Nomenclatures
ADMs Anaerobic digestion models
ADM1 Anaerobic digestion model No. 1
AI Artificial intelligence
AR Augmented reality
ASMs Activated sludge models
BOD Biochemical oxygen demand
BOD5 Five-day biochemical oxygen demand
BIM Building information modeling
CAD Computer-aided design
CFD Computational fluid dynamics
CFX Computational fluid dynamics X
D-N₂O Dissolved nitrous oxide
DEXPI Data exchange in the process industry
DO Dissolved oxygen
DTD Digital twin domain
EPA Environmental Protection Agency
FMI Functional mock-up interface
FMU Functional mock-up unit
GIS Geographic information system
GPS Global Positioning System
HMI Human machine interface
ICM Integrated catchment modeling
IMM Information Mirroring Model
IoT Internet of Things
LC-MS/MS Liquid chromatography-tandem mass spectrometry
MBSE Model-based system engineering
MCSSM Multi-objective control strategy mixed soft-sensing model
ML Machine learning
MLVSS Mixed liquor volatile suspended solids
MR Mixed reality
MUCL Mycothèque de l'université catholique de louvain
NSGA-II Non-dominated sorting genetic algorithm-II
PCA-LSSVM Least square support vector machine optimized with principal component analysis
PD Physical domain
PE Pumping energy
PCF Piping component file
PI&D Piping and instrumentation diagram
PLM Product life-cycle management
PLS Partial least-squares
RFID Radio frequency identification
ROM Read-only memory
SBR Sequencing batch reactor
SCADA Supervisory control and data acquisition
SCD Sensing and controlling domain
SS Suspended solids
SWR Sludge wastage rate
SWMM Storm water management model
TIS Tank-in-series
TKN Total Kjeldahl nitrogen
TOC Total organic carbon
TP Total phosphorus
UD User domain
VFAs Volatile fatty acids
VR Virtual reality
Wi-Fi Wireless fidelity
WWTPs Wastewater treatment plants
4-CP 4-chlorophenol
1. Introduction
As an essential resource common to all cities, water and its management are closely related to the quality of life in urban environments. Water management significantly impacts other urban services and their management, making it an essential part of the United Nations’ description of smart cities: “inclusive, safe, resilient, and sustainable cities.” As part of smart city initiatives, smart water management brings multiple benefits to cities facing risks such as water shortages, water quality deterioration, and security challenges, which are aggravated by aging infrastructure, lack of investment, growing urbanization, and climate change. Therefore, accelerating the water industry’s digitalization is imperative, with the adoption of digital twins [
1] being a key element. The focus of this paper is on wastewater treatment engineering, which is an essential element of the urban water cycle [
2].
Currently, the Fourth Industrial Revolution (Industry 4.0) is in the process of integrating digital technologies and industrial processes to bring about innovative solutions in manufacturing [
3], enhancing its dependence on real-time feedback. As part of this revolution, the application of digital twins has been extended from the manufacturing sector to a variety of fields such as medical interventions and virus response [
4], biomanufacturing [
5], earth system simulation and environmental monitoring [
6], climate change mitigation and transportation (specifically smart electric vehicles) [
7], food processing and manufacturing [
8], energy production (particularly in enhancing methane production through anaerobic co-digestion) [
9], and urban planning and development [
10]. Along with the application of modeling, simulations, and digital threads, digital twins will accelerate progress in the planning, design, and management of wastewater treatment engineering [
11].
As the connection between digital twins and smart water management becomes increasingly evident, there is a growing imperative for wastewater treatment plants (WWTPs) to adapt and optimize their wastewater treatment strategies accordingly [
12], [
13]. This review synthesizes research and applications of digital twins across various facets of WWTPs and sewage networks, aiming to offer insights and guidance for enhancing operational efficiency and advancing sustainable wastewater treatment engineering. Section 2 presents a definition of digital twins in the context of wastewater treatment engineering. Section 3 showcases several advances in integrating and implementing digital twins in WWTPs and sewage networks, highlighting the potential of digital twins to significantly enhance these facilities by utilizing the former’s core technologies. Section 4 describes two cases that exemplify the capacity of digital twins to elevate operational efficiency and decision-making. Finally, Section 5 summarizes the present challenges, future prospects, and conclusions from this study.
2. Digital twins
2.1. Concept
In the early stages of a budding technology or concept, the clear definition of relevant terms is pivotal. While the digital twins concept bears some resemblance to modeling, simulation, cyber-physical systems, and the Internet of Things (IoT), it has unique characteristics and applications [
14], which play a crucial role in the intelligent oversight of WWTPs—a topic that will be delved into in 2.3 Domain, 3 Applications for WWTPs and sewage networks.
Originally developed by Grieves in 2005 [
1], the digital twins concept was initially devised for industrial and space technologies. Various adaptations have led to a spectrum of definitions catering to different professional needs [
5].
