Next Article in Journal
Technical Analysis and Application Prospects of Magnetic Source Transient Electromagnetic Coil Devices in Hydrogeological Survey of Mining Area
Previous Article in Journal
Driving Factors and Variability of Cyanobacterial Blooms in Qionghai Lake, Yunnan–Guizhou Plateau, China
 
 
Search for Articles:
Title / Keyword
Author / Affiliation / Email
Journal
Article Type
 
 
Advanced Search
 
Section
Special Issue
Volume
Issue
Number
Page
 
Journals
Water
Volume 17
Issue 2
10.3390/w17020170
water-logo
Submit to this Journal Review for this Journal Propose a Special Issue
► Article Menu

Article Menu

share Share announcement Help format_quote Cite question_answer Discuss in SciProfiles
first_page
settings
Order Article Reprints
Font Type:
Arial Georgia Verdana
Font Size:
Aa Aa Aa
Line Spacing:
Column Width:
Background:
Review

Use, Potential, Needs, and Limits of AI in Wastewater Treatment Applications

by
Andrea G. Capodaglio
* and
Arianna Callegari
Department of Civil Engineering & Architecture, University of Pavia, 27100 Pavia, Italy
*
Author to whom correspondence should be addressed.
Water 2025, 17(2), 170; https://doi.org/10.3390/w17020170
Submission received: 15 December 2024 / Revised: 8 January 2025 / Accepted: 9 January 2025 / Published: 10 January 2025
Browse Figures
Versions Notes

Abstract

:
Artificial intelligence (AI) uses highly powerful computers to mimic human intelligent behavior; it is a major research hotspot in science and technology, with an increasing number of applications to a wider range of fields, including complex process supervision and control. Wastewater treatment is an example of a complex process involving many uncertainties and external factors to achieve a final product with specific requisites (effluents with prescribed quality). Reducing process energy consumption, greenhouse gas emissions, and resources recovery are additional requirements of these facilities’ operation. AI could extend the purpose and the expected results of previously adopted tools and present operational approaches by leveraging superior simulation, prediction, control, and adaptation capabilities. This paper reviews current AI research in the wastewater field and discusses present achievements and potentials. So far, almost all applications in the sector involve predictive studies, often at a small scale or with limited data use. Frontline research aimed at the creation of AI-supported digital twins of real systems is being conducted, with few encouraging but still limited applications. This paper aims at identifying and discussing key barriers to wider AI adoption in the field, which include laborious instrumentation maintenance, lack of process expertise in the design of current software, instability of control loops, and insufficient incentives for resource efficiency achievement.

1. Introduction

Artificial intelligence (AI) uses highly powerful computers to mimic human intelligent behavior, including learning, judgment, and decision making, by taking knowledge as its primary object, acquiring and analyzing it through a combination of computer science, logic, and other disciplines [1]. Its recent, rapid progress is promising huge and sometimes controversial changes in many aspects of human life, promoting a new era of social development, with revolutionary impacts in labor efficiency and cost reduction, structure of human resources, and creation (or elimination) of jobs. So far, AI has achieved remarkable results in varied applications such as speech recognition and natural language and image processing, mathematics, decision making, and intelligent robotic development. AI is a major research hotspot in science and technology, with an increasing number of companies committed to its applications to a wider range of fields.
Since the 1940s, AI “precursors”, i.e., artificial neural networks (ANNs) and expert systems (ES), were experimented with, limited at the time by an inadequate availability of computing capacity and excessive cost. In the 1980s, backpropagation neural networks (BNN) were widely developed, while computing hardware’s capabilities underwent a rapid improvement phase. The general diffusion of the Internet increased AI application scenarios and, with deep learning (DL) being developed in the early 2000s, AI achieved breakthrough progresses. AI models such as ANN, ES, genetic algorithms (GA), and fuzzy logic (FL) are now capable of solving ill-defined problems, configure complex data mapping, and forecast results. Therefore, its ideal application area is that of complex operations governed by non-linear parameters and multiple interconnected processes with highly variable factors affecting the overall systems.
Wastewater treatment is an example of such complex processes, involving many uncertainties and external factors, such as sewer flow variability [2] and nonlinear process dynamics [3], leading to fluctuations in effluent quality and operational costs [4], as well as potential environmental risks [5]. In addition to conventional pollutants (BOD/COD, solids, N compounds, and P), with the improvement of analytical monitoring, new types of pollutants in wastewater have been increasingly detected [6], which are challenging to treat, e.g., pharmaceuticals and personal care products (PPCPs), per- and polyfluoroalkyl substances (PFAS), and other contaminants of emerging concern [7]. These may possibly require more effective, operationally complex, and expensive treatment processes [8]. An emerging issue in wastewater treatment, superimposed to the one of pollutants removal, is the intensive use of energy for operation [9], linked to the corresponding emission of greenhouse gas (GHG) that, under the current prevailing approach, calls for the implementation of decarbonization measures [10]. This additional constraint adds complexity to systems’ optimization under multiple, often conflicting, objectives. Multiobjective system optimization is a problem that, ideally, is well suited to AI approaches.
As an indirect confirmation of this trend, the number of published papers on AI in wastewater treatment has rapidly increased since the year 2000 at a double-digit rate, especially in the last 10 years [11]. This paper reviews current AI research in the wastewater treatment field, limited to the final treatment plant and excluding the collection system, which deserves a specific review on its own due to the equal or greater number of published studies, discussing its present achievements and future potential; it also critically discusses the uses, needs, and limitations of AI technology in the wastewater sector.

2. AI Drivers and Enablers

Graphic processing units (GPU), commercially introduced in the late 1990s, are electronic circuits (“chips”) able to perform mathematical calculations at high speed by using massive parallel processing: they can elaborate in a day results that a traditional CPU would take a much longer period (weeks, or more) to obtain. Due to this capability, GPUs have become the essential technological foundation of AI. One drawback of massive parallel processing for AI purposes is its high demand of resources. At present, data centers consume 1–2% of total global generated power (about 200 TWh/y), a fraction predicted to rise to 3–4% by the end of the decade due to the increasing requirements of AI processing: a single ChatGPT query in fact requires nearly 10 times more energy than a Google search (2.9 Wh compared to 0.3 Wh) [12]. It is estimated that, by 2028, AI applications alone will represent about 1/5 of all data centers’ power demand [12]. By comparison, the entire water industry uses between 2 and 4% of the energy produced globally [13].
Data centers are also water-thirsty installations: while the average center uses over 1000 m3/d of drinking-quality water, larger ones could require up to 20,000 m3/d (a supply sufficient for approximately 3000 to 60,000 people, respectively) to remove the heat generated by IT equipment [14,15]. As of March 2024, just under 5400 data centers were recorded in the US alone (more than in any other country, roughly 10 times more than in Germany, the second per number of installations) [16], with an estimated water footprint equivalent to that of at least 16 million people. This is starting to raise sustainability issues in several water-stressed areas.
In addition to computational muscle, energy (and hydration), Big Data is another essential prerequisite for AI successful application. The “Big Data” concept embodies a technology for storing, processing, and managing complex datasets, for which conventional processing is not appropriate. These data are characterized by a substantial volume of information, as well as by high-speed input, often greater than the available processing speed [17]. Big Data in production processes is connected to the Internet of Things (IoT), which refers to a network of physical objects embedded with sensors, software, and network connectivity, allowing the collection, storage, and sharing of a continuous flow of data. In the water sector, IoT devices can enable the collection of vast amounts of real-time data from water/wastewater treatment plants, distribution networks, and even individual households [18]. Industrial automation is one of the IoT’s most common applications, as it can greatly improve process accuracy through online control and reduction of human intervention for specific tasks; intelligent robotics systems are being developed for IoT-enabled factories working at unprecedented precision scales (up to a few nanometers). In the water sector, “smart” pump technology with native IoT connection has recently been promoted as being able to achieve possible energy savings of up to 70% in urban water distribution networks, compared to conventional machines [19].
Finally, machine learning (ML) enables computing machines to learn and analyze Big Data automatically, as well as to make decisions (predictions) about real events. Within AI applications, ML algorithms analyze the massive datasets generated by IoT devices to discern patterns; make predictions; and, in essence, transform raw data into actionable insights. This enables us to optimize operations of complex systems, predict potential issues, and adapt to dynamic environmental conditions. The basic idea underlying machine learning is the use of an algorithm continuously improving its performance by learning from data [20].
Deep learning (DL) is a subset of ML that uses multilayered, deep neural networks to simulate the decision-making power of the human brain. Some form of DL powers most of AI applications today. Figure 1 depicts the relationship between AI, ML, and DL.
In traditional computer simulations, researchers summarize physical laws and analytical methods for the benefit of computers; in AI, the approach is inspired by human babies: no one teaches them how to recognize objects, nevertheless they learn. AI discovers the statistical structure of the data and establishes rules for automating a task, rather than being explicitly programmed by humans. Based on this model, a computer can learn the characteristics of “objects” (either structured (numerical) data or unstructured (text, sound, images) data) through appropriate training models (e.g., ANNs) and can accurately recognize them in other circumstances [21]. ANNs are declined in many variations, including recurrent neural networks (RNNs), convolutional neural networks (CNNs), fuzzy neural networks (FNNs), and deep neural networks (DNNs), depending on structure and algorithms used [22]. Table 1 summarizes some of the common AI algorithms used.

3. Development of AI Applications in the Wastewater Sector

Wastewater treatment involves a complicated industrial-like process influenced by many physical, physicochemical, and microbiological factors subject to arbitrary random fluctuations over time whose purpose is multifold: to reduce pollutants released to the environment, to recover residual resources embedded in wastewater [23] and in treatment byproducts [24], and to reduce energy consumption and GHG emissions.
Systematic wastewater treatment plant (WWTP) modeling was initially introduced in the 1980s as an attempt to better understand their behavior in term of pollutant removal efficiency [25], and the later introduction of online instrumentation to gain real-time process observations and characterization (often by indirect determinations) started to allow actual control and automation (ICA—instrumentation, control, and automation) of these facilities. The key reason for ICA was the recognition that all processes are subject to irregular disturbances, external or internal, that will propagate to units within a system and impair treatment objectives [26]. Recently, ICA application was also directed at reducing WWTP energy consumption and GHG emissions, encompassing all aspects of these facilities. Full or partial ICA applications are used nowadays to improve WWTP operations [27].
AI, hence, does not come as a totally new concept in the sector, but rather as an extension of the purpose and the (expected) results of previously adopted approaches, while leveraging far superior simulation, prediction, and adaptation capabilities.

