 All right, I think we're about to start. Thank you, everybody, and for attending the first of what's going to be the series of webinars on advanced control systems for nitrogen removal. This one is in full scale with with water facilities. And as you can see it's hosted by the International Water Association who kindly is helping us organize these series of webinars. The webinar in particular has been organized by the members of the specialist group instrumentalistic and control automation group that we will be introducing shortly. So this webinar will be recorded and it will be made available on the month on the IWA connect plus platform, and that will include presentation slides and any other information that we produce the speakers. We are responsible for securing copyright permissions for any work that we present and that of which we are not the legal copyright holder, and any opinions hypothesis conclusions, recommendations containing the presentations. They are the sole responsibility of the speakers so please bear in mind this does not reflect the, the opinion of the IWA. If you have any questions, please please use the Q&A box to send questions to the panelists. There's two moderators, me and Alexandria, and we will basically make sure that your question is redirected to write the speaker and at the right time. It's only for general requests and interactive activities. So please do not post your questions there. Questions for speakers only in the Q&A box. And agenda for today's webinar. Initially we'll have a welcome note from the instrumentation automation group. First, the chair of the group. Continuation, general opening and introduction where we'll explain the webinar and provide some initial concepts to help people who is on the early stages of the journey through ICA. How to help them follow the presentations, a continuation. And then we'll have the speakers, Jeff Sparks, Victoria Rano, and Stephanie Kloss. After all the presentations have finished, there will be an interactive panel discussion for all the speakers and participants. And then we'll finally close this webinar. So you're a bit more familiar with all of us, although I will prepare introductions shortly before each one of the talks. Go Jeff Sparks, speak Rano, Stephanie Kloss, the three speakers, Genelci, the chair of the ICA group that will be presenting the group that is behind this webinar, myself and Alexandria who will be moderating the Q&A session. Right, so Genelci, feel free to start when you're ready. Perfect. Thank you, Paul. So welcome everyone to this interesting webinar, advanced control for nitrogen removal. And as Paul mentioned before, these women are having organized by the instrumentation control and automation, especially group, better known as ICA. And so my name is Genelci Alferes. I'm working at Vito. Vito is a research center located in Belgium, working mostly on topics related with monitoring technologies and also digital water. Yes, so I wanted to take the opportunity before the webinar start to tell you a few words about the group and our make activities. So all our activities are really based on three main pillars. So we have first knowledge man, we try to promote an international discussion forum for to collect exchange and share information on methodologies experts experience in all aspects related to ICA for a wide amount of water system applications. Second, we work towards disseminations. So we would promote the collection also the summarize and publish all about practical experience and related with the application of ICA in practice. And the third one is towards applications taken into account aspects, not only technical but also socio economical sustainability aspect of ICA really towards the implementation of practical solutions. Next slide. And so our main activities are short summary. And we try to be active, releasing relevant information related to the ICA community, towards the IWA Connect channel and also our own social media channels, we release regularly group newsletter that are also available to the to the IWA Connect page. We organize a support conference and workshop. We support as well as group working group and cluster, working in a specific topic. We organize a webinar like this one, for example, we try and we encourage the publication of ICA related topics in different conference but also in scientific journals. We try to also partnership with the other industry organizations are working on similar topics, for example, the smart water network forum. One before. We work with a management committee that at the moment have 10 members, including young water professionals, what quiet spread geographical distributions, and you can find all the information of our group in the IWA Connect plus websites. As soon as you become a member you you have access to several type of information, including our newsletter, where we try to exchange information about, for example, new projects, activities, and different initiatives. So upcoming events, more research topics like PhD thesis, etc. news from the IWA headquarters, and other type of information that is really relevant for the ICA community. Talking about the upcoming events, our main event is really the IWA conference on instrumentation control automation last one was in 2022. This is in 2025 in Oslo, Norway is still to be organized. And in a more shorter term, we have a set of webinar that are currently in preparations. We have for example one on remote and remote and remote control, following the one that we are going to hear today. And we have also another series of webinars in preparation related to into measurement control and mitigation strategies. So it's something important we are currently opening new positions for members of our management committee. So we will release the call in the coming days. So if you want to also know more about it, and all about all the information that I mentioned before, just follow us and join our group in the IWA Connect plus channel, and also you can follow our information in our LinkedIn page. I conclude my short introduction, I give the hand to Bo, Juan Garcia, and also Alexandria Gagnon. They will guide you through the webinar. We have really highly qualified speakers today. So I thank you very much and I hope you all enjoy the webinar. Thank you very much. Okay, I'm getting the hang of this. If I use the keyboard, it doesn't get the same issues. I'll do that. Apologies for the technical issues, everybody. Right, so thanks for the introduction today. I see a group channel, see it's truly busy and very interesting, all the things that we are recognizing. Now I'll proceed to introduce the webinar. First, a little bit about me. So I have a PhD in what the science technology that I worked on evaluating the resilience of which were the treatment plans for which I had to take a look at the control systems and finding out how to make them more robust and resilient against different shocks such as power outages. In the moment I'm working more on the water quality modeling field but ICA is always an important part of my work as you are any models on this code as the instrumentation, the data that you collect throughout the instrumentation. My colleague, Alexandria Ali Gagnon. So Ali is a treatment process engineer Hampton Roads Sanitation District. She's currently pursuing her PhD in environmental engineering from Virginia Tech. And so far his research is an optimizing and stabilizing existing control systems using traditional control engineering and data-driven methods. Right, so as I said at the beginning we wanted to make sure that everybody could follow the webinar and I'm guessing that will have people from many different backgrounds. So we're going to provide a very short introduction from the very beginning. Starting with the activated sludge process. So this is probably you will know a widely used biological treatment method in which water treatment plants and issues to remove organic pollutants and nutrients from wastewater. This process involves the growth of microorganisms, hence the name activated sludge, ineration tanks, and these microorganisms will consume and break down organic matter in present in the wastewater. It's a process that was invented over 100 years ago, and although it has been perfected and studied for over 100 years, it has surprisingly not changed as much as you might think. The microorganisms will metabolize the organic pollutants, and as they do that they also form flux that settle in what we know as the secondary clarifier. Now what is nitrification and denitrification? Nitrification is the biological conversion of ammonia to nitrate, and this is a two-step aerobic process. The nitrifying bacteria oxidize ammonia first to nitrate and