Table 1 [
11], [
15], [
16], [
17], [
18], [
19], [
20], [
21], [
22], [
23] presents these definitions chronologically, offering a lens into the diverse interpretations borne from different sectors’ requirements.
Definition: Using digital models of wastewater treatment structures, digital twins analyze real-time data to predict and adjust their conditions. As they evolve, digital twins enhance environmental decision-making, effectively streamlining control, data use, and integration between the wastewater industry and socioenvironmental interactions.
2.2. Entity
Digital twins serve as a digital counterpart for various entities, including manufacturing assets and production networks. The systems and processes involved in the digital twins paradigm are detailed in
Fig. 1.In this context, a
system is a network of interrelated entities fostering enhanced decision-making throughout different life cycles, integrating a digital model with an actual entity through well-structured subsystems such as control and security mechanisms [
24]. This integration has two main categories: physical entities and abstract entities.
A
physical entity is present in the real world and stems from human-made elements such as vehicles and products. As each digital twin advances, it encompasses broader scopes such as supply chains, farms, and agriculture [
25], [
26], [
27], [
28]. These entities are further subdivided into artifact, natural, and social entities, each with its own distinctive roles and origins. An
artifact entity is a traditional human-made physical entity resulting from the transformation of a
natural entity for a particular purpose. A
social entity refers to social groups.
In contrast, an
abstract entity is formed by isolating universal characteristics from specific entities, serving functions such as scheduling and health monitoring. These entities, which include conceptual models and theories, have the ability to interact and collaborate harmoniously [
29].
An integral component of the digital twins concept is
process, which refers to a chain of interlinked activities or continuous phenomena undertaking a series of changes to achieve desired outcomes [
17], [
28]. These encompass diverse simulations and analyses and fall under categories such as physical processes and virtual processes, each facilitating a transition between the real and virtual realms. These transitions involve meticulous connections comprising different stages to simulate and realize physical and virtual attributes [
28].
2.3. Domain
The digital twins framework encompasses several domains: namely, the user domain (UD), the digital twin domain (DTD), the sensing and controlling domain (SCD), and the physical domain (PD). The cross-domain functionalities and their relationships are illustrated in
Fig. 2.In the UD, elements such as human interaction, interface design, application software, and co-intelligent digital twins work in harmony, facilitating the optimal utilization of digital twins [
11]. The DTD is responsible for representing the characteristics of physical entities through three vital functions: modeling management, simulation services [
30], and twin co-intelligence [
31]. These functionalities enable detailed visualization, dynamic simulation, and safe resource accessibility, aiding in data flow and transfer with assured security. The SCD plays a critical role in establishing real-time communication between the DTD and the PD [
32]. There are two primary components: the sensing domain and the controlling domain. The sensing domain helps gather vital data from physical objects, whereas the controlling domain effectively implements strategies devised in the DTD. The industrial IoT is leveraged to perceive and convey physical world data. The PD is the realm of tangible components—that is, people, equipment, and processes, where the actual subjects of the digital twin models exist [
33]. Cross-domain functionalities ensure a seamless and secure exchange of information across all these domains, promoting the integrated functioning of the system.
2.4. Key technologies
Leveraging the surge in data across diverse fields, digital twins create virtual replicas of physical entities and use cutting-edge tools such as artificial intelligence (AI) and virtual reality (VR) to digitally control and optimize these entities. The architecture of digital twins hinges on model-based system engineering (MBSE), which encompasses modeling, simulation, and digital threads as its core technologies, supported by the IoT as the foundational technology. Cloud computing, machine learning (ML), big data, and the blockchain constitute the ancillary technologies enhancing the capabilities of digital twins.
Fig. 3 describes the interrelationships between these technologies.
Modeling, which is foundational to digital twins, simplifies the understanding of the physical world and its problems by portraying the causality or interrelations within a system through a model [
34], [
35], [
36], [
37]. This critical process involves creating detailed digital representations of physical entities, encompassing their three-dimensional (3D) geometric structure and shape, operational mechanisms, interfaces, and the software and control algorithms they incorporate [
38]. These digital twin models can vary significantly depending on the distinctive characteristics of different physical entities. Currently, tools such as computer-aided design (CAD) and MATLAB are used for foundational modeling, Revit is employed in building information modeling (BIM) [
39], and CATIA is leveraged for advanced product life-cycle management (PLM) endeavors [
40].
Virtual models are central to the digital twins concept, facilitating high-fidelity digital representations of physical entities across multiple dimensions and scales [
33]. This core part of digital twins seeks to represent physical entities accurately and enhance their functionality through an immersive integration of the virtual and real. This necessitates visual and real-time depictions, supported by technologies such as VR [
41], augmented reality (AR) [
42], [
43], and mixed reality (MR) [
44]. VR serves as a foundational technology, employing computer graphics and dynamic environment modeling to depict the various attributes, behaviors, and rules of physical entities as vividly and realistically as possible. Building upon VR, AR and MR introduce real-time data acquisition, scene capture, and real-time tracking to synchronize and fuse virtual models with physical entities, effectively enhancing the detection, verification, and guidance functionalities. Moreover, the metaverse concept represents an evolved and expansive virtual environment that integrates AR, VR, and MR technologies and content [
45].
