The clam is a sensor platform for measuring water quality. So far, I have experimented with thermistors as stand-ins for more complex water quality sensors. I am developing a new saturometer, in order to bootstrap the capabilities of the clam. Saturometers are used to measure the total dissolved gas (TDG) in water. Monitoring TDG is important for understanding and mitigating issues related to supersaturation of water, which can cause large fish kills. I will take a naturalistic approach to how I develop the new saturometer probe, and will apply a machine learning technique, linear regression, for its calibration.

Acknowledgements

I am basing my work off of one of my dad's patents for the saturometer he developed in the early 1990's. I want to thank him for talking with me about my new design, and for letting me learn in his shop while I was a kid.

Introduction

My dad developed his saturometer before the internet was widespread. Powerful computers, SMD technology, and 3D printing were not widely available, nor was knowledge of machine learning techniques. My design for the clam is taking advantage of the developments over the last two decades. The redesign of the saturometer will focus on improving the ease of fabrication of the probe and improving the ease of calibration by incorporating linear regression, a simple machine learning technique.

Saturometers are used to measure total dissolved gas (TDG) in water. It is important to measure TDG, in order to study and mitigate supersaturation (for example, in this study of using microbubbles). Supersaturation can be caused by spillways at dams, in water turbine vents, hatcheries and other man-made or natural structure that entrain air in water. Additionally, it can be caused by other processes, such as photosynthesis of algae. Gas bubble disease (GDB) is caused by supersaturation and can cause large fish kills. Fish have a limited ability to detect supersaturation and cannot avoid it if present, so it is imperative that we monitor and mitigate it.

Sweeney Saturometer

Saturometers are able to measure TDG by taking advantage of a property of silicone: permeation. Gases are able to pass through silicone tubing, so by measuring the pressure within the silicone tubing, the saturometer is able to measure the TDG within the fluid. The saturometer probe is composed of silicone tubing wound around a support structure and connected to a pressure sensor inside a protective mesh cover. The probe is attached to a metal tube that produces waterflow across the probe when moved in the water column. Waterflow is important for getting accurate readings. Here is a Youtube video of the saturometer in action.

To calibrate the saturometer, it is necessary to use an external device. The probe can be calibrated without the silicone tubing using an air compressor, a manometer and a coupling. It can also be calibrated submerged in water with a controlled pressure and another saturometer.

In order to make fast measurements, the thickness of the tubing needs to be decreased, and the surface area of the tubing increased by increasing the length of the tubing. The volume of the tubing can be decreased by inserting a filament like fishing line into the tubing. Additionally, by adjusting the calculations of the measurements and looking at the rate of change, fast estimates can be made before the pressures have reached equilibrium.

Siphon: Initial Design & Fabrication

My design for the saturometer probe will use a coil of silicone tubing without an armature. The tubing will be placed inside a silicone siphon, and be packed loosely. 

I will use the same methods and materials that I used for the previous clam experiments. I will use the same electronic circuit but add op-amps and an absolute pressure sensor. I will pick through the old inventory of saturometer parts and use what is available. There are some inexpensive modern SMD pressure sensors that I would like to experiment with, but that part of the design can be updated later.

There is a lot of fascinating work being done in soft robotics and printing soft 3D structures, but I will construct this siphon in the same way that I constructed the initial bladders.

The saturometer needs waterflow for accurate measurements, so the siphon needs to be pumped. I think that the existing adductor solenoid can be repurposed to pump the siphon. The design for the siphon is inspired by the heart, and in effect, is half a heart. It will be made up of one atrium and one ventricle with two valves. The atrioventricular valve closes under pressure from the ventricle, and the semilunar valve opens under pressure from the ventricle. There are many ways to make artificial valves, including using encapsulated balls.

I will need to experiment where the silicone tubing is placed inside the siphon. It would be less disturbed before or after the ventricle, but there might be an advantage to having it inside the ventricle.

ML Calibration: Linear Regression

The calibration process that my dad did by hand is well-suited for linear regression. In his saturometer, there are a series of potentiometers that need to be adjusted to fit the sensor input to the target output. I will use the bluetooth module to receive the target output from another device, and on the atmega fit the incoming sensor data on the ADC to the target output.

There are many interesting potential applications of machine learning techniques, like anomaly detection. Additionally, if underwater communication could be established, group calibration and other interesting behaviors could be developed.