then further to nitrate in the presence of oxygen, that's the key. And to promote nitrification, aeration to make sure that the oxygen is replenished is essential. So we need to maintain a sweet spot of dissolved oxygen levels in the aeration tank. On the other hand, denitrification is the biological conversion of nitrate to nitrogen gas. This only happens under an oxygen condition, so low oxygen conditions. The denitrifying bacteria, different group of bacteria, will use the nitrogen from the nitrate molecules rather than the dissolved oxygen, and will convert this nitrate into gas that is then released into the atmosphere. So aeration control, as you can imagine, is critical to maintain those control zones of low oxygen and high oxygen in specific parts of the treatment process. So now we've gone from tibetase slats into nitrification to denitrification and now into aeration control. Aeration control consists of managing the dissolved oxygen levels in the different aeration tanks to create favorable conditions for both nitrification and denitrification processes. So to achieve this, we will employ very advanced processes and control strategies and also equipment such as the soft oxygen probes or other kind of probes and online monitoring systems. So limiting aeration, although we are we're focusing so far on the nitrification, the nitrification needs. It's also very important to reduce the energy consumption and also to improve the biop removal, the removal of orthophosphate from the wastewater, as well as making your process more robust and reducing the extent of ammonia peaks in your effluent. And energy is very important. So this, especially the more we take into account climate change and carbon emissions, reducing the energy consumption, it's always a priority for any industrial process. And that is also the case as we all know for wastewater treatment. As you can see in the picture on the right slide. The picture that I have seen repeatedly through my career aeration is considered to be by far the largest consumer of energy in the wastewater treatment process and therefore that makes it usually the target of any strategies to reduce the consumption of energy. We want to have a perfect control. What are the challenges to controlling these dissolved oxygen in the aeration tanks. So for starters, we have a high variability of incoming load and temperature. So the load coming into the wastewater treatment plant will vary, will oscillate from day to night, and sometimes it will have different patterns. We work with fixed reactor volumes, which is often not essential. We also will have to work with wastewater treatment plants that might have been designed for peak load, making that inefficient if we need to operate the plant at any other range of conditions as it is usually the case. And of course, in order to have control, we'll need additional infrastructure, we'll need an implementation methodology and we need regular maintenance. So there's lots of challenges to an efficient control system. However, we have obvious opportunities. We can make use on top of all the ones that we have already mentioned, we could make use of all the available capacity of the plant. And we can increase the efficiency at the robustness of the treatment process. I wanted to touch on very specific process requirements that we asked from control for nitrification and demitrification. So sufficient provision of dissolved oxygen to make sure that we have enough nitrifiers in our tanks. Constant we'll have to see how we need to work in order to make sure that those nitrifiers have a constant supply those bacteria have a constant supply of substrate ammonia and all the essential nutrients and carbon. We'll have to make sure that the aerobic sludge retention time is high enough in order to keep a constant mass, especially of nitrifiers but also for all the other bacteria is needed. And as we will see, this can get really complex. So in this webinar, now that you have an overview of what the activated sludge nitrification denitrification and control and the challenges behind it are. We're going to see, we're going to explore the potential of different approaches for optimizing nitrogen removal in these activated sludge systems, but the full scale with water treatment plants taking a look at real examples. We'll explore the benefits and the limitations of using different type of sensors such as pH and ORP sensors, ion selective ammonia sensors, nitrate sensors, and also different control structures, different reactor configurations will also be discussed. In particular, the first presentation will be focusing on ammonia based aeration control and using machine learning to implement a predictive controller to optimize this type of process. The second one will be focusing on successful experiences of full scale implementation of a slightly different type of control pH or P base control for optimizing the biological nitrogen removal. And finally, the third presentation will look at how to carry out control on mainstream animal processes. So without further ado, I will pass the work with to my colleague Jeff Sparks Jeff has been working in the wastewater industry in consulting utility management and as a process engineer for the last 15 years. He's currently a process engineer with Hampton Roads Sanitation District as well in Eastern Virginia US, and he's also registered professional engineer in the state of Virginia. I'm going to be discussing with you advancing ammonia based aeration control using particular modeling my advisors before I started here Peter Van Bollaham. I should also mention a part of this work. Okay, so as I mentioned, I'm a process engineer with Hampton Roads Sanitation District. And we are a regional wastewater utility district. HRC facility we're going to be talking about today is the Nansman treatment plant. And that's located in Suffolk region. Nansman treatment plants a five stage barcode process and you can see the Nansman treatment plant depicted here. So, within the biological process, our focus is on the aerobic zone right there in the middle. We're going to be focusing on ammonia control in that in that zone. The specific objective here is to and I'm just going to read this for a point out for you is to use a hybrid model feed forward approach to control. So yeah, the specific objective here is to use a hybrid bottle feed forward approach to control the aerobic volume and the asset point, keep positive ammonia at the end of the aeration tank and stay below the max total organic nitrogen concentration of five milligrams pleater in the secondary effluent at all times. And the reason we have to do that is because we have this requirement due to an indirect portable reuse facility on the back end of this plant. So this graphic here shows what the existing a back controller looks like at the Nansman treatment plant. So here we are, we are only showing one tank, and we are only showing the aerobic and second stage anoxic zones from the five stage barcode so flow is moving from left to right, which you can see there with that arrow and within the aerobic zone we have to deal control zones, and there are zones two and three zone two is about two times larger than zone three. And each zone has a dedicated deal probe and aeration control valve. There's an ammonia measurement that you can see there off to the right. And that's made with a wet chemical analyzer and we use a feedback PID controller to set the deal step point with the objective of keeping the ammonia constant at the end of the tank. And that point that's generated currently from this feedback PID controller is currently applied the same to both zones two and three so this is the existing feedback a back controller. We also have an influence ammonia measurement coming into the into the aeration tanks using an ions elective electrode sensor so this is the blue line that you see in this trend here off to the left. And in this trend you can clearly see the diurnal variation in the in the aeration influence ammonia concentration and in this particular trend, it looks like the ammonia is varying somewhere between seven and 11. And one important thing to note is that the flow and the concentration seem to move together here and then that can result in some pretty significant diurnal variations in ammonia load. So then we also have these disturbances in that concentration, and that causes the trend to sort of deviate from the night the normal diurnal pattern you can see that highlighted there with that circle. Now these could be due to century or wet weather industrial load, and given the long narrow sort of flood flow ish tanks. This becomes a challenging application for feedback only controller so the performance. You