Technically, modeling and simulation are intertwined. Modeling articulates our comprehension of the physical world or specific challenges, while simulation validates the accuracy and relevance of this understanding [
36], [
46], [
47]. In industry, simulation employs software to replicate the physical world based on models that integrate both deterministic rules and comprehensive mechanisms, as well as stochastic, knowledge-based models in some cases. If the model is accurate and the input data are complete, the simulation effectively mirrors the attributes of the physical world.
Traditional WWTPs were not conceptualized with digital twins in mind. To modernize these plants for enhanced wastewater treatment efficiency, it is crucial to incorporate both two-dimensional (2D) and 3D model information into one digital model. However, many of these plants currently operate with 2D designs rather than the more advanced BIM model, making graph matching and model reuse challenging.
The functional mock-up interface (FMI) was introduced to address issues such as fragmented simulation tools, limited model reusability, and intellectual property protection [
48]. The FMI offers a universal interface standard for model reuse, focusing on both model exchange and the co-simulation of functions and performance. The adoption of FMI makes model integration straightforward. Files exported using the FMI standard are labeled with a “.fmu” extension (functional mock-up units).
Digital threads act as connective bridges between entities, serving as an adaptable enterprise-level communication framework [
49]. This structure enables a comprehensive perspective, encompassing cross-level, cross-scale, and multi-view models that span the entire system life cycle and value chain. The primary function of digital threads is to guide system activities throughout their lifespan and aid decision-makers. Essentially, digital threads ensure timely and appropriate information delivery to the right stakeholders throughout the system’s life cycle [
50].
MBSE is a structured approach for developing digital twins [
51], [
52], [
53]. As the cornerstone of digital threads, MBSE leverages IoT data to ensure that simulations can detect potential failures, thus facilitating continuous improvements in existing operating systems.
The IoT acts as the foundation for digital twins [
54], gathering information from the Internet, traditional telecommunications, and various tools such as sensors, radio frequency identification (RFID) [
55], Global Positioning Systems (GPSs) [
56], and laser scanners. This allows standalone objects to be transformed into a connected network, enabling seamless interactions between objects and people. The IoT allows for the intelligent recognition and management of items and processes, facilitating the timely, dependable, and efficient transmission of twin data.
Read-only memory stores fixed programs and data that can be read but not altered, working in a non-destructive readout mode. This system has a simple structure and offers stable data storage, ensuring that data remain unchanged even in the event of a power outage, making it both reliable and user-friendly.
The scalability of digital twins varies based on the demands. While unit-level digital twins might function on a local server, system-level and complex digital twins demand more computational and storage power. Cloud computing [
57] caters to these needs by offering extensive resources and data centers, allowing digital twins to adapt to diverse computing, storage, and operational requirements. Fog computing [
58] extends cloud computing by distributing resources across numerous decentralized nodes, thereby bringing data processing closer to the edge of the network. This approach enhances operational speed and efficiency by leveraging localized processing. Complementarily, edge computing directly processes data at or near its source, further optimizing real-time data analysis for improved perception, calculation, and control at edge nodes [
59]. In tandem with cloud computing, edge computing forwards intricate twin data to the cloud for advanced processing. This cloud-edge collaboration addresses varied requirements, boosts data processing speeds, minimizes the cloud data burden, and curtails data transmission lags. In this way, the real-time functionality of digital twins is significantly enhanced. Notably, system-level digital twins are well aligned with fog computing, given their primary concentration within manufacturing enterprises and specific geographical locales.
Data are a dynamic and rapidly changing asset, requiring innovative processing techniques to enhance decision-making, insight, and optimization. Big data [
60] leverages the voluminous data created by digital twins to elucidate and forecast real-world outcomes and processes, thereby extracting precious information. As a complement to this, ML [
61] facilitates the automatic analysis of data to derive rules that can be used for predictions. Digital twins use ML to forecast future states and behaviors through data harvested from the PD through the IoT, offering valuable (albeit potentially imprecise) insights. Hence, big data and ML are invariably linked, working in tandem to provide a rich analytical foundation.
Digital twins represent digital assets and participate in digital transactions. Leveraging blockchain technology [
62] has the potential to enhance the security of digital twins by preventing unauthorized alterations that could result in errors and deviations. This fosters a safer environment that promotes innovation. Furthermore, the decentralized trading mechanisms facilitated by the blockchain ensure secure, distributed, and real-time digital asset transactions, thus providing an optimal medium for digital twins’ asset trading and fostering user trust in the services provided by digital twins.
In conclusion, the successful implementation and application of digital twins hinge on support from emerging technologies [
63]. Through deep integration with these technologies, digital twins can achieve an authentic and comprehensive perception of physical entities. This involves the precise creation of multidimensional, multiscale models, extensive data fusion, customizable service usage, and full-scale, dynamic, real-time interaction.