3.1. AI in System Simulation

ANNs are a key component of the AI universe, designed to reproduce the human brain neuron interactions to accomplish parallel complex computations. Simple ANN models have been used since the 1990s to simulate WWTPs’ unit/process behavior [28] or evaluate their overall performance [29], but their application potential has widely expanded with the increasing computational power provided by GPUs.
Many examples of predictive modeling use of ANN models have shown better forecasting capabilities, compared to mechanistic models, for example concerning N2O production (a powerful GHG of interest) in a nitrification lab-scale plant [30] and hydraulic retention time (HRT) influence on membrane bioreactor fouling, where ANNs were highly efficient in predicting effluent COD and membranes’ transmembrane pressure (TMP) trends [31]. ANNs were also used for predicting removal of metals and other pollutants from wastewater [32,33].
ML algorithms were found useful to predict the dynamic performance of WWTPs. ML was used predict daily long-term energy consumption at the Melbourne East WWTP by considering various parameters (time, climate, hydraulic flow, and wastewater characteristics) with the aim of performance enhancement [34]. Results of that study, however, were not used for implementing online corrective measures. In fact, most AI applications in WWTP systems simulation so far have not overstepped the boundary between off-line or past-data-based prediction and real-time, online process modelling, as reported in the published literature. In an in-depth review of ML-integrated dye removal applications, Bhagat et al. [35] determined that no particular ML model could be currently considered as an established benchmark in this area, and that future ML models should be able to handle information stochasticity, non-linearity, non-stationarity, and intricacy.
Table 2 summarizes some of the most recent literature-reported AI applications in WWTP modeling and output prediction. An in-depth review of AI technologies to tackle environmental challenges in pollution control was published by Ye et al. [36].
Different algorithms can be used to predict process behavior; however, potential dataset size and their possibility to contain non-informative features provide serious challenges to effective learning: a crucial element in their successful use is therefore the reduction of complexity and irrelevant or redundant features. A study on feature selection methods for the application of multiple ML algorithms to optimize output prediction in a real WWTP was recently published by Ekinci et al. [51].

3.2. AI for WWTP Monitoring

IoT is forecasted to usher in the next phase of the information revolution into modern society, introducing transformations that will compete with the Internet as known today. As mentioned, the IoT has become increasingly popular for a wide range of applications, which include monitoring of water and wastewater quality. Platform devices such as Raspberry Pi or Arduino [52], as well as a novel class of online compact sensors measuring various parameters such as temperature, dissolved oxygen (DO), pH, etc., make it possible to continuously monitor some quality parameters in real time [53]. A novel type of surveillance-based sewage wastewater monitoring system (SSWMS) with IoT was recently proposed and discussed with the purpose of sewage system surveillance, wastewater treatment monitoring, and water quality improvement [54].
WWTPs often use online sensors for key parameter monitoring and basic automated process control, e.g., maintaining constant DO levels through rule-based blower control in response to real-time DO readings [55]; recently, a focus of operational interest has been moving towards real-time measurements of nutrients, GHG emissions, and parameters, leading to process optimization. A purposely built and programmed Arduino-based control system was developed by Daneshgar et al. [56] for online control of a pilot reactor for side-stream precipitation of P compounds from aerobic sludge.
Many studies provide operators with ample insight for expanding instrumentation beyond well-established approaches such as pH, DO, and ORP probes. These methods often employ measurement of surrogate parameters to determine, through correlations or specific inference models, the presence of non-directly detectable pollutants [57,58,59]. Online sensors may replace continuous mode measurements otherwise pertaining to the lab realm, e.g., absorbance TSS sensors as a surrogate for traditional MLSS analysis, or sensors for ammonium and nitrate to achieve more consistent effluent nitrogen and decrease aeration. An overview of current commercial online instrument capabilities is reported in Table 3.
Evolving regulations may soon include requirements for the detection of other pollutants, including metals and emerging contaminants, and the quantification of global warming compounds such as N2O and CH4 emissions that are not, or are poorly, covered by available commercial instrumentation. For this purpose, studies on different determination techniques are ongoing [62,63].

Biosensors and Soft Sensors

Biosensors rely on microorganisms or biological molecules to detect chemicals, or the onset of biotoxicity, with potential benefits including sensitivity, specificity, low cost, and fast response time [64,65]. A biosensor is an analytical device consisting of a bioreceptor able to recognize the target analyte in order to produce a signal. Bioelectrochemical biosensors offer a promising approach for real-time bioprocess monitoring: based on bioelectrochemical systems (BES) [66], they exploit the interaction between current-bearing electrodes and microorganisms, and in general, they do not require additional reagents for target molecules, simplifying the monitoring process and reducing its costs. Bioelectrochemical sensors use biological materials such as enzymes, antibodies, DNA, or bacterial cells as receptors to detect target analytes in a variety of samples [67]. Few biosensors are commercially available to date [68], with even fewer published reports on their deployment and long-term performance. Research on the expansion of the range of parameters that could be detected by biosensors is currently targeting N compounds, metals, and some emerging contaminants [69].
AI contribution to BES-based sensors is multifaceted: it can predict the development of microbial community, product, or substrate generation, as well as overall reactor performance [70]. In a first-of-a-kind study, coupling of BES-based sensors with data-driven biological nutrient removal (BNR) modeling, the ability to capture the complex dynamics of C uptake was demonstrated, with a successful development of advanced predictive models of an industrial-scale WWTP performance. While the developed ML models were not directly used for automatic process control, the insights offered by this methodology present realistic opportunities to develop feed-forward control for BNR optimization. In addition to obtaining high-frequency and uninterrupted data from BES without calibration and minimum maintenance requirements, integration of BES sensors and AI could lead to considerable operational cost savings, related to chemical and aeration requirements, and increased process resiliency [71]. The most challenging aspect for AI in biosensors is the elimination of the matrix effect of a sample, possibly leading to interferences in compounds determination [72].
Another AI contribution to online WWTP monitoring could come in the refinement of “soft sensors”, which essentially create surrogate data from multiple hardware sensors of various kinds to create a virtual model for target parameters estimation: accuracy and reliability of this class of “virtual” sensors have significantly improved with the advancement of AI techniques based on ANN, DT, RF, or DL [73,74,75]. As an alternative to laboratory testing that may take minutes to days to obtain a parameter reading, or to hardware sensors that require regular maintenance and reagent/part replacement and may be skewed by interference, biofilm build-up, and measurement drift or deterioration induced by the harsh WWTP environment, soft sensors have been developed to translate easy-to-measure parameters into more complex operating parameters. The use of a physical sensor in fact requires device calibration, i.e., a model (or formula) to convert the observed effect (e.g., resistance or absorbance) to the desired parameter (unless used solely for the detection of shifts in the target variable, e.g., “fingerprint” reading) [76]. Contrary to physical sensors, a virtual (soft) sensor produces signals entirely autonomously by aggregating and combining—through mathematical or empirical models—those received from other physical and/or other virtual ones (synchronously or asynchronously) [77,78]. A recent review considering a large number of applications concluded that ANNs are the most widely used models for soft sensor development, and they will likely remain the dominant approach in the coming years [73]. Among the soft sensors reported by the current literature, some are undergoing development, while others have met full-scale industrial applications. Physical input sensor readings utilized for soft sensor development typically include influent and effluent parameters, including TSS, COD, TP, TN, NH4, ORP, and DO, and estimated target parameters are primarily predictions of effluent properties [79]. Paepae et al. [80] summarized the state of the art of soft sensing of water quality parameters. Table 4 summarizes the target wastewater parameters, the originally monitored variables, and the inference method used. All these sensors are natively (or could be) compatible with IoT networks.
A case study of soft sensor application for the identification of activated sludge bulking in the Sitkówka-Nowiny WWTP was carried out, taking into account several aspects, including data-mining method complexity, time of analysis of independent variables in simulations, accessibility and reliability of recorded data, and model usefulness for process control and optimization. The results showed that a suitably selected simulation method enables the prediction of activated sludge bulking with high accuracy, starting from WWTP contaminant input and process operating parameters [85]. The proposed soft sensor model enabled on-line diagnosis of WWTP operation despite including in the tests unforeseen events such as failure of hardware sensors, measurements discontinuity, and errors in measurements.
A CNN model combined with laser-induced Raman and fluorescence spectroscopy (LIRFS) monitoring achieved real-time WWTP monitoring of micro-pollutants carbamazepine, naproxen, and tryptophan in the range of a few µg/L (correlation coefficient R2 = 0.74), showing that this method can lead to consistent measurements of compounds that cannot be monitored using standard WWTP monitoring methods [86].
An upcoming application in WWTP monitoring is provided by electronic noses (e-noses), instruments designed to detect mixtures of hundreds of odorant compounds and discriminate complex odors using an array of nonspecific chemical sensors. Many types of commercially available e-noses exist today with a wide range of applications in various industries, including the water sector, mostly for the evaluation of odor disturbances from WWTP gas emissions [87]. Gases emitted during wastewater treatment can contain up to 450 individual compounds, with ≈ 100 of them being strong odorants, and which concentration at the water surface is closely related to the pollutants present in the wastewater. Attempts to link standard wastewater parameters to e-nose responses were made using “gas fingerprints” that employ multiple physical sensors to identify each gas sample’s distinctive profile, using appropriate AI analysis of multidimensional data. AI-supported e-nose readouts were able to identify wastewater quality at different stages of the treatment process [88]. Failure of wastewater treatment processes could thus be related to the biological units’ odor emissions detected by e-noses [89].
The characterization of the biomass composition and status by microscopy inspection can provide important information concerning the operational conditions of WWTPs, as changes in floc structure can be induced by microbial community imbalances or the presence of specific pollutants [90]. Regular monitoring of biological flocs may therefore provide important information on the dynamic changes of wastewater quality and operational status; however, the acquisition of information from their microscopy inspection requires significant specialized human expert resources and time. Digital image processing and analysis offer an alternative to monitor the state of biological WWTPs and predict their future state. At present, such technologies are recognized as important monitoring tools, although they still require extensive image pre-processing and enhancing operations that are laborious and time-consuming and may induce instrumental bias. The characterization by image processing and analysis is done by correlating the time evolution of parameters extracted by image analysis of floc and filaments with the physico-chemical parameters [91]. AI-based techniques (image classification models) for imagery analysis have been studied to allow feature representations and extraction from raw visual microscopy data without any pre-processing. AI-CNN algorithms were trained on more than 1 million database images and used without any subsequent adaptation for the automatic recognition of floc characteristics [92]; a computer vision approach to assess floc settling characteristics based on microscopy images and transfer learning of deep CNNs was tested on an offline 2-year-long image dataset collected at a full-scale industrial WWTP [93]. The model could predict early signs of bulking, indicated by surges in SVI predictions in a real-world application. Research in AI-supported image analysis shows increasing importance to produce feasible and easy solutions for control of biological WWTP operations, including those aimed at resource recovery through bioprocesses [94].

