Outlook

All our knowledge

has its origin

in our perceptions.

– Leonardo da Vinci

Arrival of a technology

Almost 25 years of AGS research has brought us many insights in the technology. Although still young compared to activated sludge, which was invented well over 100 years ago, with the arrival of the 100th full-scale Nereda^® installation in the very near future AGS technology can by no means be considered a novel technology anymore. Nevertheless, we are only starting to discover the full potential of the technology. A deeper understanding of the mechanisms at work in an AGS reactor, will help to unleash its full capacity. In this dissertation I aimed to add to this deeper understanding and hopefully I transferred some of the insights I gained during my PhD research to the reader.

Degasification control

Degasification of nitrogen gas in AGS reactors remains an elusive problem. The main reason for this is that the occurrence of degasification depends on many factors, like temperature, both current batch size and previous batch size, nitrite/nitrate effluent quality, residual COD from the previous batch (or endogenous respiration), aeration and mixing intensity in the main aeration, granule size, biomass concentration, and probably a few more. As a result, occurrence of degasification problems, resulting in elevated levels of biomass in the effluent, can be quite erratic. Although the silk-like appearance of a degasification scum layer should be easily recognized by the operator, because it is quite distinct from other scum layers that can occur in wastewater treatment plants, identification remains a problem. We showed in chapter Effluent suspended solids how the process of degasification works, but translating knowledge into a simple process control to prevent degasification, is less straightforward. The main obstacle is the fact we cannot easily measure the dissolved nitrogen concentration in the liquid to control the nitrogen deficit in the stripping phase. In practice this means the stripping phase needs to be carefully balanced between long enough to always strip enough nitrogen to be safe under all conditions and short enough to not limit the treatment capacity of the AGS reactor.

The major obstacle to having a proper process control for the stripping phase is the fact that we cannot measure the nitrogen concentration in the water phase. We can measure ammonia, nitrite and nitrate, but there is no adequate sensor for measuring pure nitrogen (N₂). A possible solution could be to create a soft-sensor, based on measurements of ammonia and nitrite, combined with the mathematical model described in chapter Effluent suspended solids extended with a more elaborate description of the biological processes involved. It should be possible to get a fairly accurate estimation of the nitrogen gas concentration after the reaction phase and adapt the process control of the stripping phase (and the settling and feeding phase) accordingly. This should minimize the problems with degasification scum layers and prevent it in most cases.

Process optimization

The AGS system can appear complex, when compared to conventional activated sludge plants. In a CAS plant the sludge flocs experience changing process conditions while traversing from tank to tank, but on average all sludge flocs are exposed to these process conditions in a similar manner. On average, all flocs receive the same amount of influent, are exposed to same amount of oxygen, and have a similar chance of being spilled. Of course, we must emphasize the words on average here, because the reader will quickly notice that it is common practice to increase floc loading rates by use of contact tanks, where only part of the return sludge is mixed with all of the influent. Also, because of residence time distribution in aerated tanks, some flocs will have shorter exposure to oxygen then others. But on average the process conditions for all flocs can be considered similar within the duration of the solid retention time. For aerobic granular sludge the process conditions are not similar for the different granule size fractions. As we have seen in chapter Settling behaviour different granule sizes have very different settling properties (with terminal velocities that can be 25 times higher when comparing the largest and the smallest granule fractions). An important effect of these different settling velocities is the process called selective feeding (see chapter On the mechanisms). This means that the substrate loading rate is extremely skewed towards the largest granules and on average the largest granules receive the most substrate and as a result will grow more. The wasting of sludge is skewed towards the smallest granules, by a process called selective wasting. The AGS process is engineered to selectively waste the worst settling fraction and to retain the best settling granules. As a result, there is a clear Granule Residence Time Distribution (GRTD). The large granules can have an age of more than 50 d, while proto-granules can be spilled within hours or just a few days. Also, on a micro-scale AGS adds complexity compared to activated sludge flocs. SND is a process that occurs in AGS because aerobic and anoxic conditions can co-exist within a large enough granule. The anoxic zones within a granule are a great benefit of AGS, because it decreases or totally removes the need for a separate denitrification phase. In a sense the aerobic and anoxic zones within a granule make an AGS reactor behave as if it were an aerated reactor and a denitrification reactor simultaneously.

Fig. 49 Optimization process in an AGS reactor.

So, we have different circumstances between granules (loading, SRT, selection) and different circumstances within granules (redox circumstances, loading). As a result, there can be multiple steady-states for an AGS reactor. We have shown in chapter Settling behaviour that multiple granule size distributions with the same amount of biomass can be stable under the same selection pressure. The same counts for biomass population distribution, conversion rates and in the end effluent quality. Baeten et al. [2018] showed in their research [Baeten et al., 2018] that it takes hundreds of days to reach a steady state for the population within a granule. In this research (chapter On the mechanisms) we showed it can take up to a year to get a stable granule size distribution. At the same time the age of the aggregates in an AGS reactor can range from a few days to several months [Ali et al., 2019] simultaneously. In my opinion this means we never reach a steady state in an AGS reactor, because the time to reach a steady state is (much) longer than the age of a large part of the granules. So, although from a process performance point of view the AGS process is generally very stable, we should always approach the AGS process as a system in a semi steady-state, with a mixed population, where granules are constantly growing and developing into larger particles.