3. Applications for WWTPs and sewage networks
3.1. Facilities
With advances in digital twins, accompanied by progress in modeling and simulation techniques, WWTPs have seen significant improvements in their facilities (
Table 2 [
64], [
65], [
66], [
67], [
68], [
69], [
70]). These developments have been observed in structures such as secondary settling tanks, biological aerated filters, and primary clarifiers, enhancing their treatment efficiency to a considerable extent. Advances have been driven by the use of ML to optimize the operation of essential equipment such as water pumps, air blowers, sludge pumps, and mixers in the trial stage, fostering their readiness for real-world applications.
CFD: computational fluid dynamics.
In recent years, substantial advances have been made in various aspects of water treatment technology. For example, in terms of settling techniques, the amended Vesilind function for hindered settling has been developed and validated, leading to a new exponential function addressing the compression settling velocity [
64]. This theme of refinement has continued with the verification and simulation of critical parameters (including the hydraulic loading rate, organic load rate, and surface area of packing materials) used in a packed-bed up-flow anaerobic sludge blanket followed by a biological aerated filter—a project undertaken by the Water Research Department at the National Research Center in Cairo, Egypt [
65].
Progress has also been observed in pump technology. In 2016, Kim et al. [
66] enhanced the hydrodynamic performance of single-channel pump impellers using a contemporary design approach. Building upon this, in 2020, the same team improved the hydraulic performance and prediction accuracy of two-vane pumps through the resolution of steady Reynolds-averaged Navier-Stokes equations [
67]. In parallel, work spearheaded by Lozano Avilés et al. [
68] capitalized on advanced flow modeling techniques to address deficiencies in fluid distribution and mixing, successfully reducing the necessary airflow to the reactor by over 3%.
Adding to the wave of innovations, advances facilitated by computational fluid dynamics (CFD) calculations in Ansys Fluent have underscored the improved applicability of vortex impellers in sludge pumps, marking a significant step in this domain [
69]. These developments, which are characterized by enhanced mixing, propulsion, and abrasion resistance of the impellers, have been successfully implemented in Zhenjiang, China, demonstrating the real-world impact of this research [
70].
3.2. Pipes
The evolution of water management has been deeply influenced by the integration of digital technologies, especially in the modeling and management of underground pipe networks. The advent of 2D and 3D pipe network models, developed with the help of CAD, Ansys computational fluid dynamics X (CFX), BIM, geographic information systems (GISs), and integrated catchment modeling (ICM) software, has paved the way for a unified standard in information digitization. This comprehensive approach not only enhances the monitoring and simulation of water flow and quality but also bolsters the prediction and verification of pollutants and streamlines the management of these intricate underground networks (
Table 3 [
3], [
71], [
72], [
73], [
74], [
75], [
76], [
77], [
78], [
79]).
SWMM: storm water management model; PI&D: piping and instrumentation diagram; DEXPI: data exchange in the process industry; PCF: piping component file; Wi-Fi: wireless fidelity.
In a notable application, China successfully launched a digital management platform for urban sewage networks in Sanya, Hainan Province. This innovative system integrates a sewage information control center with digital management, dynamic simulation, and emergency grid management. The system’s prowess lies in its ability to perform dynamic simulations of drainage networks, employ GIS spatial management and analysis, and conduct meticulous sewage network grid management [
71].
Further advances include the creation of an urban rainstorm model in Guangzhou’s Donghaochong Basin. This model analyzes the interception efficacy in regard to combined sewer overflow pollution and assesses flood mitigation levels [
72]. Using hydraulic and hydrological datasets generated by the US Environmental Protection Agency’s Storm Water Management Model (EPA-SWMM), Sun et al. [
73] explored how the flow rate, rain intensity, and pipe length influence the outputs from total suspended solids models in the Bordeaux region of France. On a more technical note, Fedorov et al. [
74] investigated the two-phase flow dynamics of wastewater streams and a mixture of air and hydrogen sulfide, pinpointing areas of intense emission. Moreover, the quality of data has been the focus of several studies. Nie’s [
75] work emphasizes the validation and refinement of semantic and topological data, ensuring minimized data loss and fostering collaboration among various underground construction stakeholders. In summary, the digital transformation—manifested in the form of unified information systems [
76] and web platforms [
3]—underscores a pivotal shift toward standardized digitization in underground water management.
3.3. Sensors for water quality
To adapt to increasingly strict environmental regulations, future WWTPs will require intelligent control mechanisms empowered by AI. A core part of this progression is the development of data collection and processing [
80], which are facilitated through the extensive deployment of water quality sensors. This makes the establishment of centralized and standardized databases imperative. These databases will not only store information for ongoing water quality monitoring but also integrate seamlessly with IoT software to foster online management systems, consequently reducing the costs associated with personnel training and other related expenses [
81].