3.3. AI for Fault and Anomaly Diagnosis in WWTP

Advances in AI and data analytics could provide opportunities for fault diagnosis, management, and decision making in urban WWTP operations. Already, to maintain desired performance, WWTPs are increasingly equipped with a number of online sensors to collect data and detect abnormalities, and sometimes also with automation systems capable of modifying process parameters in real time (with or without human supervision) [95]. Supervised control and data acquisition (SCADA) systems have been developed for WWTP control, ensuring that operations could be conducted and monitored remotely (even without local service personnel), securing intervention and adjustment of operating parameters. SCADA systems consist of several components: sensors and actuators; programmable logical controllers (PLCs); human–machine interfaces; supervisory computers; and communication networks. Generally, operators of SCADA-equipped WWTP highly appreciate the additional monitoring and control features that these represent in day-to-day operation. Their efficacy in supporting WWTP operations has been demonstrated globally for improving supervised plant control [96] and analyzing opportunities for energy requirement reduction, simulating efficiency improvement, and comparing design options for WWTP upgrade [97]. Operators familiar with SCADA use might be willing to add something to these systems in order to enhance their performance: AI implementation could provide such additional features.
WWTPs’ SCADA systems generate large data amounts, and thus the automatic detection of faults in the system, manifested by irregular collection of instances with respect to actual data trends, could be improved by using AI algorithms to process data, enabling facilities to maintain a high performance record by acting upon detected faults in a timely manner. An offline study investigated the use of DNN (long short-term memory) algorithms to capture temporal patterns in a real 5.1 million point dataset from WWTP sensor data, comparing results with those of traditional statistical methods. The system achieved a fault detection rate ˃ 92%, outperforming traditional methods and showing the potential to enable timely fault detection in real operating conditions [98].
An important limitation of this and similar anomaly detection studies is their ex-post nature and infrequent real-time validation. In many cases, they are only validated on a WWTP simulation model, or on pilot-scale plants, and their transferability is not demonstrated [99]. In a recent study, however, AI method (vector machine, multilayer perceptron, RF, LightGBM, XGBoost) performance in detecting and classifying plant anomalies in different WWTPs were compared through a rigorous validation procedure that included previously unknown datasets. Gradient boosting algorithms detected 96% of anomalies, with up to 84% and 62% of them correctly classified on different datasets [100].
As mentioned, IoT communication technology is used to transfer data from physical system to a device provided with intelligent analysis capability and as such can be considered, in a sense, an evolutionary feature of existing SCADA systems, aiming at improving their drawbacks (high cost, limited expansion capability and maintenance needs), as well as providing the basis for more dynamic and sophisticated approaches to smart wastewater management achievement.
Wireless sensor networks (WSN) have proven an effective technology for environmental monitoring applications: their use in wastewater monitoring is particularly appealing due to low sensor nodes’ cost and cost-effectiveness. Monitoring water quality through the development of microcontrollers and GSM modems has already been implemented in a few instances; however, the majority of past research on this topic lacks focus and clarity on how to make decisions and control, based on gathered sensor data. The reported examples are usually based on pre-existing third-party cloud platforms for data monitoring and analysis, without options for data exchange rerouting in the event of Wi-Fi connection failures, which makes these systems poorly robust [101].
The ability to communicate among different system entities is thus a fundamental aspect of SCADA systems; in order to optimize performance, they have thus evolved from isolated environments to interconnected networks. Despite this evolutionary advantage, a serious drawback lies in the fact that reliance on communication networks to transmit information could increase the possibility of intentional external attacks. Physical plants have thus become more vulnerable to cyberattacks, which may disrupt processes and impact public health and environmental safety. AI methods can be designed to analyze and detect ongoing cyberattacks on running SCADA systems [102].
An IoT system for autonomous monitoring and sampling of wastewater discharges in sewer mains was developed within the EU-H2020 Micromole project, consisting of a dedicated hardware architecture for wastewater monitoring, distributed real-time algorithms for anomaly detection, and the localization of harmful wastewater discharges. This prototype is currently at Technology Readiness Level (TRL) 7 according to European standard classification concerning R&D. A new IoT architecture was developed to overcome issues of unstable Internet connection in underground pipes, lack of power grid supply in sewage networks, and the need for triggering IoT-connected device actuators in real time [103].

3.4. Future Developments in Wastewater Treatment by AI Applications

Wastewater treatment generates large quantities of excess biological sewage sludge residuals, whose disposal has significant impact on the energy and economic balances (≈50% of each) of wastewater treatment operations and may contribute to almost 40% of GHG emissions related to wastewater processing; however, these same residuals contain large amounts of recoverable resources in terms of energy, nutrients, and other materials [9,104,105]. A variety of potential options for embedded resource recovery are available, ranging from process improvement and integration, to thermochemical technologies, to biorefinery, etc. Traditional recoverable resources, such as biogas from sludge digestion, are now complemented by liquid and solid energy end products (e.g., biodiesel, biochar); nutrients; and high-value-added products, such as polyhydroxyalkanoates (PHA), extracellular polymeric substances (EPS), and others [106]. Resource recovery requires well-designed treatment trains, i.e., sequences of unit processes that treat wastewater to the required compliance standard while recovering resources, in which each discreet unit process performance can affect that of the whole train. This in turn can be affected by short-/long-term influent quality and quantity changes. Sustainable recovery strategies should contemplate an assessment of environmental impact, carbon footprint minimization, and logistic and local conditions for the development of appropriate socioeconomic contexts. Facilities should be flexibly designed for resilient operation following possible variations in residuals and final product characteristics, as well as market demand for the latter. This requires processing of large amounts of heterogeneous data, and it could be supported by specific AI applications. Multi-criteria decision support systems, considering technological and non-technological aspects, were proposed to assist decision-makers to deal with such complexities, consisting of weighted multi-objective mixed integer nonlinear programming models for the selection of optimum process configurations [107].
A recent report analyzed over 100 studies in which various AI algorithms were applied to biowaste remediation and valorization. AI models were adopted to predict and classify feedstock characteristics and resulting fuels and product formation, as well as to predict volatile organic compound (VOC) content, waste generation, and accumulation rates by observing social indicators [108]. ML-AI technologies have the potential to revolutionize the way nutrients are recovered from wastewater streams by identifying intersections between nutrient influent data, the environment, farmers, stakeholders, and consumer and household trends, thus improving the circular economy’s performance and viability [109]. This type of application, going beyond the mere boundaries of the physical wastewater treatment facilities, could become decisive for the effective implementation of a wastewater-based circular economy.
The next foreseeable frontier of AI application in the water sector is through digital twins (DiTw). Generally speaking, a DiTw represents an object (i.e., a smart building) or a process (i.e., an industrial production process) as a detailed virtual replica in a modern interpretation of “as above, so below”, Hermes Trismegistus’ famous quote. DiTw is a comprehensive term representing various services and functions; its initial definition was provided in the “Gemini Principles”, a report setting out proposed principles to guide the UK’s national digital twin and the information management framework to enable it, as “a realistic digital representation of something physical in order to work using interoperative services in a suitable environment to understand, monitor, inform, optimize, or simulate an asset, system, process, or organization from planning to inception and throughout its full life cycle” [110]. A DiTw may be developed for a variety of purposes, operate at different scales, and adopt different modelling approaches while accurately reflecting its physical twin behavior under any circumstance, with the purposes of automatic fault diagnosis, run-time monitoring, and full process control, among others [111].
Once a detailed computer model sufficiently represents its real counterpart, it can be considered a digital twin of the system it represents. Examples of digital twins have already begun to appear within the built environment, as well as in many industrial sectors, serving a variety of purposes. However, a unique comprehensive definition concerning water industry applications has not been established yet [112]. AWWA, the American Water Works Association, defines DiTws as “A digital, dynamic system of real-world entities and their behaviors using models with static and dynamic data that enable insights and interactions to drive actionable and optimized outcomes”; this definition could well be extended to the wastewater field.
In fact, such systems already exist in an embryonic form at many utilities today: the ongoing digital transformation has developed tools to measure, monitor, and simulate portions of these system and their associated infrastructure. SCADA or other model systems, as argued previously, are less developed forms of DiTw that can evolve into more functionally complete applications though stepwise integration. According to AWWA, several levels can be identified: Level 0 (Digital Twin Ready) consists of existing models or systems performing more complex functions than those originally intended functions; Level 1 (Informational Twin) connects multiple systems and data sources into an interaction enavling environment; Level 2 (Operational Twin) uses real-time data stream analytics to enhance operational functions; Level 3 (Connected Twin) consists of virtual representations of connected assets (WiFi, IoT) that use real-time data analytics to coordinate functions across different domains [113].
A DiTw system is characterized by reciprocal physical-to-virtual and virtual-to-physical connections, allowing the physical object to benefit from the corresponding DiTw’s output (Figure 2), as recently discussed by Torfs et al. [114]. Both AI and DiTw applications require large amounts of data to function, and therefore their integration is not only logical but ineluctable. Integrated AI-DiTw systems have been proposed in many application domains, where the AI component can make predictions based on data flowing from the DiTw through the IoT.
In the wastewater sector, DiTw applications have been proposed for the representation and improvement of individual unit processes, such as secondary settling tanks, biological aerated filters, and primary clarifiers at a theoretical level. Advances in DiTw have seen the use of ML to optimize the operation of equipment such as pumps, air blowers, and mixers, with fallback in real-world applications [115]. However, the full development of WWTP DiTws poses huge technical challenges, requiring advanced monitoring and control systems able to handle large numbers of variables and parameters.
So far, there are only a few reported developments of full-scale operational WWTP DiTws. One such example is that of a facility in Eindhoven (the Netherlands) [116]. Eindhoven’s WWTP DiTw was developed with the active involvement of the facility’s engineers, making it a valuable and user-friendly interactive tool for decision making with long-term reliability. It includes fully automatically validated data collection, a detailed mechanistic model of the of the aeration and anoxic bioreactors’ processes, and a user interface co-created with the plant’s operators. The DiTw runs quasi-real-time simulations every 2 h with the possibility to simulate hypothetical scenarios, such as process and equipment failures and changes in controller settings.
Another DiTw for Gothenburg’s regional sewage system and WWTP was developed to allow reactive optimized real-time control of the facility’s inlet pumping station using the sewerage system’s DiTw flow forecasts for optimizing flow pumped to the WWTP, as well as to minimize untreated overflows [117]. In this case, the DiTw concerns just the sewer connector network and not the WWTP itself, being able to predictively quantify and optimize CSO reductions and inflows to the facility, minimizing pollutant loads on local catchments, according to predefined objectives.