When designing/researching/operating/modelling a AGS reactor, one must be aware of these semi steady-states, as illustrated in Fig. 49. This figure is an abstract representation of the optimization process of an AGS reactor. On the x-axis we have the state of the reactor and the y-axis represents the error regarding the performance of the reactor. One can see the latter as the actual performance minus the desired performance. Optimization can be seen as traversing towards the desired performance. Sometimes during the optimization process we need to move away from the desired state, to break out of the semi steady-state. For example, lowering the selection pressure, decreasing the load, worsening the effluent quality can be necessary to come to a better performance of the reactor. It is important to be at least aware of this when making designs, deploying start-up strategies or troubleshooting a AGS reactor.

Mathematical models

We tend to use many models in design, operation and research of wastewater treatment plants (Fig. 50). Design models are generally less complex than models used for operation and models used in research can be very complex, with a clear example in chapter On the mechanisms. Design models can be simple, because processes can be lumped into simple parameters, and some safety margins are added to the model to deal with the uncertainty caused by lumped parameters. Because of the batch-wise operation, some dynamics are generally added to AGS design models, adding complexity. There is always a trade-off when adding complexity to a model: more complexity means more effort to calibrate and validate the model.

Fig. 50 Modelling pyramid, showing different levels of modelling complexity with different applications.

In wastewater treatment there is a special role for the Activated Sludge Models (ASM). ASMs are widely used to model CAS systems, an also some attempts are made to model AGS systems with ASM type of models [Baeten et al., 2021, Dold et al., 2019, Layer et al., 2020, de Kreuk et al., 2007]. These models generally miss the essence of the AGS process, because they miss the complexity as described in the previous paragraph. They lump essential parameters, such as the GRTD, in average values, missing the dynamic behaviour of the AGS reactor. Somewhere in the near future an effort should be made to develop an ASM suited for AGS modelling. This could help greatly in the further development of the technology.

Alternative process control

Process control in the full-scale AGS process in essence is quite simple: in the feeding phase a batch of fresh influent is fed to the reactor, in the reaction phase the reactor is aerated, until the COD, ammonium and phosphate requirements are met (possible combined with pre-denitrification, intermediate denitrification or post-denitrification), then the sludge bed is allowed to settle, and the selective wasting is done. On the other hand, process control can become difficult rapidly, when one realizes that the optimal oxygen concentration does not exist or at least depends on the granule size distribution. Larger granules need higher oxygen concentrations, to get the maximum reaction rates inside the granule. During aeration a concentration gradient exists inside the granule, and the oxygen concentration can drop to zero in large granules. Increasing the bulk oxygen concentration will increase conversion rates for oxygen dependent processes in the large granules. For the smaller granules and flocs this effect will be limited, because oxygen will completely penetrate the biofilm. Also, during the reaction phase the anoxic zone within an aerobic granule will shift inward while the COD on the outside of the granule is depleted (see Fig. 51). Small granules will reach a state of full oxygen penetration much faster than large granules. Large granules might remain partially anoxic throughout the whole reaction phase. As a result, a mature granular bed with mainly large granules will produce a better effluent quality regarding nitrate compared to a granular bed with only small granules. The process control needs to be adapted accordingly. Realizing these dynamic conditions in the granules during the reaction phase, it is a small step to a more adaptive process control, where the oxygen concentration is varied over the cycle. First steps to create such a process control have already been made [Layer et al., 2020, van Dijk et al., 2020], but there is still a lot to gain.

Fig. 51 Effect granule size on denitrification.

Process control is mainly targeted at effluent requirements. The research on N₂O emissions described in chapter Nitrous oxide emission showed that there are opportunities to minimize the N₂O emission from AGS reactors. There was a clear effect of the change of process control at the end of the measurement campaign at the wastewater treatment plant of Dinxperlo. We showed the dynamics of the N₂O emission during the cycle: periods of increasing production of N₂O were alternated by periods of reduction of N₂O by denitrification. Because N₂O concentrations can be measured in both the water phase and the off-gas, it should be possible to minimize N₂O emissions by controlling the dissolved oxygen concentration in the reactor and adding denitrification phases when N₂O concentrations become too high. Such a process control could minimize the ‘environmental’ impact of the wastewater treatment plant. An optimum between the best effluent quality and the least greenhouse gas emission should be the focus. The latter would probably not be a trivial optimization, not only from a process control point of view, but even more from a regulatory perspective.