The supervisory control and data acquisition (SCADA) system is central to this transformation, assimilating data from various sensors to enable the autonomous optimization of process parameters and overseeing the functioning of aeration systems in WWTPs (
Table 4 [
81], [
82], [
83], [
84], [
85], [
86], [
87], [
88], [
89]). Leveraging SCADA will streamline water quality monitoring, allowing for real-time parameter measurement without the need for sampling or extensive user training, thereby supporting informed decision-making in wastewater treatment [
82], [
83], [
84].
PLCs: programable logical controllers; HMI: human machine interface.
Another pivotal component in this landscape is metadata, which is primarily used for data collection and storage. Platforms such as Bluemix facilitate the acquisition and integration of both historical and real-time water data, spanning quantitative and qualitative metrics over extensive stream distances and thereby enhancing real-time water quality monitoring [
85]. This inclusive database structure, underscored by a focus on metadata, ensures easy access to standardized, centralized data, thoroughly documenting all pertinent information associated with measurements [
81].
Recent initiatives highlight the fruitful application of these technologies. In Xiamen, China, an online water quality management system has successfully stabilized urban scenic river water quality, leveraging data analytics to regulate the freshwater supply from Xinglin Bay [
86]. AquaSat also promises to be a rich resource for future
in situ water quality assessments [
87].
The efficacy of SCADA has been demonstrated globally, including in Romania, where it governs WWTP operations autonomously while maintaining optimal technological parameters and recording vital operating data [
84]. Similarly, in the Republic of Korea and China, SCADA has played a critical role in monitoring aeration systems and enhancing water quality, illustrating its crucial role in the modernization of WWTPs [
88], [
89].
3.4. Sensors for activated sludge
In WWTPs, sensors play an integral role in identifying and detecting the diverse states and properties of activated sludge, and their application is set to expand in the future. Technologies such as soft sensors, liquid chromatography-tandem mass spectrometry (LC-MS/MS), low-field
1H nuclear magnetic resonance, and independent component analysis facilitate a range of assessments, enabling the identification of sludge bulking [
90], the detection of quorum sensing signal substances from both water and solid sludge phases [
91], and the measurement of water content and moisture distribution within the sludge [
92]. Moreover, these tools determine the quantities of different water types in wastewater sludge by assessing the relevant parameters [
93]. Research related to these advances has been extensively undertaken in Poland, China, and Finland (
Table 5 [
90], [
91], [
92], [
93]).
3.5. Hydrodynamics
The design of a WWTP is predominantly influenced by the target pollution removal rate, with the efficiency largely depending on the hydrodynamics of the bioreactors incorporated in the WWTPs. The development of CFD, the compartment model, and the tanks-in-series (TIS) model has enabled the modeling and prediction of these hydrodynamic characteristics, which are crucial in predicting the relevant parameter values and aiding in the removal and degradation of pollutants (
Table 6 [
94], [
95], [
96], [
97], [
98], [
99], [
100], [
101], [
102], [
103]).
4-CP: 4-chlorophenol; TOC: total organic carbon.
The use of CFD has facilitated the prediction and determination of a range of parameters. For example, Matko et al. [
94] leveraged CFD to enhance the design of oxidation ditches and aerators by forecasting the gas-liquid flow pattern and dissolved oxygen distribution. Similarly, CFD was applied by Elhalwagy et al. [
95] to identify the relationship between suspended solids (SS) and the efficiency of a new disinfectant in a municipal contact tank. Other research groups have used CFD to optimize conditions for pollutant degradation in different reactor setups [
96], [
97], [
98], [
99].
Complementing this, the compartment model and TIS model have further refined the prediction accuracy. Hormann and Fischer [
100] improved the forecasts of radioiodine movement in public sewer systems, while Ng [
101] focused on predicting the transport of urban radiocesium during wet weather events. Moreover, studies have examined the qualitatively the reactor's mixing regime [
102] and quantified accurate values of mobile volume and immobile volume [
103].
3.6. Power consumption
The design of WWTPs necessitates careful consideration of power consumption—a demand propelled by global population growth, industrial advancements, lifestyle alterations, and climate change. The surge in energy requirements poses a substantial challenge for WWTPs, especially in the context of the push for carbon neutrality and the imposition of energy limitations. Leveraging data-driven soft-sensor methodologies, which incorporate traditional time series and deep learning, enables the formulation of power consumption predictive models for WWTPs, facilitating the reduction of energy use during the initial stages of water treatment (
Table 7 [
88], [
99], [
104], [
105], [
106], [
107], [
108], [
109], [
110], [
111]).
The ASM1Temp model is an extension of Activated Sludge Model No. 1, which considers carbon removal, nitrification, and denitrification with temperature correction.
LSTM: long short-term memory.