4. Discussion

AI introduction in WWTP operation could be a significant step forward compared to the current mainstream operational practice, even in relationship to state-of-the-art SCADA systems. In general, however, this technology is still perceived at the potential level, rather than as a mature possibility. Even though many applications were developed and implemented in other industrial fields [118], they are not easily adoptable in urban wastewater systems, mainly due to a significant stochasticity/uncertainty of inputs’ forcing functions, subject to the occurrence of extreme events; the combination of biological, physical, and chemical reactions, themselves dependent on external influences; and the resulting superior complexity of these systems compared to the relative simplicity of engineered industrial processes.
The perception of control technology itself still differs substantially between municipal WWTPs (and within them, between small and large facilities) and process industries: many differences in day-by-day process control, operation, and maintenance approaches often originate from fundamental differences between private and public sector operating environments, despite the fact that in most countries, the water sector has been privatized, whether effectively or nominally. Historically, the water sector has been very conservative, risk averse, and slower to adopt and disseminate new technologies: as pointed out some years back, its current underlying engineering paradigm is still based on a technical consensus dating to the late 19th century, which still hinders the development of possible, more efficient approaches to present issues and challenges [119]. The sector, furthermore, relies on extensive infrastructure for its operation, which is not easy to substitute and may require substantial revamping in order to allow installation of modern technology. In addition, general underfunding; lack of resources, both human and financial; and constraints on revenue raising affect the sector [120].
In standard small-scale WWTPs, classical operational parameter measurements are conducted manually and sporadically (daily or at longer intervals), and staff often rely on personal experience for processes control. Small plant operators often lack specific technical expertise in the ICA/AI domains and tend to be skeptical towards automated control approaches. In addition, particularly at smaller facilities, the cost of online instrumentation installation and management is still perceived as too high, compared to the potential benefits deriving from it: a 2018 survey among 90 WWTP companies in Belgium showed that only dissolved oxygen and pH sensors were widely applied (in 96% and 69% of the plants, respectively), despite the availability of many other reasonable priced, established commercial sensors [121].
Modern, large WWTPs are often monitored by automated, connected sensors that can provide high frequency (minutes/seconds) measurements of some parameters (see Table 3), while others are typically measured manually every day (e.g., SVI) or at a few months’ intervals (e.g., disinfection byproducts). In general, data collection frequency should be sufficiently high to track irregularities and account for instrument noise, but not so high to require excessive analytical or computational power for processing. Short-term random events, like clogging of a pipe or blower shutdown, for example, require different monitoring windows than long-term trends such as TMP increase due to membrane fouling. Merging and interpreting data of different frequencies and formats is difficult: a common approach is to downscale data to the lowest time interval; however, datasets with very large frequency differences cannot be treated with this method because of the time-dependent, co-correlated, and intrinsic nonlinearity of measured data, which makes downscaling unreliable, with potential erroneous attribution of changes’ causes between samplings. The lack of deterministic components in AI computation tools may induce biases in systems that are typically subject to a higher range of disturbance and uncertainty, as well as lower monitoring capabilities than industrial processes.
A major obstacle in AI application in WWTP is in fact related to data. Correct, validated data are essential for improving sewer network efficiency and wastewater treatment reliability. Several factors contribute to the lack of reliable WWTP data, including
-
old infrastructure: older WWTPs often lack modern monitoring and data collection systems. This can result in sparse, outdated or incomplete data, as these systems may not accurately capture all relevant information.
-
human factors: manual data entry and processing is prone to errors, which can significantly affect the overall quality of the data. Human errors can occur at various stages, from data collection to reporting to manual instrument calibration, leading to inaccuracies and gaps.
-
sensing technology: even with regular maintenance and calibration, sensor measurements may drift randomly and differently over time, even between same-type sensors exposed to similar environmental conditions. It is not uncommon to experience issues with sensors: examples include failure, calibration difficulties, fouling and blockage, connection problems between sensors, actuators, and the data management system.
-
lack of data processing: even in large WWTPs equipped with sensors and SCADA systems, operators rarely monitor the total amount of collected data or have sufficient computer processing capacity to collect, sort, clean, analyze, and interpret them quickly and effectively.
It should be pointed out that very similar issues were raised in a survey concerning the adoption of AI in US drinking water operations [122].
Nevertheless, in the few actual applications to WWTPs, AI has so far provided better results in terms of data analysis than conventional methods [36] since it can detect complex relationships through the analysis of large datasets. However, most data-driven tools commercially available are designed as “black-box”, turnkey solutions with limited insight into actual processes and their causal factors. An operator receiving information from multiple sources (e.g., laboratory analyses, online sensors, operations and maintenance management notes, and technology manufacturer information) is used to make an educated decision on the course to follow; black-box systems may not differentiate between numerical or categorical variables important for data-driven analysis, and they are frequently mistrusted by WWTP operators, who may also lack a data science background to apply these tools properly. AI techniques to improve data validation and prevent misinterpretation are being investigated and should strengthen operators’ confidence in these systems [99,102].
The development of specific AI-based automated control systems using IoT data to dynamically adapt WW treatment processes would constitute an important advance in the field, with the aim to optimize efficiency and reduce manual intervention and human error, eventually resulting in cost savings and lower process upsets, as well as improving energy optimization and resource recovery. Often, however, the design of existing facilities does not allow for the ready implementation of real-time control. In process industries, incentives for resource efficiency are implemented more systematically, and more effort is made to prevent process failure and costly production downtime, resulting in more rapid introduction of innovative schemes [123].

5. Conclusions

Although in many ways the advent of AI has the potential to entirely change water utilities’ operation and design of their facilities, successful full-scale applications in the sector are still limited, consisting mostly of on- or offline operational support and limited real-time control. So far, almost all applications in the sector involve predictive studies, often at a small scale or with limited data use. Identified key barriers include laborious instrumentation maintenance, lack of process expertise in the design of current software, instability of control loops, and insufficient incentives for resource efficiency. Deficits in data collection and processing also negatively impact the perceived cost–benefit of AI.
The introduction of AI technologies to replace traditional modelling methods is a consolidating trend in many different fields, due to their observed reliability and rapid response in many circumstances. Some caveats must be kept in mind, though: AI models based on small datasets may fail to achieve the desired objectives and accuracy; on the other hand, training processes can be computationally expensive when it comes to large datasets. The water community traditionally favors mechanistic or knowledge-based approaches, however simplified in terms of incorporated process knowledge. Being of a “black-box” nature, AI models do not expose correlations mechanisms between input and output variables, and may this be not considered acceptable by operators.
From an examination of the state of the art of AI applications developed so far in the wastewater sector, the following conclusions can be drawn:
  • Almost all AI applications in the sector involve predictions of some sort, such as influent or effluent quality, energy consumption, mass flow, or specific process variables and parameters like aeration time, sludge bulking, and settleability.
  • Several algorithms, ANNs being the most popular, have been tested using real WWTP data for that purpose. Comparisons among algorithms in different studies provide initially encouraging results.
  • Soft sensors are becoming increasingly important in WWTP operation, with most recent applications (at any scale) involving soft sensors. Direct field application of soft sensors with AI control would provide more meaningful information than mere predictive-model-based sensors.
  • Online hard sensors provide reliable information supporting WWTP operation, especially for the detection of faults or anomalies. A large number of data points at a high frequency is needed for building reliable data-driven AI models.
  • AI applications for online image analysis (e.g., identification of biomass) and implementation of WWTPs’ DiTws requires further investigation.
  • WWTP DiTw implementation, although very limited at the moment, is a promising research area that could generate tools to assist in facility revamping and optimization.
The following recommendations could be made for the future:
  • Solutions should be designed with an understanding of the specific processes involved in WWTPs, along with improved data sharing.
  • AI algorithm selection should take into account existing WWTP process knowledge; further field experience in AI application development is needed, especially requiring collaboration between wastewater engineers and operators and computer domain researchers.
  • Online and soft sensors should be combined in online learning and training, allowing AI models to more efficiently interpret real-time data.
  • AI imaging and olfactory analysis of biological and water quality parameters, biomass, and sludge should be developed to help optimize data gathering and, consequently, treatment processes operation and maintenance.
  • Further study on water systems’ DTWs building and implementation is needed before they could become an operational tool.

Author Contributions

Conceptualization, A.G.C. and A.C.; investigation, A.G.C. and A.C.; writing—original draft preparation, A.G.C. and A.C.; writing—review and editing, A.G.C. All authors have read and agreed to the published version of the manuscript.

Funding

This research received no external funding.

Data Availability Statement

No new data were created.

Conflicts of Interest

The authors declare no conflicts of interest.

