How NXP is Revolutionising Water Quality for Everyone

Ruth: Our topic today is water quality and everything tech can do to help make drinking water safe for everybody. Depending on how you define access and safety, there are 40% of the global population that do not have access to safe drinking water. Could you walk us through what traditional pH monitoring looks like and how an AI-assisted system changes this game?

Fabrizio: Typically, to check if the water can be drunk or not, there are samples of this water that are being analysed by laboratories. And then, of course, there is some time needed when you pick the sample, send it to the lab, and wait for it to be analysed. The analysis is very accurate, but it’s slow, and the response time is slow.

This means that sometimes, by the time the feedback goes to the people who should use the water, it’s too late. And in the meantime, you don’t know—especially if there has been an event like a flood that may have contaminated the water—whether the water is safe to drink. So the time they receive the feedback is long. That’s really a problem for those people.

Ruth: And what led NXP to prioritise water quality monitoring? Was this customer-driven, or how did this come about?

Fabrizio: Measuring the pH is a typical application for electronic systems, like analogue front-end chips and microcontrollers. What is new is that, thanks to artificial intelligence at the edge—when I say "at the edge," I mean locally. Typically, we think about using artificial intelligence in our PCs, which is cloud-driven. So, our inputs go to the cloud, are analysed by data centres, and feedback is provided to us.

Edge AI means that the data is processed and an output is generated right where the input is coming from. This allows for fast feedback from the machine learning algorithms or models to the people using the system. So basically, this system is able to decrease the time it takes to get feedback and provide reliable, continuous data output.

Ruth: For our listeners who aren’t hardware engineers, what exactly is an analogue front end, and why does it work so well for this application?

Fabrizio: Let’s break down the three words. Analogue means processing analogue signals. Typically, our world is full of analogue signals—every system or quantity we observe is analogue. In order to be processed by a PC or artificial intelligence, whatever the application, it should be converted to digital. So all the processing we see is done in the digital domain.

An analogue front end is an analogue product that stays on the front end. It’s the first interface to the sensors. In this case, we are talking about pH. So there are probes that are, of course, in contact with water and provide analogue signals. These analogue signals are very delicate, so the information they provide is easily corrupted by temperature variation, interference, and so on.

So you need a device that is able to convert these delicate signals into robust data. This is done through an analogue-to-digital converter, which converts the analogue information into the digital world. Once the data is digital, it’s very robust. There are also algorithms that ensure this data isn’t corrupted by electromagnetic interference or other factors. But the key role of the analogue front end is to convert small, delicate signals into robust data output.

Ruth: And I think you already mentioned why this needs to happen at the edge. Why not just send it to the cloud for analysis? I think you mentioned speed as one reason, but are there others?

Fabrizio: Yes. Speed is one reason. Another could be that sometimes these water sources are in remote areas where cloud connectivity is not available or is intermittent. Having the information processed locally allows for better resilience and also, as we can see, a kind of equity—so people in areas where networking isn’t easily available can still benefit from on-the-edge artificial intelligence processing.

Ruth: Mm-hmm. Yeah. Terrific. As I imagine, those are actually the communities hit hardest by climate change and its consequences for water quality, right? So that combines really well.

Fabrizio: Yeah. Also, pollution is another element. Typically, polluted elements change the pH, so a pH change can also be a first sign of polluted water. Even if it’s not deterministic, of course, it can at least raise a flag that something needs to be analysed further to understand if something has happened.

Ruth: Okay. Got it. And what does the machine learning model actually learn when monitoring pH levels? You said it’s an indicator that when these levels change, it could be a sign of pollution. Is it spotting patterns, predicting anomalies, or something else?

Fabrizio: A machine learning algorithm is deployed in different phases. In the first phase, you put your system in a known environment and verify that the information is double-checked with manual checks, as before. So you compare the results of your machine learning algorithms to the real data you still measure with traditional systems. This is the learning phase, where you teach the algorithm how to interpret the data. Then, typically, there is a trial where you install the system in several places—not too many, but enough to control the environment—and the algorithm is refined with better data from more sources. It’s a step-by-step process that, in the end, can provide better results. And of course, after many systems are installed in the field, there is continuous improvement, where the algorithms are always refined.

Ruth: What are some use cases? I heard you mention communities that don’t have access to safe drinking water in general, but I think agriculture and municipal water systems also come to mind. Can you share some examples?

Fabrizio: Yeah. For example, in agriculture, pH is important because if the pH is altered, the water going to the plants can also damage them. In industrial treatment, for example, industries need to control the water in their processes. Water is heavily used in manufacturing, so controlling its quality is essential.

