Traditional ozone dosing methods are not adapted to increasingly variable water. Laura Ferràndez Galceran's research, within the framework of an industrial doctorate between the UdG, the LEQUIA research group and the company FISERSA, demonstrates that artificial intelligence can solve this challenge with real, measurable and transferable impact.
The current climate scenario, marked by extreme droughts and episodes of torrential rain, has rendered water management based exclusively on historical experience obsolete. Every day, drinking water treatment plants (WWTPs) everywhere face an equation that becomes more complicated year after year: producing safe water from a raw material such as raw water, which is increasingly unpredictable . Extreme droughts concentrate natural organic matter to levels that alter the effectiveness of conventional treatments. On the other hand, torrential rains, on the other hand, bring sudden pollutant loads that require immediate responses. In the midst of this situation, European regulations on disinfection by-products (such as trihalomethanes (THMs)) narrow the permitted operating margins. The result is a scenario where the accumulated experience of operators and control systems based on fixed thresholds are no longer sufficient to simultaneously guarantee efficiency and safety in the treatment of drinking water.
ETAPs today face a complex puzzle: ensuring microbiological and chemical safety under increasingly changing inlet water conditions. In this context, the ETAP in Figueres has served for four years as a full-scale laboratory for research that proposes a paradigm shift: replacing reactive control with predictive and adaptive control based on artificial intelligence . The industrial PhD student Laura Ferràndez Galceran, within the framework of an industrial PhD project between the company Figueres de Serveis SA (FISERSA) and the LEQUIA research group of the University of Girona (UdG), has designed, implemented and validated a control system based on fuzzy logic that adapts ozone dosage in real time to the variable conditions of the treated water. The thesis, directed by Dr. Hèctor Monclús and Dr. Alba Cabrera, was defended on March 13, 2026 at the Faculty of Sciences of the UdG, in an event open to the public.
"Working in a real environment provides a fundamental advantage: it allows us to understand the complexity of the system beyond ideal laboratory conditions, both at a technical, operational and human level."
The limitation of conventional control systems
WWTPs function as critical barriers between raw water collected from the environment and the water that reaches homes. Treatment with ozone, a powerful oxidant, is one of the key stages to eliminate organic contaminants and ensure disinfection. However, ozonation also presents an operational dilemma: an insufficient dose does not eliminate enough contaminants; an excessive dose generates potentially harmful by-products, such as the aforementioned THMs . The balance depends on variables that constantly fluctuate, such as the organic load of the water, ambient temperature, collection conditions, among other factors.
Conventional control systems tend to operate with fixed thresholds that do not always respond accurately to fluctuations in organic matter or temperature. This operational challenge is what led Ferràndez to propose a decision support system that ” allows us to understand the complexity of the system beyond ideal laboratory conditions,” the researcher explains. Her work has allowed us to capture the real variability of water in an operational environment where uncertainty is the norm.
Therefore, Ferràndez-Galceran’s research addresses this limitation head-on. ” Even without developing a model, this type of research already offers a differential value, because it requires managing real uncertainty, living with operational limitations and understanding how decisions are made on the plant ,” acknowledges the researcher about the experience of working in real operational conditions, far from the predictability of the laboratory.
A control system that formalizes expert knowledge
The core of the thesis is a fuzzy inference system that is fed by two continuously measured variables: ultraviolet absorbance at 254 nanometers (A254), an indicator of the quantity and nature of dissolved organic matter, and water temperature, which conditions the solubility of ozone and, therefore, its oxidizing capacity.
In this technical context, the system developed by Ferràndez works as a digital operator that makes decisions in real time . To do this, it continuously reads two key data from the water: the amount of dissolved organic matter (measured by ultraviolet absorbance) and the temperature. Based on these two readings, a set of decision rules, built with historical data and expert knowledge, translate them into a specific ozone dose.
The difference with respect to traditional control systems is conceptual. Conventional systems work with fixed thresholds: if a value exceeds a limit, the system acts in a certain way. Ferràndez’s system, on the other hand, reasons with gradual categories: “moderately high temperature”, “high organic matter”, etc.; just as an experienced technician would do in the same situation. ” It is much more similar to how an experienced person interprets reality “, explains the industrial doctor. The system does not replace the operator: it learns the criterion and applies it automatically.
In practice, the result is measurable: “ the system can adjust the ozone dose in a way that is consistent with the conditions of the incoming water, helping to maintain the efficiency of the process and minimize the formation of by-products” , clarifies Ferràndez. When the temperature rises, the system automatically increases the ozone dose to compensate for the fact that the gas dissolves worse. When it detects more organic matter, oxidation increases accordingly. Tested in a real installation, the solution improves oxidation efficiency, reduces reagent consumption and minimizes the generation of unwanted by-products.
