Results and calibration routines

[jonnor]

Hi, thank you for developing and publishing this project! I'm looking into sensors use in a personal off-grid system.

The documentation says the goal is to have accuracy of "0.05 NTU from 0-5 NTU". Have you tested a completed device, compared it to other methods? What were your findings?

Futhermore, I presume that after making the device, one should verify/calibrate it with known samples. Do you have any references for procedures on how to do this?

[alexkrolick]

Hi Jon,
You might want to check out the blog and store

You can purchase commercial calibration samples from turbidimeter vendors in a compatible size; @iamchriskelley would know which ones.

[iamchriskelley]

Hello Jon,

Thanks for raising the issue. You can definitely purchase commercial-grade calibration kits to validate an open-source turbidimeter (Here's one that's suitable), but they're generally expensive. Please note, commercial turbidimeters have sophisticated techniques for linearizing the turbidity response curve so that a three-point calibration is sufficient. For an open-source turbidimeter without that feature, you'll need a few more calibration points (the turbidity response curve is typically closer to a cubic curve) than a commercial calibration kit will give you.

The best method is to borrow a commercial turbidimeter to calibrate against. You can then use suspensions of hydrophillic cutting oil in water, or even milk in water, as your turbidity samples.

There are at least two things that would make this process easier for makers and users of open-source turbidimeters:

(1) Cheap, shelf-stable turbidity calibration samples.
(2) A mechanism that allows users to enter in their calibration data (known NTU's of samples versus raw sensor readings of an uncalibrated instrument), perform a polynomial regression on the dataset, and transmits calibration constants (and goodness of fit) back to the uncalibrated instrument. This could be done in a smartphone app, or onboard the turbidimeter's microprocessor if the microprocessor is up to the task.

I'm working on (2) now as an Android and iOS app; it's about halfway done but I'm currently pretty busy with other projects. Hopefully I'll get a chance to tackle (1) this year with some solid turbidity standards made of translucent polymer.

As for the goal of 0.05 NTU accuracy over 0-5 NTU, it remains a goal. By increasing the sampling time of a light-to-frequency sensor to several seconds per read, or swapping the light-to-frequency sensor for a photodiode and transimpedance amplifier, you could get a turbidimeter with that level of accuracy. Issues like case stability would become very important; a case printed on a home 3D printer might have too much flexibility to ensure precise alignment of the optical instrumentation. And you might find that the anisotropy of regular (borosilicate) glass cuvettes is too much of a hassle to deal with in calibrations, so you could go with quartz cuvettes instead. Even quartz cuvettes have some anisotropy though; my lab turbidimeter gives a reading variation of almost 0.2 NTU when I take a series of readings while rotating a quartz cuvette of distilled water in the turbidimeter. Which brings to mind a third thing, not currently available on the market, that could help open-source turbidimeter users and makers: one-inch path length square acrylic cuvettes. I hope to have time before April to put together some prototypes of those.

Sorry if this is an info overload, and please let me know if you have any more questions.