Sump pump monitoring with a Raspberry Pi

I am not much of a mathematician, and even less of a programmer.  But I am at my most dangerous when wielding a screwdriver, attempting to be “handy.”  So this project was a lot of fun.

My house is situated on a hill, and the sump pump in my basement acts up seasonally, rain or shine, making me suspect a possible underground spring or stream.  To test this theory, my goal was to monitor the behavior of the sump pump, measure the rate at which water is ejected from the pit, and as a safety feature, to automatically send myself text messages in the event of any of several possible failure conditions (e.g., pump or float switch failure, excessive depth of water in the pit, etc.).

The setup is shown in the diagram below.  I was able to use mostly existing equipment from past educational projects, with the exception of some wiring, a length of PVC pipe, and an eTape liquid level sensor that I wanted to try out.  The sensor is a length of tape submerged in the sump pit (inside a PVC pipe to keep it vertical); it acts as a variable resistor, with the resistance decreasing linearly as the water rises in the pit.  Using a voltage divider circuit, the resistance is (indirectly) measured as a voltage drop across the tape, using a DI-145 analog-to-digital converter.  The A/D has a simple USB-serial output interface with a Raspberry Pi, where a few Python scripts handle the data recording, filtering, and messaging.

Raspberry Pi 3 monitoring DI-145 A/D converted voltage across eTape liquid level sensor.

Raspberry Pi 3 monitoring DI-145 A/D converted voltage across eTape liquid level sensor.

The figure below shows a sample of one hour’s worth of recorded data.  The y-axis is the voltage drop across the tape, where lower voltage indicates higher water depth.  Note that the resistance is linear in the water depth, but the voltage is not.

One hour of recorded voltage: raw 120 Hz input (green), filtered (red), with detected pump cycles (blue).

One hour of recorded voltage: raw 120 Hz input (green), filtered (red), with detected pump cycles (blue).

The green curve is the raw voltage; I’m not sure whether the noisy behavior is due to the tape sensor or the A/D.  Fortunately the DI-145 samples at 120 Hz (despite all documentation suggesting 240 Hz), so that a simple low-gain filter, shown in red, cleans things up enough to allow equally simple edge detections of the pump turning on, shown in blue.  (And this is after nearly two weeks of no rain.)  Based on those edge detections, the program sends me text messages with periodic updates, or alerts if the water depth gets too high, or the pump starts cycling too frequently, or abruptly stops cycling.

All of the Python code is available at the usual location here.