By Clare-Joyce Ngoran, Data Analyst at nLine
I’ve been a Data Analyst with nLine for more than a year, scraping buildings’ data from Google maps to factor into our measurement of the duration (SAIDI) and frequency (SAIFI) of power outages in Accra, Ghana, analyzing the data we have collected in different countries, and creating visualizations that aid in communicating our findings. SAIDI and SAIFI are calculated using power quality data measured with nLine’s plug-in sensor. However, this post isn’t about that work! Rather, in August 2021, I had the opportunity to see, touch, and gain a deeper understanding into the capabilities of our power reliability monitoring strategy. This opportunity, exciting on its own, showed up at a time when COVID-19 restrictions and curfew in Rwanda were at their peak. Hence, this was not only an opportunity to get a deeper understanding of how the data I analyze is collected, but also a means to step outside of my house in Kigali and stretch my legs.
This blog will share some background on why and how we deployed sensors in both on and off-grid electrified health clinics, discuss challenges experienced in the field, and recap lessons learned.
The reliability of energy systems is a major topic of interest, especially in light of increased extreme weather events in recent years. This topic is crucial to explore in Sub-Saharan Africa where even access to electricity remains a myth for a huge percentage of the population and access to incessant energy supply seems to be a far-fetched luxury across the continent. Knowing the importance of steady and sufficient energy supply and the unpleasantness that comes with intermittent energy flow, one can imagine how mind-boggling it is to live with such unsteadiness during a pandemic, in an era where almost everything — every activity — seems to be fueled by electricity.
Access to dependable energy supply is critical for quality health-care and well-being. Unreliable nature of electricity has huge costs for clinics including about 50% loss in vaccine supplies and 70% of electrical medical equipment malfunction. From research conducted in a number of health facilities across 11 countries in Sub-Saharan Africa, researchers found that only 26% of health facilities had access to electricity and only a small percentage (28%) of these powered facilities had access to reliable energy. This presents a fascinating area of study to identify insights that can help detect the root cause of the problem and provide recommendations that can alleviate the situation.
nLine’s power measurement sensor monitors and reports on the power quality of the grid at the individual consumer level, independent of the utility. We then use our backend analysis and algorithms to generate insights on the reliability of the grid, by clustering data reports from more than one sensor. As aforementioned, this is a huge part of my daily tasks at nLine.
Samuel Miles (Sam), a PhD student at UC Berkeley, is working to understand the state of electricity at health clinics in Sub-Saharan Africa. He was interested in deploying a handful of our sensors to gain real-time insights about the reliability of power supply at health clinics in Rwanda, with the longer-term goals of sharing insights with electricity planners to better design infrastructure deployments and with health sector professionals to identify cold chain vulnerabilities. This was exciting for me as a data analyst at nLine residing in Rwanda because I would gain a better understanding of how the data I analyze is collected.
The plan was to install sensors at five health clinics (two sensors per clinic). The power supply at the clinics varied widely, and illustrated the range of solutions clinics turn to in order to meet energy demands. For example, the health clinics where sensors were installed included the following power supply sources:
- OffGridBox only
- OffGridBox, a generator, and (defunct) roof solar panels
- OffGridBox, a generator, and roof solar panels (which “kind of work sometimes”)
- Grid connected, solar panels on the roof, and a generator (as last resort)
- Grid connected only
Figure 2: These photos illustrate the mix of power solutions used to meet energy demands at a single health clinic, a common occurrence. From top right (clockwise): A generator, an OffGridBox, a utility pole reaching just up to the premises but not yet connected to the clinic building, and solar panels on the roof. (Photo Credit: Sam Miles)
I traveled with Sam to our first of three clinics we visited together. When our point of contact at the clinic arrived, we first had a briefing session where we introduced ourselves, explained the purpose of our visit to the clinic and the research study, and described how the sensor operates and the type of data it collects.