can see here of the existing feedback only controller it's up on the top right with the red orange and green lines. And again these are the aeration effluent ammonia from the wet chemical analyzers. Okay, so what I'm showing here is the the proposal for the upgraded a back controllers so the proposal here is to use that aeration influence measurement and a hybrid model to feed forward the do set point. And the hybrid model would also inform the aerobic volume. And that basically what that means is, is whether zone three gets aerated or only mixed. One important point here is that the feedback controller will remain to correct the feed forward output for when the feed for controllers not getting those do set points. Exactly, exactly right. So this architecture that you can see here on this slide is what allows us to build a hybrid model and and have it making forecasts for us on on live data so solely for the for this purpose of deploying this controller, we installed this separate server at the dance retreat and play and you can see that highlighted in red here on this slide and that connects to the distributed control system that's represented by that symbol in the top left corner. And the DCS passes data across this communications platform called OPC to this to this server, where we have two software packages installed one's called CCI cake, this is the one where we do our, our data driven one's called sumo by dynamite and that's where we do our our mechanistic modeling. And if you want to see what the nansman treatment plant looks like in sumo, you can see it off to the left there coming out of the sumo block. And I should mention what we do is we combine the outputs from these two software packages on the separate server in the DCS and that's what gives us our hybrid model. So now we're no longer on the separate server so now we're in the DCS and when the outputs get written from the separate server back to the DCS they get inserted into this into this formula here and this formula solves for the deal set point based on the one you load to be removed it comes from a 2003 publication which you can see here by by Brecco called feed forward feedback control of an activated sludge process and simulation site. So to understand this formula and where all the inputs come from we will go variable by variable here and also color by color. So first, let's start with the the blue various. These are either known or measured so the flow rate is measured that's q with a meter the influence ammonia is measured with the IC sensor I've already talked about that. And the volume is known from the record drawings when we perform this calculation in the DCS we actually perform it twice. Once with zone two volume aerated and once with zone two and zone three volumes both aerated and that helps us determine what the total aerobic volume should be. All right, so let's move on to the green variable that's the ammonia set point which is currently set at 1.5 milligrams per liter the yellow variables they come from an optimization that we do and Python and those are the nitrified kinetics based on historical ammonia removals and operating conditions using the mechanistic model on the modeling server. The red variable here is the modification that we've made to the Brecco formula and it's really what makes this a hybrid model and that's the mechanistic model error. So if we use historical mechanistic model errors what we're able to do is then forecast what the future ones should be and get a more accurate estimation of the needed do set point. So this comes from a univariate time series data driven model and this again is really what makes this a hybrid model, and we use for data driven models that all run in parallel and we choose the best one and those four data driven models are listed there linear regression triple moving extra boost in an LSTM neural network. The other one variables that you see here that I'm not going to focus on too much today just come strictly from the sumo model on the modeling server. So remember those those nitrified kinetics that I was talking about that we get through an optimization. That's what we're going to talk through here on this slide so the idea here is that we get the mechanistic model as accurate as we can, so that the data driven model doesn't have to work as hard to forecast the mechanistic model error. So every 24 hours, we use the historical TK ends coming into the plant temperatures apply DOs waste rates and observed ammonia removal to calibrate the mechanistic model and sort of back into these these nitrified connects we use the last seven days of data and average nitrification performance across the takes and service. We also take two approaches one uses a Nelver need optimistic optimization algorithm the other one uses randomly selected points in a defined sample space. So these charts here show the results from the two approaches Nelver need on the left and the random approach they are on the right so on the x axis we have new the maximum specific rate on the y axis we have KO, and then on the z axis we have the means for error. So in this case the error is based on the difference between the bottom and removal and the observed ammonia removal and the points on the charts. Here are the minimum errors from the two approaches and their color coded reds Nelver needs blues the random approach and then the black point that you see there is using default parameters in sumo. So we can see that on the slide the results are comparable across another meeting random approaches and their errors are really not that different from using the default parameters. And I think that that suggests that maybe it's these these parameters could be hard to identify especially this time year when we have too much SRT temperatures are high and do is are just slow all the time so we don't see a big variation in DO. And so we're sort of just saying here, yes, new is high. And so, yeah, we're getting pretty pretty similar results with those, those three approaches there. All right, so here are the results from those data driven models that I was talking about that we're using to forecast the mechanistic model error. So we have drive on the x axis there, and we have milligrams per liter of nitrogen on the y axis and then the four data driven models that we're experimenting with listed there on the right linear regression triple exponential smooth and she boosts in the LST and neural network. The dark black lines that you see moving horizontally are the model forecasts than the light gray lines are the observed data so the data that you see here is actually test set data. So it's unseen data to the models during model training. And at each one of these solid black vertical lines. This is where we're making a forecast, what's going to happen with that mechanistic model error over the next four hours. The model actually makes a forecast every 20 minutes, even though those black lines are separated by, by four hour time gaps, but for ease of viewing, I'm just showing one forecast and then what the forecast was four hours later so in reality the model results are really much better than this, but if you try to show the model forecasts every 20 minutes the plot just gets way too messy. What having the four hour prediction horizon allows us to do is ultimately ultimately implement a model predictive controller however in this case, we're not talking about MPC. We're talking about a much simpler feed for controller so for this controller, we make our forecast, and then we use the error one HRT into the future and we insert that into our formula for our calculation to be a separate. Now if we look at the results from the data driven models, what we see is that the LSTM neural network is the crew winner here, but it's also the most complex model we're using. And so by the way, the models here are listed in order of complexity linear regression being the simplest model and then LSTM of course like I said being the most complex. So I know that this slide looks very similar to the last slide and what I wanted to mention about this is that we can see the results of the models here 24 hours later. So the way this works is we retrain the models every 24 hours and we are always using yesterday's data as our test set. So in this case, the LSTM neural network is still the best performing data driven model forecasting mechanism model error. But one of the things that could have led to a deterioration in model performance because you can see that error went up is we have a new test set now, and also the model was retrained. Still LSTM is the winner but maybe there's an argument to be made here especially with this with this particular model that that maybe we could see we could achieve the