The N₂O trial in Dinxperlo also showed the potential for process control based on off-gas measurements. The current process control of Nereda^® reactors relies on measurements in the water phase (ammonia, nitrate, phosphate, oxygen) for the installations with the most stringent effluent requirements. These sensors are expensive and need regular maintenance. In contrast, the off-gas unit we used in Dinxperlo did not need much maintenance, because it was measuring only the clean off-gas. The unit can be used to get respiration rates, biomass growth and other relevant parameters [Baeten et al., 2021]. Possibly in combination with a simple sensor like a pH probe, this off-gas measurement could provide a whole new approach for process control of aerobic granular sludge.

In the wastewater treatment plant of Zutphen a new control strategy is developed. At this location the first Kaumera extraction is built, producing biopolymers from aerobic granular sludge. The AGS is grown on dairy wastewater, which is one of the streams treated at this wastewater treatment plant. Because the product here is biopolymers and not clean effluent (the Nereda^® effluent is polished by the CAS system that treats the domestic wastewater from the municipality of Zutphen), totally different goals arise for the process control. It is much more about amount and quality of biopolymers produced and about the stability of the sludge production. In the case of Zutphen, these goals are relatively clear, because the sole purpose of the plant is to produce biopolymers. For future AGS plants, this goal could, similar to minimization of the greenhouse gas emission, be a secondary goal of the process control: to produce good effluent quality, but also to deliver the optimal properties of biopolymers.

Continuous aerobic granular sludge

The current full-scale application of AGS is based on a batch system. During my PhD research I have been closely involved in the PhD research of Viktor Haaksman, who is one of researchers looking for a continuously fed application of AGS. It seems it is not a question if continuous AGS system will be developed but more the question when this technology will arrive. There appear to be some benefits to continuously fed AGS systems, the largest one maybe being the fact that retrofitting existing CAS systems into the AGS process could be a straightforward method for increasing the treatment capacity, without many investments. These benefits might in many cases outweigh the downsides of a continuously fed system - for example loss of concentration gradients, loss of flexibility in the process control. The mechanisms for aerobic granulation (chapter On the mechanisms) are closely related to the batch system in which the AGS process was originally developed. In a continuously fed system, the presence of these mechanisms is not per se evident, and I believe it will always remain more difficult to grow aerobic granules in a continuously fed system, compared to a batch system.

Process knowledge versus big data techniques

In a batch system much more process information is generated than in a continuous flow through process. In CAS systems the sensors measure (more or less) a constant value because process conditions are kept constant over time. In a batch system all cycle measurements are done under varying process conditions. As a result, in a batch system for every reactor a vast amount of data is generated. This data can be put to good use. We could, for example, use artificial intelligence techniques to predict process failure or to minimize energy usage. The possibilities seem endless. But also from a more ‘old school’ perspective, a batch system gives a lot of process information. Every cycle gives nitrification and denitrification rates, phosphorus release and uptake rates, endogenous respiration and so on. On a daily basis, process engineers and operators can monitor these rates and act upon them. Based on knowledge rules, we can provide early warnings about the process performance. But the many degrees of freedom in the system - as discussed earlier, size distribution and GRTD make the process more complex - and the abundance of process information ask for machine learning techniques and other artificial intelligence applications.

This all starts with the collection of valid measurements. Data collection sometimes struggles with common issues familiar to anybody working in wastewater, like rags, fouling and lack of maintenance. These all influence the validity of the measurements. We can setup fancy process control or give early warnings based on historical and current process behaviour, but if measurements are incorrect, it will malfunction. Data validation and reconciliation techniques could help greatly with this. There is some redundancy in the system: different reactors under the same process conditions should mimic each other, allowing for comparison of sensors. Redox sensors should make sense if we compare with oxygen and nitrate, and vice versa. Nitrate production cannot be (much) higher than ammonia concentrations and so forth. Using modern data validation and reconciliation techniques could improve process reliability.

Microbial control

Understanding the mechanisms for aerobic granulation gives us new possibilities. These mechanisms, especially the mechanisms of selective feeding and selective wasting, give us some control on the microbial populations which grow in the AGS system. Over the past few years, we learned that different microbial communities exist in different granule sizes. Analysis of the metagenome and, more recently, of the proteome provide insights in the behaviour of different species resulting from the GRTD. We could use this knowledge to reverse the process: influence the GRTD to get the microbial communities we prefer. For example, we could favour PAO over GAO to improve phosphorus removal. We could enrich for nitrifiers to increase nitrification rates. But we could also try to influence the desired product characteristics of the Kaumera produced from the AGS waste sludge. Soon online measurement of the metagenome will bring us a new tool to directly monitor and influence the microbial communities. This will bring new opportunities to minimize the footprint of our wastewater treatment plants, to enhance the effluent quality and to generate new products from the waste AGS.

Not the end

The future of the aerobic granular sludge process is bright. When compared with CAS systems, with AGS one can save both on investment costs (decrease of total reactor volume) and operational costs (less energy consumption and less chemical usage). On top of this, recovery of potentially high-end biopolymers, but also recovery of for example phosphorus make the technology well suited for the current day demands of modern wastewater treatment. There are still so many opportunities to increase the treatment capacity of the AGS process, so many discoveries to be made: the journey only just started, and we will see where we go from here.