea is that the "crisp" quantities of grain, the amount of acid, the alkalinity of the water etc. are "fuzzified" into "lingusitic" variables that have values like "added acid is moderate" or "caramel malt content is low". Then "rules" using these variable such as "if alkalinity is high AND caramel malt content is low AND black malt content is low AND.... then pH is medium. These rules are formed based on what "experts" know about mashing and go into a "knowledge base". An inference engine combines (using very simple math) the rules that apply (and there will be several) to determine "membership" in output sets corresponding to "mash pH is midrange", "mash pH is a bit low", "mash pH is quite low" etc. and a "defuzzifier" turns those membership values into a crisp pH estimate.
The algorithm is really quite simple (but I said that about the chemical model too, didn't I), simple enough that the Wikipedia description ought to be able enough to get an interested investigator started.
The big advantage is that as knowledge is gained in can be incorporated into the knowledge base or the knowledge base can be modified if observations show that the current one is is faulty in some regard. This should be of tremendous appeal to those that basically favor the empirical approach to the deterministic one for whatever reason.
Caveat/Disclosure: I don't really know much about this. I've been aware of fuzzy control for years but never really used it in a project.