same objectives with a little less complexity here than using the LSTM. Okay, so really I think the conclusions here are that the iteration takes having a plug flow orientation long and narrow with relatively long detention times may have a residence time distribution that does not favor feedback only a back and so I think that this situation that we're in here at NASM treatment plant and nitrifier kinetics can be reasonably estimated at similar results achieved using random search space and an LREMEET optimization algorithm and we can find those nitrifier kinetics live and real time, but perhaps they're not identifiable if the DO is not is not varying significantly when we do those optimizations. And also finally that the mechanistic model errors can be reasonably forecasted using univariate time series algorithms and machine learning. But the choice of that algorithm may depend on, you know how complex you want to be with your model how accurate do you need to be. And then of course what's with the composition is of the test set data. And that's all I have for you today and Thanks, yes. That was great. The first thing I've seen that we got up to, I think there's nine questions. So, thank you, please keep them coming. We'll try to reply. So Jeff, I think you might be able to take a look now and feel free to type a reply directly, if you can. And otherwise there will be some of them that probably it's worth answering live. All right, before we start the next presentation I just would like to ask you all when you ask a question if you can type initially the name of the presenter. colon and then the question that way will make it easier for the moderators for us to know to whom that question is addressed. And yeah, we'll make sure the question arrives to the right person. Great. So on to our next presentation, false skill implementation of pH or B base control. This presentation will be by Bikki Rwano system professor. Bikki. Well she's an assistant professor in the chemical engineering department at the University of Valencia. She's a member of color research group, which is formed by professors and researchers from the University of Valencia and also a polytechnic department at the University of Valencia. He has a PhD and a postdoc instrumentation coder and automation with what the treatment plans. She has worked with a quality on control system implementation and full scale. And to date, her research mostly focuses on modeling and control of conventional but also emerging technologies specialized on sewage treatment. Whenever you're ready, feel free to start. Thanks pal for the nice introduction and good afternoon for all the attendees. It's a pleasure to be here and give this short presentation. So, let's see if I can control the presentation. Okay, yes. So, I'm going at sorry. I'm going to show you our experience as research group on full scale implementation of pH and or B base control systems for optimizing biological nitrogen removal. So, mainly up to date, most of these control systems are based on online nitrogen sensors that is ammonia nitrate, and few applications are based on pH and or B sensors that are mostly applied to sequences batch reactor configurations, or intermittently rated systems. And indeed there are scarce applications for activated systems that are rated continuously. Based on the know how acquired by the research group. After intensive full scale process monitoring of several with water treatment plans with different treatment schemes, and also with different operating conditions. We develop we developed an advanced biological nitrogen removal control system based on pH and or B sensors that is based on fuzzy logic and a set of knowledge based rules. And basically consists of three controllers, the notification controller, which modifies the diesel Foxy and set point in the different kind of examples that are in the biological reactor. In order to maximize the notification process, based on the pH profile along the biological reactor, these diesel Foxy and set points are regulated by a lower layer controller. The second controller is the simultaneous notification the notification controller, which modifies the diesel Foxy and set point in a faculty chamber with and within low values of this diesel Foxy and more or less between 0.2 and 0.5 pps of diesel Foxy and to favor simultaneous notification and identification process, whenever the notification process is not limited. This is made based on the information provided by pH and or B sensors located at the beginning of biological reactor. We have the denitrification controller, which regulates the nitric acid flow rate to maximize the denitrification process, based on pH and or B sensors located in the anoxic chamber that gives information about this denitrification degree in this chamber. All these control actions for these controllers are limited or regulated by a set of knowledge based rules. In order to obtain a trade off between complying with the effort criteria and also reducing the energy demand and also to take into account system boundaries operators limits and so on. Afterwards, we validate this philosophy this control system, statistically, by executing PCA and parcellist squares model validation, which confirm the relationships between this pH and RP and nitrogen based sensors. Particularly, the profile of pH located at the very beginning of the biological reactor predict the nitrogen loading rate dynamics entered into the biological process. The RP the tendency of the RP also located at the very beginning of the biological reactor predict the organic loading rate dynamics into the process. The combination between the profiles of the pH located at the very beginning of the biological reactor and at the end also predict the denitrification degree of the process. And the pH and RP located at the end of the anoxic chamber represented the denitrification degree and the nitric concentration perspective. So, just as an example of the implementation of this advanced control system. Here we have a full scale facility whose biological reactor is based on an A2O scheme. So here we have an anaerobic, anoxic, facultative and two aerobic chambers. The denitrification controller implemented, regulated the dissolved oxygen set point of these two aerobic chambers based on the information provided by the pH located in the anaerobic and in the last aerobic chamber. So here for instance, we have the control system performance for two periods, more or less winter season and summer season with different temperatures operating temperatures. And what we have here for instance is the input to the control systems that are in orange the pH at the beginning of the biological reactor and on purple the pH at the end of the biological reactor. And as output the dissolved oxygen set point of the two aerobic chambers here in black. Also here is plotted the ammonium in front in green and ammonium effron concentration in blue. So here for instance, during this last period in this figure, what we can see is comparing both pH profiles, a high notification degree, that is also validated due to the significant peak of ammonium entering the biological reactor. And here the control system sets the maximum dissolved oxygen set points in these aerobic chambers. Conversely, here for instance in this period, what we have comparing the both pH profiles, these and these, we have a low notification degree, also validated because the decreasing trend of ammonium entering into the biological reactor. And here the control system sets the minimum dissolved oxygen set point for both aerobic chambers. Then the second controller that was implemented was the simultaneous notification and the notification controller, which manipulates the dissolved oxygen set point in the faculty chamber, based on the information provided by the or P and pH sensors located in the an aerobic chamber. Here we can see an example of the control system performance both for winter and summer periods. And what we can see here is the input to the control system, the or P and yellow. And in red we have the pH profile. And as output, what we have is the dissolved oxygen set point in purple. And here what we also have is the ammonium entering to the anoxic chamber to validate the control system performance. So what we can see here more or less is that the or P and pH profiles give us information about the nitrogen and organic login entering to the system. So when this load is decreasing, the dissolved oxygen set point can be increased. But conversely, it can be seen here during summer period that most of the time the dissolved oxygen set point is set at its minimum value. And in winter season it is increased because the load entered into the system is higher and