Sean et al. [
88] used current and water quality data to forecast optimal airflow rates and energy expenditure, offering a valuable reference for the initial phase of plant operation, while Saini et al. [
99] explored the energy dynamics of pumped recirculation in an existing anaerobic digester, focusing on the inlet planes. Moreover, Harrou et al. [
104] and Cheng et al. [
105] have attempted to predict the short-term energy needs of WWTPs using flow rates, temperature data, and biochemical oxygen demand, fostering data-driven management of these plants. In addition, De Canete et al. [
106] applied ML to determine variables affecting influent quality, such as chemical oxygen demand (COD), total nitrogen (TN), and total suspended solids (TSS), thus optimizing energy consumption and minimizing violations in biological wastewater treatment facilities. WEST, a Belgian simulation platform initially created for wastewater treatment, is a versatile environment for dynamic network modeling and long-term simulation development [
107]. Cechinel et al. [
108] considered the prediction of effluent quality, while Muoio et al. [
109] identified the optimum solid retention time of a large industrial WWTP in an attempt to minimize the operating costs. Kovács et al. [
110] modeled biofilm reactors that contributed to a base module in SUMO. Kirchem et al. [
111] proposed a flexible demand scheme for the power sources of WWTPs.
3.7. Knowledge-based and data-driven models
Since their introduction, data-driven models have served three main purposes in WWTPs: fault detection, variable prediction, and advanced control. Deeper insights into activated sludge models (ASMs) and advanced control are presented in 3.8 Mechanistic models, 3.10 Control methods, respectively [
112]. Section 3.7 will focus on knowledge-based and data-driven models (
Table 8 [
113], [
114], [
115], [
116], [
117], [
118], [
119], [
120], [
121], [
122], [
123]). This approach uses methods such as control charts, principal component analysis, partial least-squares (PLS), and neural networks for fault detection while employing tools such as transfer function models, multiple regression, and neural networks for variable prediction.
MLVSS: mixed liquor volatile suspended solids; SBR: sequencing batch reactor; DO: dissolved oxygen; TP: total phosphorus; BOD5: five-day biochemical oxygen demand; BOD: biochemical oxygen demand; D-N2O: dissolved nitrous oxide.
The domain of fault detection, which is crucial for the smooth operation of WWTPs, leverages various methodologies. Control charts enabled Santos et al. [
113] to monitor membrane permeability and dictate necessary interventions, and allowed Trubetskaya et al. [
114] to identify specification limits using industrial data. Similarly, principal component analysis has aided in pinpointing suitable sub-period division strategies for paper mill sequencing batch reactor (SBR) processes and performing statistical analysis of WWTP quality parameters [
115], [
116]. Using PLS, Liu et al. [
117] detected sensor faults within processes with nonlinear and dynamic features and improved the prediction performance and stability of effluent quality indexes [
118]. Moreover, neural networks have facilitated a nuanced understanding of the intricate relationship between raw influent and treated effluent water quality data in Iraq [
119].
Concerning variable prediction, a range of studies have attempted to forecast elements such as sedimentation reservoir outflow turbidity, removal efficiency of different wastewater treatment technologies, and daily urban wastewater discharge, thereby contributing to more efficient management of WWTPs [
120], [
121], [
122]. Moreover, detailed evaluations have been conducted to explore periodic fluctuations in water quality parameters over extended periods [
123].
3.8. Mechanistic models
In the field of wastewater treatment, technological improvements have fostered the evolution of mechanistic models, notably ASMs [
124], [
125], anaerobic digestion models (ADMs) [
126], [
127], and soft-sensor mechanisms [
128], [
129]. These models are used to simulate specific treatment processes, facilitating the prediction of target outputs and aiding in the assessment of the entire operation, as outlined in
Table 9 [
90], [
104], [
105], [
106], [
125], [
127], [
128], [
130], [
131], [
132], [
133], [
134], [
135], [
136], [
137], [
138], [
139], [
140], [
141], [
142], [
143], [
144], [
145], [
146], [
147], [
148].
TKN: total Kjeldahl nitrogen; ADM1: anaerobic digestion model No. 1; VFAs: volatile fatty acids; SWR: sludge wastage rate; PE: pumping energy.
ASMs and ADMs have been instrumental in the optimization and estimation of numerous variables. Employing these models has allowed researchers to enhance the operational conditions of coking WWTPs, thus reducing costs [
130], as well as to gauge the removal efficacy pertaining to antibiotics [
131], explore the implications of sludge vertical stratification on the spatial and temporal distributions of ASM components [
132], and predict biogas production in various phases within anaerobic reactors [
133].
Soft sensors occupy a substantial segment of the mechanistic model domain, with applications across a wide spectrum in WWTPs. They aid in the forecasting of intricate variables, such as the nitrate concentration in denitrifying post-filtration units [
134] and the emulation of weather predictions for controlling WWTPs [
135]. Various studies have leveraged soft-sensor mechanisms to devise real-time control strategies for diverse parameters, including the sludge lysate return ratio under fluctuating influent low C/N ratios [
136], total Kjeldahl nitrogen (TKN) estimation in long-term complex wastewater treatment processes [
137], and the enhancement of soft-sensing model efficiency and precision in predicting effluent quality [
138]. In addition, soft sensors have facilitated the prediction of challenging-to-measure yet quality-relevant variables in WWTPs [
139], the online monitoring of pivotal variables in wastewater procedures while capturing nonlinear and non-Gaussian data [
140], and the extraction of dynamic characteristics for quality variable prediction [
141].