Installing two sensors was a trouble-free, straightforward procedure that lasted 5 minutes. In these 5 minutes, we did the following:
- Identify all power outlets in rooms where the clinic tells us a sensor may be plugged in. Since we were plugging in two sensors, we chose two available rooms with the best cellular service and then chose the best available outlet location. An available outlet in a seldomly accessed or least visited corner of the room is recommended. This can help prevent the sensor from being knocked out of the socket or unplugged.
- Next, connect a power extension cord to the outlet to ensure that the clinic is not inconvenienced by the presence of the sensor and outlet plugs are still available for other devices.
- Plug the sensor into the extension cord.
- Monitor the behavior of the sensor (via the light indicators) and ensure that the sensor is capable of connecting to a cellular network in the area. This cellular connectivity allows the sensor to transmit collected data to nLine’s database.
- Complete a brief survey with the clinic’s point of contact.
My main responsibilities during this process involved recording survey responses from the clinic point of contact, troubleshooting sensors as needed, and ensuring that data was being transmitted from the sensors into nLine’s data visualization system. (The extent of sensor troubleshooting involved monitoring the light indicators and trying different power outlets in the room if the original outlet was not suitable for connectivity or was not functioning.)
Six days later we visited the furthest and most remote two of the five clinics. There was a slight difference in the deployment strategy at these clinics which had an OffGridBox at the location. Instead of both sensors being installed within the clinic’s building, one sensor was installed in the building and the other in the OffGridBox outside the clinic building (this was the sensor deployment strategy at all four clinics with an OffGridBox on the premise). We followed the same procedure of meeting the point of contact at the clinic, explaining the reason for our visit and purpose of the research study, describing how the sensor functions, and then installing and observing the sensor for network connectivity to ensure we were getting real-time data.
- Locating remote clinics: Our first installation was a straightforward drive and the clinic was by the main road. Hence, it was not a challenge to locate the premises. During the second trip, however, the clinics we visited were far off the main roads, which were untarred, and the winding mountainous drives were exhausting (I was nauseous for most of the trip). We were following Google Maps which oftentimes led us to unmotorable areas and we were forced to re-route. Our drive from the second clinic to the third took about 2 hours going through some very difficult paths. And what’s worse? Upon arriving at the clinic, we realized we had passed it by when we were locating our first clinic and the drive that lasted 2 hours could have been 15 minutes!
- Collecting accurate installation GPS locations: To complete the sensor installation process, the coordinates of the location of the installed sensors are required. A good amount of time was spent trying to capture accurate coordinates but in most cases, we ended up with coordinates that were up to 15m accurate (up to 3m is a suitable margin). This is a challenge because, should one need to relocate this sensor in the future, the navigation will indicate that they have arrived at the destination when they enter a 15m radius around the sensor. This location vagueness can make timely sensor pickup or troubleshooting challenging.
- Relativity of scale of measurement: One of the questions in our survey is “how would you describe the power situation in this clinic?” Possible response options are “good”, “okay” and “bad”. One participant chose the option “good, which seemed to contradict what they had earlier said in a conversation when they mentioned experiencing blackouts about eight times a day. I would personally see such a situation as “bad”. This interaction shed light on the relative difference in an individual’s sense of what constitutes “good” or “bad” power supply.
The importance of measuring power quality and reliability cannot be understated. Health care has evolved so much over the years such that most quality medical facilities today require a stable and reliable power source for optimal operation.
As mentioned, one major reason why I accompanied Sam for this pilot sensor deployment was to get a better understanding of how the data I analyze is collected. I come from Cameroon, where frequent blackouts have been seen as normal for too many years. Meeting the clinic contact who described her power situation as “good”, even though power supply goes off about eight times a day, aroused sympathetic feelings and reminded me of where I come from, the reason why I decided to merge two fields of studies (energy systems and applied machine learning) during my Masters, and why I joined nLine.
I was reminded of the huge importance of the work we do at nLine. I believe we will go a long way in accomplishing our mission of measuring and improving the performance of the dire energy infrastructure situation in Sub-Saharan Africa. As they say, “if it cannot be measured, it cannot not be improved.”