the notification process is limited so it needs more dissolved oxygen in this faculty chamber. And lastly, we have the denitrification control implemented in the in this facility. This controller regulates the nitrate nitrate recycling flow rate to the beginning of the anoxic chamber, based on the information provided by the or P and P sensors located at the anoxic chamber. Similarly, as the other controllers we have here an example of the control system performance for winter and summer seasons. What we have here is the or P and in yellow and in green, the pH. And here we have the output of the control system that in this case are the frequency of the two available recycling pumps that are in the facility in red and blue. And also we have in red, the nitrate concentration at the end of the anoxic chamber to validate the control system performance. And what we have is that the decreasing trend on on our P, or also a low values of our P at the note low nitrate and nitrate concentration in the in the anoxic chamber, and also there is an increasing 2088 source higher potential of the anoxic chamber. So in this case, the nitrate recycling flow rate increases by the controller. Conversely, when we have a higher or P values or increasing trend in our P values as in this period, and also decreasing trending P8 showing a lower denitrification potential also validated here by the significant nitrate peak that we have here. The control system set at its minimum value the nitrate cell flow rate. So to show results about the implementation of this advanced control system in several water treatment plants. Here we have a successful results before and after control system implementation in three full scale facilities. So to sum up more or less in terms of effluent quality, we have total nitrogen concentration that was reduced between 10 and 30% in average. And in terms of energy savings between 14 and 33 in terms of kilowatts per hour and per kilogram equivalent COG removed, taking into account also the nitrified ammonium. So to conclude, as the home messages, this control system showed great potential to comply with effluent criteria with the reduced operating cost. And to highlight for this control strategy the importance of knowledge based rules, because with them, you can get more feasibility in order to adapt to different which water treatment plant layout to different treatment and to obtain a trade off between energy reduction and effluent criteria compliant. And also to remark that this also really need to implement online for detection methods to guarantee proper control system performance, not only for this control system, but for all the control systems that depends on data. So with that, I conclude my presentation. Thanks. That was really interesting. Great. I can see you've got a few questions already waiting for you. Great. So I hope everybody enjoyed that. And now on to our last presentation for today, Stephanie Claus process control for mainstream and amongst Stephanie, she's a process treatment process engineer have to have done road sanitation district as well. She's a PhD and masters, both from Virginia Tech in civil engineering. And she has researched the barrier topics. She has a very broad interest but in particular, all of them pertain to size stream and mainstream shortcut nitrogen removal. Hence, yeah, she's an expert on this particular topic. Thanks Stephanie, and whenever you're ready, please go ahead. Thank you. So how did a good job introducing the background on the nitrogen cycle and the introduction. And so in my presentation, I'm going to be talking about I'm shortcutting the that cycle using animox. So the acronyms here I'll explain on the next slide but we're using both the partial denitrification animox route and the partial nitritation animox route at HR ST. And then about a decade of research of applying animox to the mainstream that we're now doing full scale. So we have two implementations currently of mainstream animox at York River and James River. Because it's been so successful and we've had so much savings, chemical, mostly chemical savings and capacity savings from this process. We're looking at implementing it at a few of our other plants and here's our region. This is the Chesapeake Bay here. This is the, the coast of Virginia. You know how we arrived at doing partial denitrification animox instead of partial nitritation animox is on the left. This is the, the PNA route. This is what we, this is what we typically use in sidestream processes and this was the topic of our research for many years in mainstream and it seems more obvious to just stop at nitrite to get the required stoichiometry for animox which requires approximately equal amounts of ammonia nitrite. But what we found through a decade of research was it's actually easier to get the nitrite by using an external carbon source and going from nitrate to nitrite. Then stopping halfway on nitrification and surprisingly it also has similar theoretical cost savings because you're only sending 50% of the ammonia through this route. You know, the 50% that you're getting, ammonia removal you're getting through animox is a huge savings no matter what it's just how you get, how you get that nitrite. So, for the implementation for process control, you can, we can either look at implementing in an IFAS process or a hybrid granular process in the mainstream. So animox are slower growing so they need a separate SRT. So this mostly for us means biofilm processes but you could also potentially do it by bio augmenting and retaining granules using a screen or hydro cyclone. But so this would be animox on the media and then integrated into the mix liquor you have the partial denitrification happening. You can also have it in a tertiary process after secondary clarifier, such as a deep bed filter and MBBR fluidized bed. Either way you need to control the ammonia to knock ratio. So, this has to do with the animox stoichiometry walks, which I'll explain in the next couple of slides. So, whether you implement it in integrated in the mix liquor or in a polishing process, you need to now control that ratio, not just control the ammonia coming there. And then we also need a carbon dosing controller to control the external carbon source. We have not. It's very readily possible to use incoming carbon to do partial denitrification. It happens very readily using methanol glycerol acetate. And so we have not researched that yet using the influence COD for partial denitrification. So this is just Jeff slide with a little bit added to it just to say that, you know, it's still the same concept as a ammonia based aeration control. It just requires one extra sensor so you have to measure, you have to measure knocks, and we'll talk about to if you need how much you didn't need to differentiate between nitrate and nitrite how important that is. So this can be accomplished through PID feedback, some sort of feed forward feedback controller. Or the work that Jeff is doing can eventually move to this direction where we do AVN control instead of ammonia based aeration control so it's just controlling this ratio ammonia over knocks. The, the results of our research have shown that you know we worked so hard on doing and to be out selection to do the partial nitritation route, and doing the partial denitrification route to implement mainstream animox was surprisingly easy from the biology standpoint. Now, the biggest challenge that we have is really in the controls to meet these low affluent TIN values. So, the required nitrite to ammonia by and for animox stoichiometry is about 1.3 and animox also produced a little bit of nitrate. So if you just subtract these two numbers, it gives you sort of theoretically, because it so the nitrate now gets fed back into the process this is actually a benefit of the PD&A route. By the way over the P&A route is, you know, you don't have to deal with that nitrate production goes back into the process. So that's like the theoretical approximate stoichiometry now it's actually much more complicated than that and I slide on that but what happens if you have to if your nitrate is much higher than your ammonia coming in, then basically you're doing more full denite you're not stopping at nitrate. And you can also have nitrite breakthrough if you're doing partial denitrification but the ammonia is not there for animox to take it in the correct stoichiometry then you can have nitrate coming out. If your nitrate is much greater than your nitrate, then you'll have ammonia breakthrough. So we typically operate conservatively more in the first case because then you're just wasting, you know, additional carbon, which is safer when it comes to your permit then having ammonia breakthrough. So this