3.9. Hybrid twins
Leveraging the digital twins concept, hybrid twins combine real data with digital replicas, creating a complementary and supplementary virtual model grounded in physical principles that encapsulate causality. Hybrid twin models refine their simulation outcomes based on actual test parameters, thereby reducing the testing costs and enhancing data precision. Within the context of WWTPs, hybrid twins play a pivotal role in mitigating ambiguities [
149]. Furthermore, they assist in the comprehensive design of operation variable systems tailored for multi-objective control of the anaerobic ammonia oxidation process [
150], as detailed in
Table 10 [
149], [
150], [
151].
PCA-LSSVM: least square support vector machine optimized with principal component analysis; NSGA-II: non-dominated sorting genetic algorithm-II; MCSSM: multi-objective control strategy mixed soft-sensing model.
3.10. Control methods
Control systems in WWTPs are differentiated into four main categories: linear control, linearizing control, nonlinear control, and AI-based control [
135]. Each category encompasses a range of strategies, as detailed in
Table 11 [
152], [
153], [
154], [
155], [
156], [
157], [
158], [
159], [
160], [
161], [
162], [
163], [
164], [
165]. Linear control strategies are applied to individual parameters within WWTPs to enhance processes. For example, aeration costs have been reduced through the efficient control of dissolved oxygen (DO) in activated sludge process-based treatments [
152], [
153], [
154], N
2O emissions have been mitigated during nitrification [
155], and effluent quality has been bolstered while conserving energy through the minimization of effluent COD and organic nitrogen content [
156].
Linearizing control serves to harmonize multiparty conditions in WWTPs, thus reducing large fluctuations in influent flow rates and concentrations, as well as uncertainties in measurement noise and kinetics. This control strategy improves the efficiency of WWTPs [
157], [
158] and enhances the reliability of processes such as denitrification and dephosphorization in anaerobic-anoxic-oxic (A
2/O) reactors [
159]. Linearizing control has also been applied to the optimization of aeration in water resource recovery facilities, in alignment with distinct management objectives [
160].
Nonlinear control facilitates the prediction and handling of dynamic parameters. Specific applications include maneuvering the reverse osmosis process to remove dimethylphenol from wastewater [
161], energy conservation without sacrificing aeration efficiency [
162], and forecasting TN peaks well in advance to modulate airflow and maintain stringent effluent standards, while also saving energy [
163].
AI-based control uses mapping models to further refine effluent quality and reduce the frequency of plant measurements [
164]. This strategy can also pre-emptively calculate outlet results to facilitate forward planning [
165].
In summary, these varied control strategies form a comprehensive toolkit, enhancing the efficiency, reliability, and predictive capabilities of WWTP operations. Overall, they are pivotal in fostering improvements in both energy conservation and treatment efficiency.
4. Case study
4.1. Boai County No. 2 WWTP
The Boai County No. 2 WWTP in Jiaozuo, China, stands as the world’s inaugural digital twins WWTP operating on PLM technology [
166]. Designed to serve 150 000 residents, it boasts a daily municipal wastewater treatment capacity of 60 000 t. The plant complies with China’s stringent class-A pollutant discharge standards, as outlined in Discharge Standard of Pollutants for Municipal Wastewater Treatment Plant (GB 18918—2002), treating influent with specified concentrations of various substances, including COD (270 mg∙L
−1), five-day biochemical oxygen demand (BOD
5, 140 mg∙L
−1), SS (200 mg∙L
−1), NH
3-N (35 mg∙L
−1), total phosphorus (TP, 4 mg∙L
−1), and TN (50 mg∙L
−1), to produce effluent suitable for release into natural water bodies. This treatment employs an A
2/O process complemented by coagulation, sedimentation, and filtration procedures.
Upon establishing its foundational model, the WWTP brought together design, operational, maintenance, and real-time data to enhance three core functionalities:
(1) Virtual inspection: Leveraging the synergy between 3D models and real-time data, virtual inspections have been introduced to address the challenges of arduous control and debugging tasks, simplify personnel training, and facilitate multidimensional data analysis by portraying data information accurately and comprehensively in real time.
(2) Digital delivery: Integrating 2D and 3D representations of real-time data, digital delivery overcomes the issues associated with paper delivery, such as complexity and reuse difficulty in operations and maintenance. This strategy minimizes the maintenance workload and enables automated control systems to operate with reduced or no human supervision.
(3) Predictive analysis: By engaging simulations across input/output, equipment, and process layers, the system can forecast the inlet flow rate, DO fluctuations, and effluent indices. This intelligent analysis uses stored data for self-diagnosis, pinpointing the causes of any issues and issuing early warnings to preempt them.
