Earth · Case study

Dripito

An open-source drip monitor for children on IV infusions.

277 commits · 19 lab-verified runs · 16 design decisions · 2 DOI releases

In many humanitarian and rural clinics, there are no infusion pumps. A study of 30 neonatal facilities in Sub-Saharan Africa found that 73 percent lack reliable access to one. A nurse sets the flow rate by counting falling drops against a watch. A wrong rate can be fatal for a small child.

Dripito clips onto the drip chamber. Two infrared beams watch the falling drops. The device counts each drop and measures its volume. It runs on one AA battery. I built it as my bachelor thesis and my semester project at ETH Zürich, with the Global Health Engineering chair.

The first version counts drops. In lab tests, it missed 1 drop in 1,430. The second version measures each drop. Drops are not uniform: within one drip set, the drop volume varied 4.7 times. So the device times each drop between the two beams and computes its volume from the time it takes to fall between them. It calibrates itself at every start.

The lab tests also showed the limit of this design. Volume accuracy depends on the mounting position, and none of the algorithms I tested brought it within 5 percent. I documented this failure with the raw data, isolated three physical causes, and specified the fix for the next revision.

The project is open hardware: three licenses, 16 written design decisions, 17 documented limitations, and two archived releases with permanent citation links (DOIs). The analysis rebuilds from the raw data with one command (Docker). The Global Health Engineering chair at ETH hosts its own copy of the repository. The repository documents how to rebuild the device, and which three problems to solve next.

Technical notes

• The firmware samples both light sensors every 30 microseconds and interpolates each threshold crossing between two samples. This gives microsecond edge timing from a polled microcontroller loop.
• The power design has two phases. First, both beams stay on to measure the real drop volume of the session. Then one beam turns on for only 0.3 percent of the time, and a simple analog circuit (a comparator) wakes the processor for each drop. This requirement determined the choice of the microcontroller.
• The reference measurement itself needed a correction: drop counts at the chamber and at the end of the tubing differ by 1.5 to 2 times. Only the weighed mass of each chamber drop is a valid reference. The invalidated analysis tool stays in the repository with an explanation of why it is wrong.