References

  1. Zhang, C.; Lu, Y. Study on artificial intelligence: The state of the art and future prospects. J. Ind. Inf. Integr. 2021, 23, 100224. [Google Scholar] [CrossRef]
  2. Capodaglio, A.G.; Zheng, S.; Novotny, V.; Feng, X. Stochastic system identification of sewer-flow models. J. Environ. Eng. 1990, 116, 284–298. [Google Scholar] [CrossRef]
  3. Novotny, V.; Jones, H.; Feng, X.; Capodaglio, A. Time series analysis models of activated sludge plants. Water Sci. Technol. 1991, 23, 1107–1116. [Google Scholar] [CrossRef]
  4. Harremoes, P.; Capodaglio, A.G.; Hellstrom, B.G.; Otterpohl, R.; Soeberg, H. Wastewater treatment plants under transient loading- Performance, modelling and control. Water Sci. Technol. 1993, 27, 71–115. [Google Scholar] [CrossRef]
  5. Rossi, L.; Chèvre, N.; Fankhauser, R.; Krejci, V. Probabilistic environmental risk assessment of urban wet-weather discharges: An approach developed for Switzerland. Urban Water J. 2009, 6, 355–367. [Google Scholar] [CrossRef]
  6. Capodaglio, A.G. In-stream detection of waterborne priority pollutants, and applications in drinking water contaminant warning systems. Water Sci. Technol. 2017, 17, 707–725. [Google Scholar] [CrossRef]
  7. Kim, J.; Xin, X.; Mamo, B.T.; Hawkins, G.L.; Li, K.; Chen, Y.; Huang, K.; Huang, C.H. Occurrence and Fate of Ultrashort-Chain and Other Per- and Polyfluoroalkyl Substances (PFAS) in Wastewater Treatment Plants. ACS EST Water 2022, 2, 1380–1390. [Google Scholar] [CrossRef]
  8. Capodaglio, A.G. High-energy oxidation process: An efficient alternative for wastewater organic contaminants removal. Clean. Technol. Environ. Policy 2017, 19, 1995–2006. [Google Scholar] [CrossRef]
  9. Capodaglio, A.G.; Olsson, G. Energy issues in sustainable urban wastewater management: Use, demand reduction and recovery in the urban water cycle. Sustainability 2020, 12, 266. [Google Scholar] [CrossRef]
  10. Capodaglio, A.G. Energy use and decarbonisation of the water sector: A comprehensive review of issues, approaches and technological options. Environ. Technol. Rev. 2025, 14, 40–68. [Google Scholar] [CrossRef]
  11. Zhang, S.; Jin, Y.; Chen, W.; Wang, J.; Wang, Y.; Ren, H. Artificial intelligence in wastewater treatment: A data-driven analysis of status and trends. Chemosphere 2023, 336, 139163. [Google Scholar] [CrossRef]
  12. Goldman Sachs. AI Is Poised to Drive 160% Increase in Data Center Power Demand. Available online: https://www.goldmansachs.com/insights/articles/AI-poised-to-drive-160-increase-in-power-demand (accessed on 30 August 2024).
  13. U.S. Department of Energy. The Water-Energy Nexus: Challenges and Opportunities; US DOE: Washington, DC, USA, 2014.
  14. NPR. Data Centers, Backbone of the Digital Economy, Face Water Scarcity and Climate Risk. Available online: https://www.npr.org/2022/08/30/1119938708/data-centers-backbone-of-the-digital-economy-face-water-scarcity-and-climate-ris (accessed on 30 August 2024).
  15. The University of Tulsa. Data Centers Draining Resources in Water-Stressed Communities. Available online: https://utulsa.edu/news/data-centers-draining-resources-in-water-stressed-communities/ (accessed on 28 December 2024).
  16. Statista. Leading Countries by Number of Data Centers as of March 2024. Available online: https://www.statista.com/statistics/1228433/data-centers-worldwide-by-country/ (accessed on 28 December 2024).
  17. Klipa, D.; Ristić, I.; Radonjić, A.; Scepanović, I. Big data and artificial intelligence. Int. J. Manag. Trends Key Concepts Res. 2022, 1, 3–14. [Google Scholar] [CrossRef]
  18. Alzahrani, A.I.A.; Chauhdary, S.H.; Alshdadi, A.A. Internet of Things (IoT)-Based Wastewater Management in Smart Cities. Electronics 2023, 12, 2590. [Google Scholar] [CrossRef]
  19. Aquatech. Tech Dive: The Rise of the Smart Pump. 2020. Available online: https://www.aquatechtrade.com/news/water-treatment/tech-dive-rise-of-the-smart-pump (accessed on 15 June 2023).
  20. Kotsiantis, S.B.; Zaharakis, I.D.; Pintelas, P.E. Machine learning: A review of classification and combining techniques. Artif. Intell. Rev. 2006, 26, 159–190. [Google Scholar] [CrossRef]
  21. Zhao, Z.Q.; Zheng, P.; Xu, S.T.; Wu, X. Object detection with deep learning: A review. IEEE Trans. Neural Netw. Learn. Syst. 2019, 30, 3212–3232. [Google Scholar] [CrossRef]
  22. Wang, Y.; Cheng, Y.; Liu, H.; Guo, Q.; Dai, C.; Zhao, M.; Liu, D. A Review on Applications of Artificial Intelligence in Wastewater Treatment. Sustainability 2023, 15, 13557. [Google Scholar] [CrossRef]
  23. Capodaglio, A.G. Urban Wastewater Mining for Circular Resource Recovery: Approaches and Technology Analysis. Water 2023, 15, 3967. [Google Scholar] [CrossRef]
  24. Capodaglio, A.G.; Callegari, A. Energy and resources recovery from excess sewage sludge: A holistic analysis of opportunities and strategies. Resour. Conserv. Recycl. Adv. 2023, 19, 200184. [Google Scholar] [CrossRef]
  25. Nemcik, J.; Krupa, P.; Ozana, S.; Slanina, Z. Wastewater Treatment Modeling Methods Review. IFAC-PapersOnLine 2022, 55, 195–200. [Google Scholar] [CrossRef]
  26. Olsson, G. ICA and me—A subjective review. Water Res. 2012, 46, 1585–1624. [Google Scholar] [CrossRef]
  27. Yuan, Z.; Olsson, G.; Cardell-Oliver, R.; van Schagen, K.; Marchi, A.; Deletic, A.; Urich, K.; Rauch, W.; Liu, Y.; Jiang, G. Sweating the assets—The role of instrumentation, control and automation in urban water systems. Water Res. 2019, 155, 381–402. [Google Scholar] [CrossRef] [PubMed]
  28. Capodaglio, A.G.; Jones, H.V.; Novotny, V.; Feng, X. Sludge bulking analysis and forecasting: Application of system identification and artificial neural computing technologies. Water Res. 1991, 25, 1217–1224. [Google Scholar] [CrossRef]
  29. Ráduly, B.; Gernaey, K.V.; Capodaglio, A.G.; Mikkelsen, P.S.; Henze, M. Artificial neural networks for rapid WWTP performance evaluation: Methodology and case study. Environ. Model. Softw. 2007, 22, 1208–1216. [Google Scholar] [CrossRef]
  30. Mehrani, M.J.; Bagherzadeh, F.; Zheng, M.; Kowal, P.; Sobotka, D.; Mąkinia, J. Application of a hybrid mechanistic/machine learning model for prediction of nitrous oxide (N2O) production in a nitrifying sequencing batch reactor. Process Saf. Environ. Prot. 2022, 162, 1015–1024. [Google Scholar] [CrossRef]
  31. Hazrati, H.; Moghaddam, A.H.; Rostamizadeh, M. The influence of hydraulic retention time on cake layer specifications in the membrane bioreactor: Experimental and artificial neural network modeling. J. Environ. Chem. Eng. 2017, 5, 3005–3013. [Google Scholar] [CrossRef]
  32. Khataee, A.R.; Kasiri, M.B. Artificial neural networks modeling of contaminated water treatment processes by homogeneous and heterogeneous nanocatalysis. J. Mol. Catal. A Chem. 2010, 331, 86–100. [Google Scholar] [CrossRef]
  33. Podder, M.S.; Majumder, C.B. The use of artificial neural network for modelling of phycoremediation of toxic elements As(III) and As(V) from wastewater using Botryococcus braunii. Spectrochim. Acta A Mol. Biomol. Spectrosc. 2016, 155, 130–145. [Google Scholar] [CrossRef]
  34. Bagherzadeh, F.; Nouri, A.S.; Mehrani, M.J.; Thennadil, S. Prediction of energy consumption and evaluation of affecting factors in a full-scale WWTP using a machine learning approach. Proc. Saf. Environ. Prot. 2021, 154, 458–466. [Google Scholar] [CrossRef]
  35. Bhagat, S.K.; Pilario, K.E.; Babalola, O.E.; Tiyasha, T.; Yaqub, M.; Onu, C.E.; Pyrgaki, K.; Falah, M.W.; Jawad, A.H.; Yaseen, D.A.; et al. Comprehensive review on machine learning methodologies for modeling dye removal processes in wastewater. J. Clean. Prod. 2023, 385, 135522. [Google Scholar] [CrossRef]
  36. Ye, Z.; Yang, J.; Zhong, N.; Tu, X.; Jia, J.; Wang, J. Tackling environmental challenges in pollution controls using artificial intelligence: A review. Sci. Total Environ. 2020, 699, 134279. [Google Scholar] [CrossRef]
  37. Oliveira, P.; Fernandes, B.; Aguiar, F.; Pereira, M.A.; Analide, C.; Novais, P. A Deep Learning Approach to Forecast the Influent Flow in Wastewater Treatment Plants. In Intelligent Data Engineering and Automated Learning—IDEAL 2020; Springer International Publishing: Cham, Switzerland, 2020; pp. 362–373. [Google Scholar]
  38. Arismendy, L.; Cardenas, C.; Gomez, D.; Maturana, A.; Mejia, R.; Quintero, M.C. Intelligent System for the Predictive Analysis of an Industrial Wastewater Treatment Process. Sustainability 2020, 12, 6348. [Google Scholar] [CrossRef]
  39. Deepnarain, N.; Nasr, M.; Kumari, S.; Stenström, T.A.; Reddy, P.; Pillay, K.; Bux, F. Artificial intelligence andmultivariate statistics for comprehensive assessment of filamentous bacteria in wastewater treatment plants experiencing sludge bulking. Environ. Technol. Innov. 2020, 19, 100853. [Google Scholar] [CrossRef]
  40. Karam, A.; Mostafa, M.; Elawwad, A.; Mahmoud, A.; Peters, R. Semi-Pilot Plant for Tertiary Treatment of Domestic Wastewater using Algal Photo-Bioreactor, with Artificial Intelligence. In Proceedings of the 2019 Annual AIChE Meeting, Orlando, FL, USA, 10–15 November 2019. [Google Scholar]