is the slide I was saying that just a little bit more about what the target AVN is. It's actually much more complicated than just the theoretical stoichiometry because it depends on the TIN coming in, what your effluent goals are, the partial denitrification percentage which is how much of the nitrate stops at nitrite compared to going the full way to nitrogen gas through heterotrophs. And then also the composition coming in how much is nitrite and how much is nitrate. So this also speaks to how, you know, we're not abandoning the partial nitritation route entirely. This controller actually the controls and everything works the same no matter how much NOB out selection you get. So the benefit whenever you have more nitrite coming into the process, coming into the Anamox zone, that means you don't need to add external carbon to do the PDN route, the ammonia nitrite are already there for Anamox to use in the PNA route. So yeah, this is just some examples of how we as when we're operating, how we think about it this could be automated. But as far as changing the AVN step point, but right now we're just we just do this manually based on performance. So that was the aeration controller. The carbon source part what we found from our research that's pretty interesting is that when you're using these external carbon sources like glycerol, methanol acetate, that we don't need to have no special controls that need to happen except that you need to maintain a nitrate ratio. I mean, sorry, a nitrate residual. So that's what this is showing here in a batch that when whenever you see the nitrate in this example getting below two. That's when you see the switch where nitrite stops accumulating and starts getting used. So we want we want to be here where it's accumulating on the left side so that Anamox can use it. What we find is we need to maintain, depending on the process and nitrate residual in the range of point five to two to keep the PDN percentage high. Again, the amount of nitrate that's stopping at nitrate. So to do this, we can just have a simple feedback carbon dosing controller which is what we have at James River. So that's just feeding the carbon based on a nitrate set point of around one. Or at York River, we use a feed forward feedback controller just to be more precise. So this is feeding forward the, the, the stoichiometry for the carbon source so like for methanol that we feed forward what we think the approximate stoichiometry we're going to need based on an Anamox measurement. And then there's a feedback trim based on the affluent knocks. So those, those are the controls. Of course this all requires reliable sensor measurements. So we need to have a good ammonia measurement even at low concentrations and depending on the implementation discrimination of nitrate and nitrite without interferences can also be very important. It depends. For example at York River coming into our PDNA denitrification filters. We can have a lot of nitrite at certain times. So then it can be important to know, you know, how much of your influence is nitrate or, or nitrate. So what one way that we've dealt with this internally is we built our own wet chemical analyzer. These are our instrumentation folks, Josh and Arba, who built this and we named it the Darbelizer for Josh and Arba. So this is just our internal instrument that we built. We still use, you know, we still use nitro tax and ammonia ISE sensors. And it's sort of a case by case basis where we need to use these but there's some cases where we really want to have, you know, the wet chemistry of being able to differentiate nitrite is also getting a really low reliable ammonia values. So the two implementations that we have are one is at York River of mainstream anamox. And at this plant we're doing intermittent aeration intermittent step feed to meet the AVN step point so it's changing the aerobic fraction to get the ratio of ammonia to NOx coming out of the BNR process. And then that's feeding into a tertiary deep bed denitrification filter that we converted to now partial denitrification anamox and that has been operating for three years. These are the aeration tanks here we have six tanks so there's 12 DO probes to in each tank and we have two Darbelizers with three sample locations to do post aeration and then filter and fluid and filter effluent to do the feed forward feedback carbon dosing control and this first analyzer does the aeration control. And lastly this is our James River plant. So these there are nine parallel aeration tanks here. And we're doing here we're doing AVN continuous aeration. So this is just modulating just like ammonia based aeration control, modulating the DO step point up and down to meet the ammonia to NOx ratio going into the PDNA zone which is integrated IFAS. So this is the the yellow dots are the DO probes in the aeration tanks. And then at the end of the tanks we're using the ammonia YSI ISC sensors and and the NitroTax hoc sensors for the carbon dosing and then we also have some of the wet chemical analyzers they're not not all set up yet but we will be using them in the future and here we're just using a simple feedback carbon dosing system to feed either methanol or glycerol. So that is that so now we'll do the discussion portion. Is this the final slide, Stephanie? Yes. Perfect. Thank you. Incredibly interesting. All right. So I think we have a few questions already prepared. I haven't been answering. So, Jeff, Vicky, thank you because I've seen that you've been answering questions. Hopefully that's been useful. And yeah, I think our audience is quite grateful for it. Stephanie, I guess you will start receiving questions now so feel free to answer them as you can but otherwise if you receive any questions to answer live then, yeah, any questions that are left unanswered we will try to answer them after the presentation and then add them as materials to the IWA Connect platform. So don't worry anybody if there's a question that is left unanswered. Ali, do you want to take over? I can take over. So I think thank you everyone who have put questions in the Q&A, keep them coming. I thought I would start off there was a good question about how the operators have adapted to the new control methods. So I thought that would be a good start for all of our panelists to kind of speak to operator interactions with these different control schemes or in cases where it's not fully deployed yet how they expect the operators to interact with them and or if they have been working with them their responses. So, Jeff, if you wanted to go first, we can go in the order of presentation. Yeah, I'll go first. Sure. So I would say that doing these advanced controls and getting the operators on board and paying attention to the data and to the control performance has been the result of a really a culture change that took probably a decade or so to really complete What we're at a point now where the operators can identify reasonably low when we have some sort of sensor fault, it's not automatically detected, and then troubleshoot those and then return the sensor back. So, you know, one of the questions was, I think geared towards like, you know, are the operators going to accept and adopt these really complex controllers, because maybe it makes their lives harder. And of course, the objective here is to make lives easier. And, and we don't want to do anything that's counter to that. And so, you know, one of my big objectives has been to basically have this thing run off to the side and I say this thing, I mean, the digital twin and machine learning algorithms run off to the side on the separate server where they can't see it. And what happens is, is that it will stay in control until there's an issue. And when there's an issue when ammonia goes high, for example, for an ammonia based iteration controller, the assumption will be that, well, we've lost control, and the digital twin or machine learning algorithm is not doing what it's supposed to do. And so we need to revert to a simple mode control. And so the control system would do the switch automatically. So we would go from, let's just say advanced control to DO control. If the ammonia exceeded three milligrams per liter, five milligrams per liter, you know, you pick a number. So you go to the control of two milligrams per liter, you have an alarm that says, Hey, we've switched for no longer in advanced control. And then someone has to do some troubleshooting. But the good thing about this approach is, is that you can stay in that simpler mode of control for a while. You're wasting energy of course, more than likely. But I think, you know, there, we can probably, at least in my case, maybe not for Stephanie, but at least in my case, we can probably still