#WaterData18 - part 2
With so much to report on, the write-up for #WaterData18 has been split in to two epic parts. Read on for the successful sequel to part 1.
Day 2 - Saturday 19 May
Inspired by the glorious weather, the Saturday was a more relaxed affair with only a short get together over breakfast tea/coffee to make sure everyone knew what they were doing. As a few people could only attend for the Friday, some teams had shrunk considerably or had disappeared altogether. Thankfully there were people who were happy to step in and help complete prototypes in time for the show-and-tell in the afternoon.
We filmed and live-streamed the final presentations, so for those of you who want to watch the best bits, the video is available on YouTube. Happy to keep reading? Then do so!
The first team to be brave and share their hard work was the team nicknamed 'The Time Lords' (no one actually knew why). They chose to focus on the Leakage DMA dataset, working towards identifying a 'background level' of flow or leakage. The only variables they used was the number of properties in each DMA (domestic and non-domestic) and one external dataset that provided the typical number of rooms per domestic property type for an area. With this, the team created expected flow predictions that could then be compared to the actual flow data to spot anything unusual. With additional data, the models could become more detailed. In order for the model to be useful, the outcomes were visualised on a map and for further investigation, you could click on a DMA to be presented with a line-graph of flow (similar to what Yorkshire Water already use). As an aside, they also took a quick stab at the acoustic logger sample sound files, plotting the signal frequency on a spectrogram. With more sound files and a sprinkle of machine learning, leaks could be found from the patterns of the sounds.
Next to step into the fray was Stephen, a one-man team who had experienced a bit of a setback after his laptop had spent all night updating (less time for number crunching). His idea from the first day - using APIs to create series data - had been put on the back-burner. However, he did do some creative exploration of the data. He turned the average sound levels of the acoustic logger dataset into ASCII art!
From little to large - this team was made up of individuals from 7 different companies! All working together in harmony to explore Yorkshire Water data and create cool things. This team also looked at identifying a 'background level' of flow or leakage. For each DMA, they analysed the data between 3am and 4am (times could be changed for future analysis) and created a 'baseload.' They then took the top 10 DMAs with the lowest number of non-domestic properties and created a profile. They did the same again but for the top 10 DMAs with the lowest number of domestic properties. These profiles were then applied to all of the other DMAs so that actual flow data could be compared to the profile and the baseload figure calculated earlier. The difference between the profile and the actual flow data gives you leaks. Although incomplete due to time constraints, the team started to look at combining things like geographic data into their model. In particular, they started with the topographical data that they could access thanks to Clive Mellor (from OS) who provided temporary access to API. The goal is to create more detailed profiles by incorporating data from other sources.
Back to a one-man team for Paul Garside from Bradford College. His initial ideas from day 1 - nano submarine-drones, using additives in water to trace leaks, and triangulation of the acoustic loggers by using mobile devices as well - actually all turned out to be currently in development or currently in use! So he explored a fourth idea that used Doppler radar to help pinpoint leaks. Because Doppler radar can detect movement, direction, and velocity, he theorised that it could pick up the changes in flow caused by leaks. He had even crafted a physical demo, using Smarties and a plastic bottle. He did admit that Doppler radar might have limitations, for example not working accurately on curved pipes, but he was clear from the start that Doppler radar methods wouldn't replace other leak detection methods. It is meant to complement what is already in place but simply aid in finding precise leak locations.
The next team came about as a part-collaboration with ODI Leeds data projects person Stuart. The original team had dwindled to just Sam from Yorkshire Water but her ideas aligned nicely with a visualisation that Stuart had been developing. Sam had started to analyse the acoustic logger data, looking at the instances that triggered alarms for investigation. How many of these were false positives? How many were genuine leaks, and was the threshold for triggering alarms appropriate? Sam then proposed that looking at the sound data itself and combining it with decision-making models might allow for better leak detection, as alarms would need to meet certain criteria before being investigated. Developed separately but related to this proposed idea, Stuart had been working on a visualisation of the acoustic logger data which allows for a visualisation exploration over time. From it you can quite quickly identify a 'baseline' frequency that a majority of the loggers sit at during normal operation. You can also quickly see where alarms were triggered.
The penultimate team also required a bit of help to finish the prototype, so the wonderful Dan from imactivate, and regular ODI Leeds contributor, stepped in. Focused on the human side of understanding water flow, the proposal was to combine the use of acoustic loggers fitted in highly specific locations (on the pipes feeding each household for instance) and asking the household members to keep a diary of their water consumption. With this data, Yorkshire Water could then look for the patterns that might indicate increased water consumption rather than leaks.
Last but not least was Kara from University of Leeds, who was also the last person standing from her original group. On day 1 the group had proposed a Shazam-like app that customers could use to help identify the sounds of leaks, with data collected going to Yorkshire Water to build a database of sounds for analysis. By day 2, Kara instead went down the route of digitally improving methods that already worked - listening sticks. Her physical prototype demonstrated the need for a good quality microphone on the listening end with a smartphone-like device on the other end. A screen would allow you to 'see' the sound, playback, erase, send to cloud, etc.
For more in-depth musings on each of the teams and their ideas, we highly recommend perusing the hackpad document that everyone contributed to. It's a long-read though - over 40 pages!
The final furlong was done, everyone had reached the finish line with no injuries, and Yorkshire Water had a wealth of fresh insight into how their data could be used and improved. They were so impressed with the collective effort and have invited the teams to discuss their ideas further internally for future development.
Before the curtains are even drawn on this event, there is talk of another Yorkshire Water innovation session in July. Keep your eyes peeled and your ears to the ground for the next innovation challenge surrounding bioresources.