  41. Kern, P.; Wolf, C.; Gaida, D.; Bongards, M.; McLoone, S. COD and NH4-N estimation in the inflow of Wastewater Treatment Plants using Machine Learning Techniques. In Proceedings of the 2014 IEEE International Conference on Automation Science and Engineering (CASE), New Taipei, Taiwan, 18–22 August 2014; pp. 812–817. [Google Scholar]
  42. Mei, R.; Kim, J.; Wilson, F.P.; Bocher, B.T.W.; Liu, W.-T. Coupling growth kinetics modeling with machine learning reveals microbial immigration impacts and identifies key environmental parameters in a biological wastewater treatment process. Microbiome 2019, 7, 65. [Google Scholar] [CrossRef]
  43. Mercier, T.; Dembele, A.; Denoeux, T.; Blanc, P. Machine learning as a decision support tool for waste water treatment plant operation. Water Resour. Manag. X 2019, 229, 103–107. [Google Scholar]
  44. Newhart, K.B.; Marks, C.A.; Rauch-Williams, T.; Cath, T.Y.; Hering, A.S. Hybrid statistical-machine learning ammonia forecasting in continuous activated sludge treatment for improved process control. J. Water Process. Eng. 2020, 37, 101389. [Google Scholar] [CrossRef]
  45. Niu, G.; Yi, X.; Chen, C.; Li, X.; Han, D.; Yan, B.; Huang, M.; Ying, G. A novel effluent quality predicting model based on genetic-deep belief network algorithm for cleaner production in a full-scale paper-making wastewater treatment. J. Clean. Prod. 2020, 265, 121787. [Google Scholar] [CrossRef]
  46. Oulebsir, R.; Lefkir, A.; Safri, A.; Bermad, A. Optimization of the energy consumption in activated sludge process using deep learning selective modeling. Biomass Bioenergy 2020, 132, 105420. [Google Scholar] [CrossRef]
  47. Zaghloul, M.S.; Achari, G. Application of machine learning techniques to model a full-scale wastewater treatment plant with biological nutrient removal. J. Environ. Chem. Eng. 2022, 10, 107430. [Google Scholar] [CrossRef]
  48. Liu, Y.; Guo, J.; Wang, Q.; Huang, D. Prediction of filamentous sludge bulking using a state-based gaussian processes regression model. Sci. Rep. 2016, 6, 31303. [Google Scholar] [CrossRef]
  49. Shao, S.; Fu, D.; Yang, T.; Mu, H.; Gao, Q.; Zhang, Y. Analysis of Machine Learning Models for Wastewater Treatment Plant Sludge Output Prediction. Sustainability 2023, 15, 13380. [Google Scholar] [CrossRef]
  50. Khalil, M.; AlSayed, A.; Liu, Y.; Vanrolleghem, P.A. Machine learning for modeling N2O emissions from wastewater treatment plants: Aligning model performance, complexity, and interpretability. Water Res. 2023, 245, 120667. [Google Scholar] [CrossRef] [PubMed]
  51. Ekinci, E.; Özbay, B.; Omurca, S.I.; Sayın, F.E.; Özbay, I. Application of machine learning algorithms and feature selection methods for better prediction of sludge production in a real advanced biological wastewater treatment plant. J. Environ. Manag. 2023, 348, 119448. [Google Scholar] [CrossRef] [PubMed]
  52. Minchala, L.I.; Peralta, J.; Mata-Quevedo, P.; Rojas, J. An Approach to Industrial Automation Based on Low-Cost Embedded Platforms and Open Software. Appl. Sci. 2020, 10, 4696. [Google Scholar] [CrossRef]
  53. Miller, M.; Kisiel, A.; Cembrowska-Lech, D.; Durlik, I.; Miller, T. IoT in Water Quality Monitoring-Are We Really Here? Sensors 2023, 23, 960. [Google Scholar] [CrossRef]
  54. Kumar, P.M.; Hong, C.S. Internet of things for secure surveillance for sewage wastewater treatment systems. Environ. Res. 2022, 203, 111899. [Google Scholar] [CrossRef]
  55. Revollar, S.; Vilanova, R.; Vega, P.; Francisco, M.; Meneses, M. Wastewater Treatment Plant Operation: Simple Control Schemes with a Holistic Perspective. Sustainability 2020, 12, 768. [Google Scholar] [CrossRef]
  56. Daneshgar, S.; Cecconet, D.; Capsoni, D.; Capodaglio, A.G. Side-Stream Phosphorus Recovery in Activated Sludge Processes. Water 2022, 14, 1861. [Google Scholar] [CrossRef]
  57. Copetti, D.; Marziali, L.; Viviano, G.; Valsecchi, L.; Guzzella, L.; Capodaglio, A.G.; Tartari, G.; Polesello, S.; Valsecchi, S.; Mezzanotte, V.; et al. Intensive monitoring of conventional and surrogate quality parameters in a highly urbanized river affected by multiple combined sewer overflows. Water Sci. Technol. Water Supp. 2019, 19, 953–966. [Google Scholar] [CrossRef]
  58. Viviano, G.; Valsecchi, S.; Polesello, S.; Capodaglio, A.; Tartari, G.; Salerno, F. Combined use of caffeine and turbidity to evaluate the impact of CSOs on river water quality. Water Air Soil. Policy 2017, 228, 330. [Google Scholar] [CrossRef]
  59. Capodaglio, A.G.; Callegari, A.; Molognoni, D. Online monitoring of priority and dangerous pollutants in natural and urban waters: A state-of-the-art review. Manag. Environ. Qual. Int. J. 2016, 27, 507–536. [Google Scholar] [CrossRef]
  60. S::CAN. Wastewater Monitoring. Available online: https://www.s-can.at/en/ (accessed on 3 September 2024).
  61. Copetti, D.; Valsecchi, L.; Capodaglio, A.G.; Tartari, G. Direct measurement of nutrient concentrations in freshwaters with a miniaturized analytical probe: Evaluation and validation. Environ. Monit. Assess. 2017, 189, 144. [Google Scholar] [CrossRef] [PubMed]
  62. Zhang, Y.; Wang, Z.; Hu, J.; Pu, C. Intelligent management of carbon emissions of urban domestic sewage based on the Internet of Things. Environ. Res. 2024, 251, 118594. [Google Scholar] [CrossRef] [PubMed]
  63. Asadi, M.; McPhedran, K.N. Greenhouse gas emission estimation from municipal wastewater using a hybrid approach of generative adversarial network and data-driven modelling. Sci. Total Environ. 2021, 800, 149508. [Google Scholar] [CrossRef] [PubMed]
  64. Ali, J.; Najeeb, J.; Ali, M.A.; Aslam, M.F.; Raza, A. Biosensors: Their fundamentals, designs, types and most recent impactful applications: A review. J. Biosens. Bioelectron. 2017, 8, 1–9. [Google Scholar] [CrossRef]
  65. Bindu, A.; Bhadra, S.; Nayak, S.; Khan, R.; Prabhu, A.A.; Sevda, S. Bioelectrochemical biosensors for water quality assessment and wastewater monitoring. Open Life Sci. 2024, 19, 20220933. [Google Scholar] [CrossRef]
  66. Sevda, S.; Garlapati, V.K.; Naha, S.; Sharma, M.; Ray, S.G.; Sreekrishnan, T.R.; Goswami, P. Biosensing capabilities of bioelectrochemical systems towards sustainable water streams: Technological implications and future prospects. J. Biosci. Bioeng. 2020, 129, 647–656. [Google Scholar] [CrossRef]
  67. Shanbhag, M.M.; Manasa, G.; Mascarenhas, R.J.; Mondal, K.; Shetti, M.P. Fundamentals of bio-electrochemical sensing. Chem. Eng. J. Adv. 2023, 16, 100516. [Google Scholar] [CrossRef]
  68. Ejeian, F.; Etedali, P.; Mansouri-Tehrani, H.A.; Soozanipour, A.; Low, X.Z.; Asadnia, M.; Taheri-Kafrani, A.; Razmjou, A. Biosensors for wastewater monitoring: A review. Biosens. Bioelectron. 2018, 118, 66–79. [Google Scholar] [CrossRef]
  69. Zhang, W.; Tooker, N.B.; Mueller, A.V. Enabling wastewater treatment process automation: Leveraging innovations in real-time sensing, data analysis, and online controls. Environ. Sci. Water Res. Technol. 2020, 6, 2973–2992. [Google Scholar] [CrossRef]
  70. Li, C.; Guo, D.; Dang, Y.; Sun, D.; Li, P. Application of artificial intelligence-based methods in bioelectrochemical systems: Recent progress and future perspectives. J. Environ. Manag. 2023, 344, 118502. [Google Scholar] [CrossRef]
  71. Emaminejad, S.A.; Sparks, J.; Cusick, R.D. Integrating Bio-Electrochemical Sensors and Machine Learning to Predict the Efficacy of Biological Nutrient Removal Processes at Water Resource Recovery Facilities. Environ. Sci. Technol. 2023, 57, 18372–18381. [Google Scholar] [CrossRef] [PubMed]
  72. Voitechovič, E.; Pauliukaite, R. Electrochemical multisensor systems and arrays in the era of artificial intelligence. Curr. Opin. Electrochem. 2023, 42, 101411. [Google Scholar] [CrossRef]
  73. Ching, P.M.L.; So, R.H.Y.; Morck, T. Advances in soft sensors for wastewater treatment plants: A systematic review. J. Water Process. Eng. 2021, 44, 102367. [Google Scholar] [CrossRef]
  74. Wang, G.; Jia, Q.-S.; Zhou, M.; Bi, J.; Qiao, J.; Abusorrah, A. Artificial neural networks for water quality soft-sensing in wastewater treatment: A review. Artif. Intell. Rev. 2022, 55, 565–587. [Google Scholar] [CrossRef]
  75. Alvi, M.; Batstone, D.; Mbamba, C.K.; Keymer, P.; French, T.; Ward, A.; Dwyer, J.; Cardell-Oliver, R. Deep learning in wastewater treatment: A critical review. Water Res. 2023, 245, 120518. [Google Scholar] [CrossRef]
  76. Wu, J.; Pons, M.N.; Potier, O. Wastewater fingerprinting by UV-visible and synchronous fluorescence spectroscopy. Water Sci. Technol. 2006, 53, 449–456. [Google Scholar] [CrossRef]
  77. Martin, D.; Kü, N.; Satzger, G. Virtual Sensors. Bus. Inf. Syst. Eng. 2021, 63, 315–323. [Google Scholar] [CrossRef]
  78. Perera, Y.S.; Ratnaweera, D.A.A.C.; Dasanayaka, C.H.; Abeykoon, C. The role of artificial intelligence-driven soft sensors in advanced sustainable process industries: A critical review. Eng. Appl. Artif. Intell. 2023, 121, 105988. [Google Scholar] [CrossRef]
  79. Moretti, A.; Ivan, H.L.; Skvaril, J. A review of the state-of-the-art wastewater quality characterization and measurement technologies. Is the shift to real-time monitoring nowadays feasible? J. Water Process. Eng. 2024, 60, 105061. [Google Scholar] [CrossRef]
