2019 has been the year of Water! Acute Water Shortage in Summer across India and now devastating flood in the Rainy Season. Our prayers with all those affected!
With rising apartments and rising cost of water, there is sudden demand for individual water metering across homes, offices. I have seen a number of start-ups jumping into the fray with “Smart” Water Meters. Its good to see the indigenous push towards water meters, with ultrasonic measurement, telemetry, accuracy etc.
While it is tempting to get the “state of the art smart” meter, it is very important to first weigh your true needs of metering with the long-term costs of continued maintenance, repair, service contracts and subscriptions over next 5-10 years, especially with the poor water quality with “hard water” and scaling – All considering that the water supply in India comprises of mix of public utilities, borewells and tankers, resulting in poor quality of water flowing through pipes.
Considering maintenance-free operation across all types of water, its natural to turn towards the good old Mechanical Meters that have been doing business in India for many years. These meters are economical (7-10$), ISI certified, easily available and above all, can deal with any water flowing through it for many years and needs no batteries to operate. We should continue to use such a reliable, mature technology for measuring water consumption.
The problem of getting the reading of mechanical meters to the cloud, or telemetry, can easily be solved by deploying latest low cost cameras on top of the meters at a fixed focus. The prices of camera modules have dropped significantly in the last year and we can pick up a 2MP Camera module operated by a Microprocessor at 5$-10$ retail price today without volume discounts.
Check out the ESP-EYE module based on the Espressif ESP32, which packs the 2MP Camera Sensor along with heavy duty Dual-core Microprocessor with WiFi/BT/BLE in a small form factor.
If we put such a low cost camera to infer the reading from a mechanical water meter, the benefits of such a simple to deploy system would be fabulous.
A basic Machine Learning system would easily be able to recognise the meter reading from the image of the water-meter on the Edge using Tensor Flow or similar tools – given that the image and the meter is always in the same lighting, same position and same model across all units. This approach can be scaled to a large number of apartment units using industry grade RS-485 and M-bus protocols to create a maintenance free, reliable and economical solution. Wise Water Meters seeks to use our method for the solution, check out their website for more details.
Here’s what we did so far; We put such a camera-module on-to a Mechanical Meter to capture the images and generate a ML model on the server. While the ML model is being developed, there is a manual backend to catalogue these images and arrive at the right reading today itself – This allows us an automated real-life scenario to get real meter images for ML learning while ensuring customer is happy from Day 1. Once the ML model will be stable, all manual classification will be avoided and the whole system will be autonomous.
I expect the final ML Model on the Edge to be ready in 2 months along with our production ready design of the telemetry unit that fits as an after-market on popular mechanical water meters in India. Checkout one of our prototype deployments at home… Reach out to us at firstname.lastname@example.org, if you want to partner in commercialisation or have better ideas about this.