achieve our TI internships. So yeah, that's all I have to say. Thank you. Yeah, Vicky. Interesting what you have to say. Yeah, the way it's quite difficult when when you install advanced control systems that and you will arrive today with what the treatment plan and operators are just used to, to, to, to work with a with a PIT or something or more simple. It's like, wow, but my experience is really good because when we install this, this type of control systems is in a, in a SCADA where they have the flexibility to change all. So for instance, with the diesel foxy and sub points they put limits is what we have mentioned about the knowledge per rule. So they, they, they can just modify the range of diesel foxy and where they want to move. And if they are not confident that at first with that they limit the range of diesel foxy and and they put just one and a half or something. And they just, when they just are confident with the control system, they just widen the range. So my experience is that first is like, they are saying that I guess, wow. But after all, they are, yes, they are happy with that. Yeah, yeah, I like that perspective that our experience has been that too it's not all in it tends to be, you kind of inch your way into it they get more comfortable and you push you push the boundaries a little bit. Yeah. That helps accelerate adoption among the facilities that they've operated it one way for so long. Um, I think for us I can give two different examples, like one being York River and one being James River that when going from, um, you know, when going when transitioning to like the AVN Erasure control which I think is the more complicated thing than the carbon dosing control. It really depends on what they're doing currently so if they already have like DO control or ammonia base Erasure control it's not that much more complicated. So like for example at James River they had already been operating again incrementally like DO control to ammonia base Erasure control to AVN control for a while so that I don't think they I don't think the operators see that much of a difference there. But at York River. We also, we were going from no DO control. So we're going from just manual valve control. We went all the way to a very complicated controller with intermittent aeration which has like many many automated valves. Um, so. Yeah, I think. I think there it's just maybe our experience has been someone still needs to be paying close attention for us that's probably the process engineer. And yeah, I think that this reiterating what everyone else has said doing, I think doing incremental steps has been helpful but there's there's no way around that it is, you know, it is additional training. Um, no, I think that that was a really good point that, you know, it is a compromise, I think, and maybe the speakers can comment on this to that. Although we don't want to make their jobs more complex. There is an elevation in the process knowledge that someone has to have on plant site in order to assess what is going on and that might be a shift in. Um, and how the plant is structured or what, what expertise is required among plants that may not be the plant operators themselves, it might be in a trustee case then process engineers who have really helped accelerate adoption. But I don't know if any of the speakers have any comment on that. Okay. Alright, so another, another question that I thought would be good for the speakers to kind of address. So you know, in general, the motivation behind this work is either for operational efficiency and reducing chemical and energy costs and or or for, for example, capex costs in terms of upgrades and reducing reducing the requirement for major construction projects through intensification processes. And I'm one of the one of the questions really was like what is the most efficient control technology. I think it's very subjective but I am curious in terms of identifying what might be the best approach for the for a particular facility, you know, from the speaker's perspectives, what would you assess. One of the one of the drivers that would take you from, say, something like feed forward model predictive control or did with a digital twin versus traditional feedback only a back control, or in the sense of of your example Vicki where you're using ORP and pH probes which are a little bit different in terms of their maintenance requirements. I can start. It's really subject to the answer because it depends on the facility depends on infrastructure depends on what you can change or not. Yeah, is. I think that there is not only one optimal for one facility there are a lot. You can go to mechanistic models and other bar control system that depends on mechanistic models, you can go to whatever I mean a PID and this PID could work well. So, mainly on the on the philosophy, and I think also on the on the range of facility that you give to the control system, if the control system is really limited for, for, for instance, the dissolve oxygen range. I think that there is no flexibility and you cannot optimize as well. I think that there is no one solution. Can you refrate can you repeat the question for me is it from going from are you saying going from different levels of control or justifying installing the sensors. So what what from a facility is looking to, you know, identify what the most efficient control system is for their specific goals, you know, what would you say they need to, to, where do they begin in terms of assessing their processes and, and their objectives. It's a I'm sure it's a whole thought could be a whole white. I mean, for, for beginning us where would you begin I feel like you'd always go. So you always start feedback PID control. So that same same thing that kind of Vicki was getting out where it would be, you start small, and then incrementally, either you're widening your range or you're increasing complexity. In terms of your objectives. But though because we have had we have had conversations like that before, like the question that, you know, the thing that Vicki brought up earlier about, you know, PID being simpler and then like, making sure they're more comfortable but when we've had these types of conversations before it's like operators don't necessarily even know what's going on in the background so like if something works it works so and from a from a graphic standpoint operating the plant and operator can't tell if it's a PID or a more complex model so if something works that I don't think the operator can tell so I don't think you necessarily need to start with PID. Yeah, I only proposed starting with PID because I mean, whether it's an operator or it's a process engineer. Why would why would anybody want to work with the additional complexity. Yeah, and you don't know if you need the complexity until you have the baseline. Yes, that that is true. Yeah. And I think on that note, the one thing that that, at least in my experience is very true is that these systems aren't maintenance free. So the more complex your controller is the more upkeep it likely requires in order to maintain its operation. So that in this this simpler it is the less the less frequently you have to interact with it to maintain the performance that you observed during your initial optimization phase. I don't know if the speakers agree with that assessment. Okay, and so moving on to another another question that I think often comes up in any sort of automation and control presentation is really discussing the sensors and all of you kind of hinted at the the sensors that are utilized in the facility. And I'm curious from, you know, from your experience, specifically the HSD facilities, you know, when it comes to the sensor maintenance and I actually I think this is a great question for you when it comes to sensor management QAQC. And then, you know, I think you mentioned fault detection also in your presentation. How, how essential and critical are some of these things to making sure these work obviously the sensor needs to be accurate for the control system to work but but making sure that the adoption of these methods to maintain the system you know and how has it been received by the facilities. I'm just curious what everyone's perspective is. Yes. That is one thing that we are just fight for this, because we also implemented ammonium based sensor ammonium based control systems. And the problem of ammonium based control systems is that mainly depends on the deterministic value so it depends a lot of the calibration and maintenance that you make to these to the sensor. And so for this reason and others, we move to other control systems that depends on tendencies that also with ammonium it could be. And also for low maintenance because ammonium