  80. Paepae, T.; Bokoro, P.N.; Kyamakya, K. From Fully Physical to Virtual Sensing for Water Quality Assessment: A Comprehensive Review of the Relevant State-of-the-Art. Sensors 2021, 21, 6971. [Google Scholar] [CrossRef]
  81. Harrison, J.W.; Lucius, M.A.; Farrell, J.L.; Eichler, L.W.; Relyea, R.A. Prediction of stream nitrogen and phosphorus concentrations from high-frequency sensors using Random Forests Regression. Sci. Total Environ. 2021, 763, 143005. [Google Scholar] [CrossRef] [PubMed]
  82. Pattnaik, B.S.; Pattanayak, A.S.; Udgata, S.K.; Panda, A.K. Machine learning based soft sensor model for BOD estimation using intelligence at edge. Complex. Intell. Syst. 2021, 7, 961–976. [Google Scholar] [CrossRef]
  83. Pattanayak, A.S.; Pattnaik, B.S.; Udgata, S.K.; Panda, A.K. Development of Chemical Oxygen on Demand (COD) Soft Sensor Using Edge Intelligence. IEEE Sens. J. 2020, 20, 14892–14902. [Google Scholar] [CrossRef]
  84. Castrillo, M.; García, Á.L. Estimation of high frequency nutrient concentrations from water quality surrogates using machine learning methods. Water Res. 2020, 172, 115490. [Google Scholar] [CrossRef]
  85. Szelag, B.; Drewnowski, J.; Lagod, G.; Majerek, D.; Dacewicz, E.; Fatone, F. Soft sensor application in identification of the activated sludge bulking considering the technological and economical aspects of smart systems functioning. Sensors 2020, 20, 1941. [Google Scholar] [CrossRef]
  86. Post, C.; Heyden, N.; Reinartz, A.; Foerderer, A.; Bruelisauer, S.; Linnemann, V.; Hug, W.; Amann, F. Possibilities of Real Time Monitoring of Micropollutants in Wastewater Using Laser-Induced Raman & Fluorescence Spectroscopy (LIRFS) and Artificial Intelligence (AI). Sensors 2022, 22, 4668. [Google Scholar] [CrossRef]
  87. Prudenza, S.; Panzitta, A.; Bax, C.; Capelli, L. Electronic Nose for Real-time Monitoring of Odour Emissions at a Wastewater Treatment Plant. Chem. Eng. Trans. 2022, 95, 169–174. [Google Scholar] [CrossRef]
  88. Łagód, G.; Duda, S.M.; Majerek, D.; Szutt, A.; Dołhańczuk-Śródka, A. Application of Electronic Nose for Evaluation of Wastewater Treatment Process Effects at Full-Scale WWTP. Processes 2019, 7, 251. [Google Scholar] [CrossRef]
  89. Piłat-Rożek, M.; Dziadosz, M.; Majerek, D.; Jaromin-Gleń, K.; Szeląg, B.; Guz, Ł.; Piotrowicz, A.; Łagód, G. Rapid Method of Wastewater Classification by Electronic Nose for Performance Evaluation of Bioreactors with Activated Sludge. Sensors 2023, 23, 8578. [Google Scholar] [CrossRef]
  90. Callegari, A.; Capodaglio, A.G. Effects of selected industrial pollutants on urban WWTPs activated sludge population, and possible mitigation strategies. Water Pract. Technol. 2017, 12, 619–637. [Google Scholar] [CrossRef]
  91. Khan, M.B.; Lee, X.Y.; Nisar, H.; Ng, C.A.; Yeap, K.H.; Malik, A.S. Digital Image Processing and Analysis for Activated Sludge Wastewater Treatment. In Signal and Image Analysis for Biomedical and Life Sciences; Sun, C., Bednarz, T., Pham, T., Vallotton, P., Wang, D., Eds.; Springer: Cham, Switzerland, 2015; Volume 823. [Google Scholar] [CrossRef]
  92. Satoh, H.; Kashimoto, Y.; Takahashi, N.; Tsujimura, T. Deep learning-based morphology classification of activated sludge flocs in wastewater treatment plants. Environ. Sci. Water Res. Technol. 2021, 7, 298–305. [Google Scholar] [CrossRef]
  93. Borzooei, S.; Scabini, L.; Miranda, G.; Daneshgar, S.; Deblieck, L.; Bruno, O.; De Langhe, P.; De Baets, B.; Nopens, I.; Torfs, E. Evaluation of activated sludge settling characteristics from microscopy images with deep convolutional neural networks and transfer learning. J. Water Process. Eng. 2024, 64, 105692. [Google Scholar] [CrossRef]
  94. Oruganti, R.K.; Biji, A.P.; Lanuyanger, T.; Show, P.L.; Sriariyanun, M.; Upadhyayula, V.K.K.; Gadhamshetty, V.; Bhattacharyya, D. Artificial intelligence and machine learning tools for high-performance microalgal wastewater treatment and algal biorefinery: A critical review. Sci. Total Environ. 2023, 876, 162797. [Google Scholar] [CrossRef] [PubMed]
  95. Kazemi, P.; Giralt, J.; Bengoa, C.; Masoumian, A.; Steyer, J.P. Fault detection and diagnosis in water resource recovery facilities using incremental PCA. Water Sci. Technol. 2020, 82, 2711–2724. [Google Scholar] [CrossRef]
  96. Tokos, A.; Micu, D.D.; Bartha, C.; Jipa, M.; Lingvay, I. SCADA Systems for Wastewater Treatment Plants. Electroteh. Electron. Autom. 2021, 69, 39–45. [Google Scholar]
  97. Sean, W.Y.; Chu, Y.Y.; Mallu, L.L.; Chen, J.G.; Liu, H.Y. Energy consumption analysis in wastewater treatment plants using simulation and SCADA system: Case study in northern Taiwan. J. Clean. Prod. 2020, 276, 124248. [Google Scholar] [CrossRef]
  98. Mamandipoor, B.; Majd, M.; Sheikhalishahi, S.; Modena, C.; Osmani, V. Monitoring and detecting faults in wastewater treatment plants using deep learning. Environ. Monit. Assess. 2020, 192, 148. [Google Scholar] [CrossRef]
  99. Elsayed, A.; Siam, A.; El-Dakhakhni, W. Machine learning classification algorithms for inadequate wastewater treatment risk mitigation. Proc. Saf. Environ. Prot. 2022, 159, 1224–1235. [Google Scholar] [CrossRef]
  100. Bellamoli, F.; Di Iorio, M.; Vian, M.; Melgani, F. Machine learning methods for anomaly classification in wastewater treatment plants. J. Environ. Manag. 2023, 344, 118594. [Google Scholar] [CrossRef]
  101. Salem, R.M.M.; Saraya, M.S.; Ali-Eldin, A.M.T. An Industrial Cloud-Based IoT System for Real-Time Monitoring and Controlling of Wastewater. IEEE Access 2022, 10, 6528–6540. [Google Scholar] [CrossRef]
  102. Yalçın, N.; Çakır, S.; Ünaldı, S. Attack Detection Using Artificial Intelligence Methods for SCADA Security. IEEE Internet Things J. 2024, 11, 39550–39559. [Google Scholar] [CrossRef]
  103. Solano, F.; Krause, S.; Wöllgens, C. An Internet-of-Things Enabled Smart System for Wastewater Monitoring. IEEE Access 2022, 10, 4666–4685. [Google Scholar] [CrossRef]
  104. Daneshgar, S.; Vanrolleghem, P.A.; Vaneeckhaute, C.; Buttafava, A.; Capodaglio, A.G. Optimization of P compounds recovery from aerobic sludge by chemical modeling and response surface methodology combination. Sci. Total Environ. 2019, 668, 668–677. [Google Scholar] [CrossRef] [PubMed]
  105. Cecconet, D.; Capodaglio, A.G. Sewage Sludge Biorefinery for Circular Economy. Sustainability 2022, 14, 14841. [Google Scholar] [CrossRef]
  106. Capodaglio, A.G. Biorefinery of Sewage Sludge: Overview of Possible Value-Added Products and Applicable Process Technologies. Water 2023, 15, 1195. [Google Scholar] [CrossRef]
  107. Sucu, S.; van Schaik, M.O.; Esmeli, R.; Ouelhadj, D.; Holloway, T.; Williams, J.B.; Cruddas, P.; Martinson, P.B.; Chen, W.S.; Cappon, H.J. A conceptual framework for a multi-criteria decision support tool to select technologies for resource recovery from urban wastewater. J. Environ. Manag. 2021, 300, 113608. [Google Scholar] [CrossRef]
  108. Aniza, R.; Chen, W.S.; Pétrissans, A.; Hoang, A.T.; Ashokkumar, V.; Pétrissans, M. A review of biowaste remediation and valorization for environmental sustainability: Artificial intelligence approach. Environ. Pollut. 2023, 324, 121363. [Google Scholar] [CrossRef]
  109. Soo, A.; Gao, L.; Shon, H.K. Machine learning framework for wastewater circular economy—Towards smarter nutrient recoveries. Desalination 2024, 592, 118092. [Google Scholar] [CrossRef]
  110. Bolton, A.; Butler, L.; Dabson, I.; Enzer, M.; Evans, M.; Fenemore, T.; Harradence, F.; Keaney, E.; Kemp, A.; Luck, A.; et al. Gemini Principles; CDBB: Cambridge, UK, 2018. [Google Scholar] [CrossRef]
  111. Kreuzer, T.; Papapetrou, P.; Zdravkovic, J. Artificial intelligence in digital twins—A systematic literature review. Data Knowl. Eng. 2024, 151, 102304. [Google Scholar] [CrossRef]
  112. Karmous-Edwards, G.; Tomić, S.; Cooper, J.P. Developing a Unified Definition of Digital Twins. J. AWWA 2022, 114, 76–78. [Google Scholar] [CrossRef]
  113. Cooper, J.P.; Jackson, S.; Kamojjala, S.; Owens, G.; Szana, K.; Tomic, S. Demystifying Digital Twins: Definitions, Applications, and Benefits. J. AWWA 2022, 114, 58–75. [Google Scholar] [CrossRef]
  114. Torfs, E.; Nicolaï, N.; Daneshgar, S.; Copp, J.B.; Haimi, H.; Ikumi, D.; Johnson, B.; Plosz, B.B.; Snowling, S.; Townley, L.R.; et al. The transition of WRRF models to digital twin applications. Water Sci. Technol. 2022, 85, 2840–2853. [Google Scholar] [CrossRef] [PubMed]
  115. Wang, A.J.; Li, H.; He, Z.; Tao, Y.; Wang, H.; Yang, M.; Savic, D.; Daigger, G.T.; Ren, N. Digital Twins for Wastewater Treatment: A Technical Review. Engineering 2024, 36, 21–35. [Google Scholar] [CrossRef]
  116. Daneshgar, S.; Polesel, F.; Borzooei, S.; Sørensen, H.R.; Peeters, R.; Weijers, S.; Nopens, I.; Torfs, E. A full-scale operational digital twin for a water resource recovery facility—A case study of Eindhoven Water Resource Recovery Facility. Water Environ. Res. 2024, 96, e11016. [Google Scholar] [CrossRef] [PubMed]
  117. Lumley, D.; Jursic Wanninger, D.; Magnusson, Å.; I’Ons, D.; Gustafsson, L.G. Implementing a digital twin for optimized real-time control of Gothenburg’s regional sewage system. Water Pract. Technol. 2024, 19, 657–670. [Google Scholar] [CrossRef]