This comprehensive approach ensures that the plant operates effectively, maintaining a commitment to environmental standards while streamlining operations through technological innovation.
4.2. Nosedo WWTP
The Nosedo WWTP is the primary municipal wastewater treatment facility in Milan, Italy, and is also Europe’s largest, with an impressive capacity that can serve 1 250 000 population equivalents [
167]. The facility boasts a handling capacity of 432 000 tonnes per day and processing rates of 5 m
3∙s
-1 in dry weather and 15 m
3∙s
-1 under rainy conditions. A significant 60%-70% of the treated water is subsequently channeled to support agriculture. The WWTP’s operational efficiency, which has resulted in yearly savings of approximately 630 000 EUR, spans three critical areas:
(1) Integrated operation: The WWTP has orchestrated the seamless integration of the sewer system and its treatment processes. This system empowers real-time decision-making, optimal control of biochemical processes, and a marked reduction in energy consumption, constituting 40% of the plant’s total energy use. Aided by this system, the plant expertly manages variations in biological load, ensuring minimal manual adjustments and offering a more comprehensive process overview.
(2) Operational savings: A strategic focus on reducing energy use, chemical consumption, and sludge production has borne significant savings. Breakdowns show a 25% energy reduction in biological treatment, 9% in grit chamber aeration, and an impressive 80% in FeCl3. The decrease in the usage of FeCl3 is attributed to the enhancement of the phosphorus precipitation process. Furthermore, chemical sludge production has been reduced by 126 tonnes per year as a result of reduced precipitation.
(3) Enhanced hydraulic capacity: The Nosedo WWTP has optimized its biological processes to better manage wet weather scenarios. With the aid of stormwater-mode assessments using rain gauges and sewer measurements, the plant has boosted its hydraulic capacity by 20%-30% during inclement weather and rainstorms, ensuring greater resilience and efficiency.
By honing its strategies in these critical areas, the Nosedo WWTP has emerged as a beacon of efficiency and sustainability in wastewater treatment. Continual optimization of its processes has helped to safeguard the environment while maintaining economic viability.
5. Concluding remarks
Over the years, the digital twins concept has steadily cemented its role as a transformative force in various industries, including the critical sector of wastewater treatment. As we navigate the intricacies of its applications and developments thus far, it is pertinent to delineate existing challenges while casting a speculative eye on future prospects.
The synthesis of real-time data and established models poses a considerable technical challenge, mandating advanced control systems that are adept at handling multiple variables. These intricacies have the potential to produce a steep learning curve, complicating the task of training personnel to manage these complex systems proficiently. Furthermore, the sector is grappling with the optimization of energy and chemical consumption, where striking a balance between efficiency and efficacy is essential. As exhibited in practical applications, the precise prediction of dynamic parameters requires enhanced focus and development to secure reliable and safe effluent standards.
The future integration of digital twins in WWTPs marks a pivotal shift toward smarter urban water management. The pioneering cases of the Boai County No. 2 and Nosedo WWTPs exemplify the capacity of digital twins to elevate operational efficiency and decision-making. This innovation transcends traditional monitoring, embracing advanced predictive maintenance and resource optimization. Coupled with the IoT, the digital twins concept is set to redefine the standards of sewage treatment and environmental stewardship [
168], [
169], [
89]. In the future, digital twins are expected to seamlessly blend into broader digital water infrastructures, heralding an era of enhanced, interconnected water services that prioritize efficiency, resilience, and sustainability.
In conclusion, we have summarized the concept, entity, domain, and key technologies of digital twins in the context of wastewater treatment engineering in this technical review. Digital tools have been developed to aid decision-making across various aspects of WWTPs and sewage networks. Furthermore, the two decision-support digital-tool cases given herein exemplify the potential for improving sewage treatment processes and environmental outcomes. It is anticipated that the integration of digital twins with emerging technologies, such as the IoT, will strengthen the monitoring, predictive maintenance, and adaptive strategies for resource optimization in WWTPs. Through the use of real-time analytics, decision-support digital tools are poised to significantly enhance the efficiency and decision-making capabilities of WWTPs. It is recommended that future efforts should expand digital integration, innovate data analysis techniques, and broaden the scope of environmental applications to further augment the potential of digital twins.
Acknowledgments
This work was supported by the National Natural Science Foundation of China (52321005, 52293443, and 52230004), the Shenzhen Science and Technology Program (KQTD20190929172630447), the Shenzhen Key Research Project (GXWD20220817145054002), and the Talent Recruitment Project of Guandong (2021QN020106). The authors would like to thank Dr. Stuart Jenkinson from Edanz for improving the draft’s language.
Compliance with ethics guidelines
Ai-Jie Wang, Hewen Li, Zhejun He, Yu Tao, Hongcheng Wang, Min Yang, Dragan Savic, Glen T. Daigger, and Nanqi Ren declare that they have no conflict of interest or financial conflicts to disclose.
Special issue articles
PDF
(1228KB)
