This is a nice, warm, fuzzy idea with very little scientific (or engineering) rigor behind it. It will not achieve what is intended.

Now, I am as much a geek as anyone, and love to play with massive amounts of data to do cross-correlations and time series and derive neat-o prediction schemes and link cause to effect….. But before I can do that, I really need to have accurate data. Otherwise, I might as well make up my own data. What this idea will generate is massive amounts of data that can’t be correlated to the same axes.

If you want to compare data from several different sensors, then the sensors MUST be calibrated to a common standard. AND the relative accuracy of each sensor must be known.

Something that is almost always ignored by those outside of the instrumentation and control field is that each data point has two parts: a measurement and an accuracy: x +/- y. Without both pieces of the datum, it is worthless. It is only good enough to produce Congressional budgets.

So here is my analysis of this wishful thinking:

“In order to be useful, we need to ensure we can compare data relatively faithfully across multiple sensors.”

True.

“This doesn?t need to be perfect, nor do they all need to be calibrated together, we simply need to ensure that they are ?more or less? recording the same thing with similar levels of precision and consistency.”

False. The author is using “perfect” in terms of “better than I need to play around with it”, instead of its real (and unattainable) meaning of having no error. In fact, since there is no accuracy linked with each datum, it is assumed to be perfect.
When he says that the sensors need to be “more or less”, then he is vaguely sensing that there needs to be an accuracy attached somewhere. See? Common sense is innate, just not being used here.

“Ultimately in a lot of instances we care about trended data rather than individual points so this isn?t a big problem so long as an individual sensor is relatively consistent and there isn?t ridiculous variation between sensors if they were put in the same conditions.”

False. This is just pure wishful thinking. Without the accuracy data, you can only trend data from each sensor. You will not be able to do any meaningful trending between groups of sensors and their associated data. Without calibration results and sensor accuracy information, the data is not “apples to apples”.

If you don’t know what you are doing, then you come up with statements like these that will start people out on the primrose path. The whole exercise will be wasted effort, or worse, bad data bringing bad decisions, unless it follows standard (and boring) practice from past instrumentation lessons learned the hard way.

I would like to point out that the lessons learned the hard way usually involved massive destruction. When I worked in the instrumentation field around 1980, the petrochemical industry blew up a plant every year. Or was it two a year? And blow up doesn’t mean localized damage in one part of the plant, it means a plant completely leveled. I remember one plastics company sending their design team to our Engineering Center after completely leveling their plant. They sent 25 people. We had talked them down from 75. They were extremely concerned that it not happen again.

Good decisions start with good data. Good data comes from good design. Good design takes into account how to correlate the data between different measurements. Talk to some of those who have made the mistakes already. It will save a lot of grief farther on down the road.

artp