  118. Plathottam, S.J.; Rzonca, A.; Lakhnori, R.; Iloeje, C.O. A review of artificial intelligence applications in manufacturing operations. J. Adv. Manuf. Process. 2023, 5, e10159. [Google Scholar] [CrossRef]
  119. Beder, S. Technological Paradigms: The Case of Sewerage Engineering. Technol. Stud. 1997, 4, 167–188. [Google Scholar]
  120. World Bank. Water Can’t Wait: Accelerating the Adoption of Innovations in Water Security. Available online: https://blogs.worldbank.org/en/water/water-cant-wait-accelerating-adoption-innovations-water-security#:~:text=The%20sector%20has%20historically%20been,testing%2C%20and%20deploying%20new%20technologies (accessed on 26 October 2024).
  121. Cornelissen, R.; Van Dyck, T.; Dries, J.; Ockier, P.; Smets, I.; Van den Broeck, R.; Van Hulle, R.; Feyaerts, M. Application of online instrumentation in industrial wastewater treatment plants—A survey in Flanders, Belgium. Water Sci. Technol. 2018, 78, 4. [Google Scholar] [CrossRef]
  122. Rapp, A.H.; Capener, A.M.; Sowby, R.B. Adoption of Artificial Intelligence in Drinking Water Operations: A Survey of Progress in the United States. J. Water Resour. Plan. Manag. 2023, 149, 7. [Google Scholar] [CrossRef]
  123. Eerikäinen, S.; Haimi, H.; Mikola, A.; Vahala, R. Data analytics in control and operation of municipal wastewater treatment plants: Qualitative analysis of needs and barriers. Water Sci. Technol. 2020, 82, 2681–2690. [Google Scholar] [CrossRef]
Water 17 00170 g001
Figure 1. Relationship between AI, ML, and DL.
Figure 1. Relationship between AI, ML, and DL.
Water 17 00170 g001
Water 17 00170 g002
Figure 2. Digital twin representation.
Figure 2. Digital twin representation.
Water 17 00170 g002
Table 1. Common AI algorithm classes.
Table 1. Common AI algorithm classes.
AI Algorithm TypeCharacteristicsLimitations
Artificial neural network
(ANN)		One of the most commonly used AI methods.
Simple topology, black-box-type model.
Suitable for large-scale complex problems.		Large amounts of training data required.
Genetic programming
(GP)		Strong robustness, global optimization search.
Straightforward application.
Absence of preliminary assumptions on model
structure. 		High computational density and continuous optimization during the prediction process required. Limited adaptability for continuous prediction.
May be subject to premature convergence.
Fuzzy logic
(FL)		Simple reasoning process.
Simulates human thinking.		Fuzzy rules may compromise prediction. It may struggle to understand complex dynamic of biological processes.
Support vector regression
(SVR)		Simplified modelling process.
High prediction precision if input data properly screened.		Prone to inaccurate results if not optimized to input data. Requires parameter adjustment.
Particle swarm optimization
(PSO)		Imitates the swarm behavior of biological populations (e.g., birds, fish) by constantly adjusting individual position and velocity in search of the optimal solutions.
Can deal with complex nonlinear problems and avoid falling into local solutions.		Sensitive to initial conditions. Requires multiple runs to obtain good results.
Random forest
(RF)		Used in classification and regression problems. Multiple decision trees (DTs) are trained on a random subset drawn from the original dataset.Sensitive to noise and outliers. Requires more computing time and resources to train.
Self-organizing map
(SOM)		ML algorithm for clustering and dimensionality reduction of complex datasets in a low-dimensional mapping space, making it easier to visualize and classify data points.
K-nearest neighbor
(KNN)		Able to deal with complex I/O nonlinear relationships, adaptively adjusting model complexity to various data types using different distance measurements.Sensitive to dataset dimensions and noise. Computationally expensive.
Adaptive neural fuzzy inference system
(ANFIS)		Combine ANN learning mechanism and
FL linguistic reasoning capacity.		Single output type.
Requires deep understanding of the system to be modelled.
Table 2. AI applications for WWTP performance prediction.
Table 2. AI applications for WWTP performance prediction.
ApplicationInputOutputTechniqueRef.
Municipal WWTP influentQ; T; Patm; air humidity; wind speed and direction; rainfall; cloudinessInfluent flow 3 days in advance LSTM, CNN[37]
Municipal WWTP influentQ; CODIN; SSIN; MLSS; MLVSS; NIN; pH; DO; F/M Next day COD effluentMLP[38]
Multiple WWTPs Filamentous species (10 types)SVIANN[39]
Algal PBR with lamella settlerHRT; Q; MLSSBODEFF, CODEFF, TSSEFF, NH4-NEFF, TPEFFANN[40]
Municipal WWTP influentTurbidity; SAC (254 and
433 nm); conductivity; T, Q; pH		CODIN, NH4-NINRF; SVM; LDA; SV; MLP[41]
UASB followed by CASMicrobial taxonomic unitsTOC; DO; effluent NH4-N, PO4-PLM[42]
Municipal WWTPQ; influent BOD5, COD, N, TSS, P, T; rainfall; sunlight irradiationAeration timingPCA, BT, LM[43]
Municipal WWTPQ, NH4-Nin, QR
NH4-NEFFHybrid LM-ANN[44]
Industrial WWTPInfluent Q, pH, T, DO, COD, SS Effluent COD, TSS DBN; GA; hybrid
GA-DBN		[45]
Municipal WWTPT; QR; QEnergy consumptionDNN[46]
Municipal WWTPDO; COD; Q; MLSS; SVI; TFilamentous bulking GPR[47]
Municipal WWTPInfluent Q; MLVSS; OLR; SVI; HRT; RAS TSS; T; pH; WAS TSS; PE TSS; SE TSS; SE TP; BOD5; TKN; TP; NH4-N; COD; TN; MLSSEffluent Q; MLVSS; OLR; SVI; HRT; RAS TSS; T; pH; WAS TSS; PE TSS; SE TSS; SE TP; BOD5; TKN; TP; NH4-N; COD; TN; MLSSANN, SVR, ANFIS,
E-ANN,
E-SVR,
E-ANFIS, E-AVG,
E-WAVG		[48]
Municipal WWTPInfluent Q; MLVSS; OLR; SVI; HRT; RAS TSS; T; pH; WAS TSS; PE TSS; SE TSS; SE TP; BOD5; TKN; TP; NH4-N; COD; TN; MLSS; rainfallSludge production LR, KRR, DT, SVR, KNN, FCNN, RF, XGBoost[49]
Municipal WWTPQ; NH4-N; NO3-N; NO2-N; DO, TSS, TN2O emissionsKNN; DT; RF; DNN; XGBoost; ADABoost, [50]
Table 3. Determination methods and online detectable parameters under current spectrometric sensor technology [60,61].
Table 3. Determination methods and online detectable parameters under current spectrometric sensor technology [60,61].
Determination MethodParameterNotes
UV–VIS spectrometry over total wavelength range 190–750 nmTSS, TS, turbidity, color, TOC, DOC, BOD, COD, NO3-N, NO3, chloramine, HS−, O3, Chl-a, BTX, UV254, chlordeconeAll parameters can be determined with fast measurement intervals (every 30 s or less)
Local multipoint calibration possible according to matrix characteristics
Long-term stable operation without chemical dosage
UV–VIS spectrometry over total wavelength range 200–390 nmNO3-N, COD, BOD, TOC, UV254, NO2-N, BTX
Amperometric (internal buffer 3-electrode system)Free/total chlorinepH range from 4 to 10+
Amperometric (membrane covered)Peracetic acid, hydrogen peroxide, chloride dioxide
Reference electrodepH
ISE (ion-selective electrodes) (with optional potassium compensation)NH4-N, NO3-NISE lifetime: typically 6 month (applications < 1 mg/L NH4-N), 1–2 years (applications > 1 mg/L NH4-N)
Optical/fluorescenceD.O. No consumables
ElectrodeConductivity
Reference electrodeORP
Analytical process
miniaturization		Dissolved reactive phosphorus, total phosphorus, N-NH4Consumables replacement needed
Table 4. Reported soft sensor developments for wastewater quality parameters.
Table 4. Reported soft sensor developments for wastewater quality parameters.
Input Variables aOutput(s) aTechniques b Ref.
EC, pH, ToD, fDOM, HP, Temp, Turb, SMTP, TNRF[81]
DO, Turb, pH, Temp, ORP, ECBODMLR, MLP, SVM-SMO, IBK, RF[82]
DO, Temp, TSS, NH3, pH, TOC, TurbCODMLR, MLP, SVM, RF, kNN[83]
EC, Turb, Temp, DO, pH, Chl-a, QTP, TNRF[84]
Note(s): a fDOM = fluorescence dissolved organic matter; HP = hydrostatic pressure; Temp = temperature; Turb = turbidity; SM = soil moisture; ORP = oxidation reduction potential; TSS = total suspended solids; TOC = total organic carbon; Chl-a = chlorophyl-a; Q = flow rate; TDS = total dissolved solids; BOD = biological oxygen demand; COD = chemical oxygen demand. b RF = random forest; MLR = multi linear regression; MLP = multilayer perceptron; SVM-SMO = support vector machine–sequential minimal optimization; IBK = instance-based learner; kNN = k-nearest neighbor.
Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content.

Share and Cite

MDPI and ACS Style

Capodaglio, A.G.; Callegari, A. Use, Potential, Needs, and Limits of AI in Wastewater Treatment Applications. Water 2025, 17, 170. https://doi.org/10.3390/w17020170

AMA Style

Capodaglio AG, Callegari A. Use, Potential, Needs, and Limits of AI in Wastewater Treatment Applications. Water. 2025; 17(2):170. https://doi.org/10.3390/w17020170

Chicago/Turabian Style

Capodaglio, Andrea G., and Arianna Callegari. 2025. "Use, Potential, Needs, and Limits of AI in Wastewater Treatment Applications" Water 17, no. 2: 170. https://doi.org/10.3390/w17020170

APA Style

Capodaglio, A. G., & Callegari, A. (2025). Use, Potential, Needs, and Limits of AI in Wastewater Treatment Applications. Water, 17(2), 170. https://doi.org/10.3390/w17020170

Note that from the first issue of 2016, this journal uses article numbers instead of page numbers. See further details here.

Article Metrics

Back to TopTop