Also, municipalities can benefit. Imagine if these systems, with a certain level of connectivity, allow municipalities to provide transparent data to people, showing the pH level and water quality level in real time. So they can also provide a service to the people living in that municipality.

Ruth: We will take a short break. Stay with us, and we will be hearing from our guest very shortly. This podcast is brought to you by Avnet Silica, the engineers of evolution. Subscribe to our Avnet Silica newsletter or connect with us on LinkedIn. If you want to learn more about us, we have put information and links in this episode’s show notes.

What are the biggest technical challenges in ensuring that these electrochemical sensor readings are accurate?

Fabrizio: Yeah, I already mentioned the delicate signals coming from the probes—that’s the first challenge. But this is also the key benefit of our front-end chip. Another challenge is that sometimes the data coming from pH measurements are just siloed—it’s only that data. Having multi-channel information, also from other probes—like temperature or humidity—allows for data aggregation, which can again be processed by machine learning algorithms. This way, the data is more consistent because it takes into account not only a single measurement but multiple measurements. For example, our NXP analogue front end has multi-channel operation with temperature compensation, which provides the foundation for this type of use.

Ruth: Are there any real-world examples you can share, maybe giving some practical insight into what problem this solved and what wasn’t possible before?

Fabrizio: Yeah. As I said, the possibility to collect data in remote areas—especially considering that some of these systems can be battery-powered—allows us to provide this kind of service in places where it wasn’t possible before. That’s the real use case.

Industries can also benefit by having continuous monitoring of the water instead of episodic measurements—whether it’s once a day or once an hour. This allows for automation and provides the possibility for automatic decisions. For example, whether to increase the chemical dosage to correct the pH can be done in real time, avoiding over- or under-dosing. If you take a measurement and the pH is too high, you might overcorrect with pH-minus chemicals. With continuous monitoring, you avoid that, which also means reducing the cost of chemicals.

Ruth: When I think about industrial applications or agricultural fields, it comes to mind that these systems must operate in very harsh environments. How rugged does the hardware need to be?

Fabrizio: Yes. Typically, these systems need to resist surges, for example. This is the most common source of robustness required. So surge protection or electromagnetic interference protection needs to be considered. Of course, our analogue front end has been designed for that, so it can resist harsh environments because it’s built for this type of application.

Ruth: Are there any limitations to what edge AI can achieve in pH monitoring? Or are there cases where human oversight is still essential?

Fabrizio: As we discussed, if the learning phase is done properly—if you teach the system and control the results in the learning phase—the data are pretty reliable. The only thing operators should do manually is maintain the probes. Probes, of course, age and degrade over time, so they need to be changed periodically. The probe itself, which is the source of information, has to be maintained properly to get reliable data.

Ruth: What advice would you give to water authorities or agricultural operators considering AI-assisted monitoring? Where should they start, in your opinion?

Fabrizio: Yeah, that’s part of the step-by-step implementation process. What they can do is test one or two systems first. They can see the data coming out of the electronic system, compare it with traditional analysis, and maybe run these tests for a few months to check if the data is reliable. Then they can start implementing it in their systems. For industries, it’s also very important to have connectivity, as I mentioned before, so they can integrate these electronic systems into their data infrastructure and have all the water information in their dashboards.

Ruth: Does it make sense to combine this information with data from other sensors?

Fabrizio: Yes. For example, temperature is the first thing that comes to mind. But also information from the cloud—like weather data. Humidity can make a difference in measurements, or you can correlate pH variations with weather events, like floods or droughts, which can influence the measurements. So, you can correlate pH data with local events.

Ruth: If you had to put together a soundtrack for this episode, what song would you put on it?

Fabrizio: I was thinking about a song by a band called System of a Down, called Toxicity. It’s about environmental problems—urbanisation puts a lot of stress on water quality for all of us. The song is about the toxicity of our cities, so that’s what came to mind.

Ruth: Oh, that’s a perfect fit. Terrific. Thank you so much. I will put it on our playlist. It is actually quite scary. A couple of years ago, we had a guest on the show talking about smart water metering, and they shared some horrific numbers—big cities, even developed ones like Barcelona, London, or Mexico City, will run out of water in a few years. So this is really important work that you’re doing, and tech is delivering solutions for issues that will affect all of us at some point, right?

Fabrizio: Yeah, sure.

Ruth: Thank you, Fabrizio, for sharing your expertise. It was really interesting to see that edge AI is now moving beyond vision and voice into environmental monitoring that can genuinely improve lives. Thank you so much for being on the show.

Fabrizio: Thank you. Thanks a lot to the people listening to this.

Ruth: Thank you for listening to We Talk IoT. Stay curious and keep innovating. This was Avnet Silica’s We Talk IoT. If you enjoyed this episode, please subscribe and leave a rating. Talk to you soon.