nitrate sensors are now are there. They are progressing a lot and they are quite feasible to to obtain, but to maintain a still are really hard in some time sometimes for operators and for calculating them and for maintaining them in in in quite in quite good quality. So, when we compare with the pH and our sensors our experience is that the calibration and maintenance is low, lower than the than the ones for for ammonium nitrate sensors. But I think that the technology would be further and would be better also. Let's see. And for that reason and for the tendency and so on. I think that is nice in that in that point. And so what what I mentioned is that the data quality assessment is that all these control systems depend on data and data for data folks so we must move to that. Also to to put in the control system is that the data quality assessment to assure that this is the data that is good, and to assure a protocol system performance so we depend on data and that is what we have. Yeah, so I think the question was regarding sensor of all perfection. And then adoption among the operators, like maintenance, the common issues. Yeah, I mean, the maintenance of course is a big deal. And then there's always this confusion about who does it right is it wouldn't operate it does it. It is an instrumentation. Who has the right set to do it because, you know, the validating instrument, for example, wow, the operator probably has a better space. So that's been a challenge is sort of like drawing that line about who does what. And then just identifying the faults so we found that if you don't have the alarm set up in the DCS to sort of automatically indicate that there's problem, just relying on human to find a bad value. So yeah, it seems like sometimes we over automate and we over alarm. And then, and then you have to strike this balance that you don't alarm so much that alarms become white noise and operate certain ones. So yeah, it's a constant battle. I don't know that there's like a good, a good answer to the question, but we're always working on it. We're always implementing trainings to try and get folks familiar with how to operate maintains things doing weekly validations for the examples out of the tanks and very samples and the sensor analogy. Yeah. Yeah. Yeah. Yeah, I agree with everything said so far. I guess maybe one additional thing we've struggled with is just so many of the maintenance items on plant site are done on this like preventative maintenance schedule. And so then it seems like the. Yeah, it seems like because it's done that way operators like okay well how often do I need to do this, and they want like a set schedule of calibration and maintenance and we know how often we want to check it, but we definitely don't. We can't just say calibrate this every month. So it that I think that's been different, you know, plant to plant how we end up who watches it, how we decide, you know, when a sensor needs to needs to be calibrated is we it's not. It can be standardized and if you say like three samples that deviate more than X percentage from the, you know, the lab value, but then again someone if that's more careful attention than people are used to doing when it comes to maintenance. Okay, so the reason I, you know, I reason I asked that question is that that comes up quite a bit. I think in a lot of discussions I've had around automating control systems is that sometimes I think we're willing and excited to put in these control systems and other simple ammonia based aeration control or something more complex. But we're also at the same time figuring out what do we need to do in order to maintain the instrumentation required to operate these systems. And, and I know that there's, there's definitely been some literature in the past on how to how to best do that, but utilities have struggled, I think to adopt some of them mainly because of the personnel and personnel and knowledge requirements when it comes to to making things happen. So thank you I appreciate everyone providing a perspective on that. I think another good question that this is more, I think, directed at Jeff, and he did answer it in the chat but I was hoping he could talk a little bit more about it was the most challenging issue you've seen when it comes from adopting your digital twin from modeling environments to the to the real world, and I know that that's something you've been working a lot on is getting this digital twin control system up and working at that enhancement plan so I was just curious if you could comment a little more on that. Yeah, sure. They were really to see if one of them was just getting the data flowing. So getting from the control system to the digital twin server where we learn the mechanism model. And then using that data inside of the model and then of course, producing output and sending that back to the DCS. So just the flow of data from point A to B and then point B back to A. So it was a pretty big challenge. And a lot of that was due to cybersecurity stuff is not doing a little house. But yeah so I mean this is the feedback that I would get from the experts for sure. So that that took a long time. We also had some issues with the model just crashing the digital twin crashing. And this happened a lot. So between getting data to flow and just keeping the model running without crashing, you know, probably talking about maybe two years of time. So a long time. And I know that the sound may be like trivial things but they were not in the space. So the issue with the model crashing is, of course, anybody who's run a simulator before knows that they crash. And that's, that's fine. You're just doing a desktop exercise or sometimes it's not having to sort of on all crashes. But yeah, it's even worse if you're trying to use that model control system. So using the good folks at dynamite, they felt solution for us and that was to do a couple things. One is is that we run into a problem that the model just kind of pauses and then tries to model again. And it just tries to pause and keep going. And if it can it will if it can't and it ultimately does just shut down and crash, then they built in this thing all rapper, and maybe you guys know what a rapper is but it would just reopen sumo and then pick back up where I left off. So yeah, I mean those were the two real two biggest things. Yeah, I appreciate that. So I think that's kind of all of the time we have for today, in terms of the discussion. I thought I would give the speakers. If you guys have any final, like, you know, at the 30 seconds comment if you'd like to make before we go any key takeaways you'd like to just emphasize. I'll go ahead and give you that opportunity now. Is there anything you'd like to say. Okay, Vicki or Stephanie. I'm here. I feel like we covered everything. Okay, perfect. All right. And I think I've got just one last thing to hit on just need to go ahead and transition the slide. Okay, so just a reminder that there are some upcoming webinars that IWA is hosting. One is on embracing indigenous perspectives to achieve sustainable development goals that's on August 9. And then there's another one on sustainable estuarine and coastal development that is on August 19 and you can register at IWA-network.org for slash webinars. And then also, there is the IWA Digital Water Summit in Bilbao, Spain is coming up on November 14 to 16. And it's going to be a great event registrations now open. And it's in reference to digitalization of the water sector where I'm sure very similar topics to today's presentations will be discussed. And it's planned as a B2B event. And I think that's all we have for everyone today. Thank you so much. And if you all could fill out the post-webinar survey, that would be great. There's also the Water and Development Congress and Exhibition on water sanitation and climate resilience. And that's 10th to 14th December and Kigali Wanda. And you can find out more information on that on waterdevelopmentcongress.org. Thank you everyone. I think it was a slide before that. Before I thank you one. Okay, oh yes. Join IWA. Yes. So, yeah, as you can see we're a big network with many thousands of professionals. So if you found this webinar interesting. Well, in IWA Connect, you have all the information and you can get a view of all the different technical groups, webinars, events, and any other materials that probably will help you during your professional career. I think that's it. Thank you all. Thank you to these speakers. Thank you Ali for moderating a great Q&A session and thank you everybody who attended. I hope you enjoyed it. I hope you found it interesting. And yeah, everything will be available and the link that one of my colleagues have posted on the webinar chat. Cheers.