Developers digitize water, address IoT solutions with Azure IoT Hub and Azure Machine Learning
Print
August 30, 2017
We usually think of the Internet of Things as a network of connected devices. Developers at the Norwegian software company Powel have discovered that IoT can also be extended to nondurable resources like water. Through the careful collection of sensor data, Powel’s software engineers transform water currents into data streams. These streams are then analyzed using machine-learning algorithms to locate problems in municipal water supplies and alert the appropriate utilities, saving both money and a vital natural resource.
Want to try it yourself? Try the hands-on labs for Azure IoT Hub.
Want to experiment and explore for yourself? Get your free Azure account and receive $200 in credits now.
How do you make water smarter?
Municipalities in Norway are seeing an average water loss of more than 30 percent due to leakages in their water distribution networks. A major contributor to this problem is the age of the various components in the water distribution infrastructure. This high rate of water leakage has both a financial and an ecological cost.
Powel’s software developers went to the whiteboard and decomposed this resource issue into two distinct but related challenges. First, they asked themselves, how do you determine normal water flow/consumption when there are already leaks in the system? Second, how do you efficiently monitor water flow over time so you can detect new leaks?
Story image 2
From water flow to data streams
The developers realized they already had some data to work with. In most municipalities, water flow is tracked using supervisory control and data acquisition (SCADA) sensors. Using the Trondheim municipality as a test case, they retrieved SCADA data between 2013 and 2015. They next built a model with the data in Azure Machine Learning Studio that could be used to predict normal water flow throughout the year. “The ability to comprehend and analyze large amounts of data that the municipalities already had was the key,” says Froydis Sjovold, Solution Manager at Powel.
For the machine-learning algorithms to be useful in determining abnormal flow due to leakage, the Powel engineers’ predictive results still needed to be compared against live data from the SCADA sensors. Team members quickly settled on a tool for ingesting real-time sensor data for analysis. According to Kevin Gjerstad, Powel’s CTO for Cloud Technology, “The customers we are talking to will have many sensors, so we need something that scales—IoT Hub, for instance.”
IoT Hub has built-in security mechanisms. IoT Hub only allows registered devices to communicate with it, and each registered device gets its own “personal” connection string, ensuring that a compromised device’s connection string cannot be used to impersonate other devices. Additionally, device management in the Azure portal allows devices to be disconnected to stop a faulty device from transmitting data that is inconsistent or damaged in any form, which helps prevent bad readings.
During the training of the machine-learning model, the team needed a way to understand data they were working with. They discovered that Power BI Desktop was an excellent tool for exploring the data using various kinds of visualizations. This allowed them to quickly look for patterns that would provide insight into the data. Using Power BI in this way, the developers could identify the fields they needed to include in the dataset and what kind of cleaning needed to be done before they started experimenting with the data in machine learning.
Through this pre-analysis with Power BI, the Powel team learned that the water flow into the water distribution system follows a daily pattern, with some exceptions during special events and public holidays. To allow for a granular model for water flow within a 24-hour period, they split every 24 hours into 5-minute intervals. They also included a flag to indicate whether a given sample time was associated with a public holiday.
By combining a machine-learning solution for predictive results and an Azure IoT Hub solution for real-time data collection, Powel found a way to make it easy for the municipal water systems of Norway to self-report on their status and even tell the utilities when that status is abnormal. Powel effectively made monitoring of the water distribution smarter.
Stopping tomorrow’s leaks
The developers who built this project are currently integrating their Water Alert solution into Powel’s main product line. As water utilities install more sensors and smart water meters over time, the solution will have the potential to provide even more intelligence as real-time leakage warnings are cross-referenced with historical repair data to isolate the sources of leaks.
Powel’s CTO no longer sees water in the same way. “These utilities are sitting on treasure troves of data,” says Gjerstad. “But that data is locked away, and they don’t have enough insight into it today.” Through machine learning and IoT Hub, Powel is unlocking that data for clients and making it give up its secrets.
To learn more about their Water Alert solution, you can review the code samples and architectural diagrams created by the Powel developers on GitHub. You can also get hands-on with Azure IoT Hub labs or start developing a new solution with an Azure trial.