"Even without developing a model, this type of research already offers a differential value, because it requires managing real uncertainty, living with operational limitations and understanding how decisions are made on the plant"
The challenge of integrating data from heterogeneous sources
A predictive system is only as good as the data that feeds it. In the case of the Figueres ETAP, the main obstacle was not the lack of information, but the dispersion of this data. Plant control instruments, laboratory point analyses and online sensors generated data with different sampling frequencies and, in some cases, with significant gaps. Ferràndez describes it precisely: ” The real bottleneck was to build a coherent and reliable data set that allowed not only to analyze the past behavior of the system, but also to support a model with predictive capacity and applicable in a real operating environment. ”
This data harmonization task is often invisible in the final results, but it conditioned the entire modeling process. Therefore, the developed method is transferable to other facilities that want to implement intelligent control systems.
The human factor as a condition for success
It is a fact that no decision support system works if the people who have to use it do not trust it. Ferràndez and the project team designed a progressive adoption strategy: first, validation of the model with historical data; then, application of the recommendations in parallel with the real operation; finally, gradual integration into the plant dynamics with iterative adjustments. ” Operators’ trust is not built only from the model, but from the implementation process, and this has been key to its adoption as a Decision Support System ,” emphasizes the researcher.
A determining element in the acceptance was the transparency of the system. ” The fact that the system is based on expert rules has been a key element in favoring its acceptance, since it allows to understand the criterion behind each recommendation ,” explains Ferràndez. The operator can inspect why the system recommends a certain dose. In short, he does not perceive a black box, but a tool that complements his professional criterion without replacing it.
A cost-effective and quality solution
From a business perspective, the immediate benefit of the project’s outcome is tangible: a more precise adjustment of the ozone dose leads to a reduction in reagent consumption and the elimination of overtreatments. But Ferràndez points to a less quantifiable and potentially more valuable dimension: ” the intangible value lies in a more stable operation, more resilient and less dependent on reactive decisions or individual experience. ” In other words, in a sector with personnel turnover, a system that preserves expert judgment becomes a strategic asset.
The system also allows working with a greater margin of safety in the face of variability in the incoming water and further reducing the levels of by-products such as THMs at the plant outlet, improving the efficiency of subsequent treatment stages.
"Operator confidence is not built solely from the model, but from the implementation process, and this has been key to its adoption as a Decision Support System."
Towards a comprehensive decision support system
The validated ozone controller is the first piece of a broader architecture. Ferràndez’s project establishes the conceptual foundations for two additional modules: one for chemical risk assessment, focused on predicting the potential for disinfection by-products formation, and another for microbiological risk, based on indicators of assimilable organic carbon and bacterial regrowth. Integrated, these three components form the framework for a future modular Environmental Decision Support System, designed to provide comprehensive and informed risk management throughout the treatment chain.
From a scientific point of view, the project has provided the LEQUIA research group with the incorporation of analytical techniques that are not common in plant operations. Ferràndez particularly highlights the determination of assimilable organic carbon through bioassays, a method that he describes as complex and with ” high sensitivity to contamination and a certain experimental instability “. Despite the difficulty, the technique has allowed the generation of new knowledge about processes that are not usually monitored in detail.
A model that can be exported to any treatment plant
The system is not a directly exportable solution, but the method is. Since it is based on local data and expert knowledge, and not on closed mathematical models, each new installation requires a prior phase of analysis and adjustment. In other words: what is transferred is not the configured system, but the way it is built. ” It must be adapted to the specific conditions of each plant before applying it directly ,” Ferràndez specifies.
The impact of the research, however, goes beyond Alt Empordà. The modular structure facilitates the transfer of knowledge to other plants with similar problems, both nationally and internationally. In addition, the project has incorporated advanced scientific methodologies such as the determination of assimilable organic carbon (an indicator of microbiological water quality that is not usually continuously monitored in treatment plants), providing a deeper insight into the process.
In conclusion, Laura Ferràndez’s industrial doctorate illustrates a model of collaboration between university and company that generates bidirectional value: FISERSA obtains operational tools validated in its own environment; LEQUIA consolidates its transfer trajectory in drinking water treatment systems.
But the researcher closes with a reflection that transcends the project and directly addresses the business community: ” Incorporating industrial doctoral students should not be a question to be justified from a gender point of view, but a natural bet to attract qualified talent and connect research and real operation .” Within the framework of the ‘Water and Gender’ focus of this 2026, Ferràndez’s career also highlights the strategic value of incorporating industrial doctoral students into R&D structures. The researcher directly addresses the business community: ” more than a question of equity, it is also a question of strategic intelligence for companies .”
In a sector in full technological transformation, where climate variability and regulatory pressure will only increase, the thesis project defended in Figueres shows that the answer lies in predictive, adaptive and resilient management systems. Artificial intelligence does not come to replace the operator, but to amplify his criteria. And the industrial doctorate, as an instrument, shows that the most solid innovation is born when research and operation share the same ground.
