Connecting digital twins to control collections systems and water resource recovery facilities: From siloed to integrated urban (waste)water management Open Access

The impacts of climate change, increased urbanization and urban growth, aging infrastructure, and community values are increasing the pressure on cities and their utilities in the water industry. Available instrumentation and online data have shown great potential for tackling the problems of aging infrastructure and the need to extend infrastructure service life at low cost (Garrido-Baserba et al. 2020). Nevertheless, increasing data availability as well as the need to address multiple decision criteria requires holistic tools to convert data into actionable information.

Rapid advances in digitalization have offered great opportunities for data collection, storage, and processing as well as fast model computation (Grievson et al. 2022), opening the path for commercial digital twin applications. Existing digital twin applications include (a) Future City Flow (FCF; developed by DHI A/S, Denmark) for catchments and collection systems; (b) Replica™ (developed by Jacobs, US; Johnson et al. 2021) and TwinPlant (developed by DHI A/S, Denmark) for WRRFs; and (c) model predictive control applications for integrated catchment–WRRF management under wet-weather conditions (Stentoft et al. 2019). Mechanistic or data-driven models (or combination of these) are used as virtual replicas of the respective systems. These applications are currently in use worldwide, providing operational support to water utilities for planning, operator training and real-time control (RTC), and optimization. Nevertheless, existing applications (with the exception of Stentoft et al. 2019) are still primarily confined to the specific system of interest (a catchment or a WRRF), and the advantages of their use in combination are still largely unexplored.

The objectives of the current study are to (i) provide an initial overview of digital twin applications for catchments and WRRFs and respective use cases; (ii) present a vision and concrete example for the integration of collection system and WRRF digital twins; and (iii) highlight the benefits of such integration for integrated (waste)water management as well as the expected challenges and response solutions. The presented results are based on existing applications of FCF and TwinPlant and their integration. Overall, the study aims at overcoming the existing barriers in the integrated management of wastewater systems and opening the path for increasing communication and cooperation within and between urban water utilities.

FCF is a family of digital tools for collection systems originally developed by a consortium of Scandinavian utilities, universities, and technology providers (Hagman et al. 2018). In general terms, FCF integrates the following digital twin components: (i) a model of the catchment and collection system of interest; (ii) continuous and automated acquisition of online data from the catchment and other sources (e.g., weather forecasts); (iii) a web-based graphical user interface (GUI) for data visualization; and (iv) real-time and predictive control of catchment operations (e.g., pump utilization, use of sewer storage volumes).

The work of the FCF consortium resulted in three tools, FCF DATA for acquiring and analysing catchment data, FCF RTC for optimization of existing catchment infrastructure, and FCF planning for long-term planning and optimization of catchments. The FCF RTC digital twin utilises DHI's software MIKE+ CS for collection system simulation and MIKE OPERATIONS (version 2022 and subsequent, DHI A/S, Hørsholm, Denmark). Modules available within the MIKE+ CS collection systems modelling package are used for describing the hydrological effects and the hydraulic performance of the collection system. MIKE OPERATIONS provides tools for task scheduling, data import, validation and exchange, model execution, online forecasting, post-processing, optimisation, alarm generation, report generation, and data exchange with the web-based GUI.

Generally, a catchment is divided into a number of subcatchments, and a model is formulated and verified for each subcatchment using a deterministic model concept found in MIKE+ CS (Gustafsson et al. 1993; Andersen et al. 2004). The hydrological processes are described with a general hydrological model (MIKE+ RDI, Lumley et al. 2024) and account for runoff from impervious areas as well as infiltration into the sewer system from the surrounding soil. Each subcatchment model is a simplified conceptualisation of the real subcatchment reflecting the salient characteristics of the subcatchment, such as sewage flows, accumulated overflows, transport times within the subcatchment, and controllable structures, with a level of detail that allows computational times suitable for RTC. This surrogate technique (Wright & Davidson 2020) overcomes some of the inherent limitations of high-fidelity models as noted by Meneses et al. (2018). The hydraulics of the main transport system, e.g., transport, storage, controllable structures, and overflows, are described in a hydrodynamic model based on de Saint-Venant's equations (de Saint-Venant 1871).

Data assimilation (Hutton et al. 2014; Lund et al. 2019) is applied to the calibrated collection system model to deal with residual anomalies and uses weighting function algorithms to optimize the online model with current and historical observations and system status. Control strategies are customized for every FCF RTC digital twin to mirror the necessary operating conditions or constraints in the catchment. After every simulation, recommended setpoints for the various control handles in the catchment are sent to the catchment's control system to be used in an advisory capacity or for direct implementation in the control system after a setpoint plausibility control (e.g., variation from previous setpoint, exceedance of setpoint boundaries).

More than 20 catchments have FCF digital tools and more than a third of these implement FCF RTC. The FCF RTC digital twin has been found to provide a sound base for modelling, simulating, forecasting, and controlling the catchment. The versatility of the FCF digital twin makes it useful for gaining insight into catchment dynamics and the simulator can be used to study the effects of different scenarios or changes in the catchment. The predictive control imbedded in the FCF digital twin recommends setpoints for controlling the catchment that can be used by the operators for decision support or even be directly applied in the control system as model-based predictive control. A digital twin requires good quality online data from the catchment (e.g., flow and level measurements at different locations) and, perhaps most importantly, high-quality rain forecasts, especially when prognosing long-time horizons. A more detailed description of the FCF RTC digital twin was presented by Lumley et al. (2024).

TwinPlant is a digital twin technology developed for WRRFs that was developed by DHI in cooperation with Aarhus Vand, one of Denmark's largest water utilities. TwinPlant integrates the following digital twin components (for further details, see Daneshgar et al. 2024): (i) a model of the WRRF of interest; (ii) continuous and automated acquisition of online data from sensors and laboratory measurements at the WRRF and from other sources (e.g., weather forecasts); (iii) a web-based GUI for data visualization and what-if scenario analysis for operator training and decision support; (iv) identification of optimal control setpoints based on the defined objectives. TwinPlant is based on the WEST process simulator (version 2022 and subsequent, DHI A/S, Hørsholm, Denmark) and uses the data integration and management system DIMS.CORE (version 10.0.1 and subsequent, DHI A/S, Hørsholm, Denmark) for automated acquisition of online data from SCADA, validation, and pre-processing (gap filling, filtering) as well as RTC tasks (e.g., setpoint modification).

What-if scenario tools allow users to evaluate alternative WRRF operation in a virtual environment both for past (The Learning Plant) and future (The Predictive Plant) operation. Available scenario functionalities include the possibility of modifying controller setpoint, simulating equipment failure and/or maintenance, modifying input data from online sensors, and subsequently evaluating plant performance in comparison with current (baseline) operation. These tools are essential in supporting the training of new operators to familiarizing with plant operations and assessing the impact of operators' decision in a risk-free environment, without interfering with full-scale operations. Furthermore, automated optimization (The Optimized Plant) provides operators with optimal controller settings to minimize KPIs, e.g., energy use, effluent loads, carbon footprint, alone or in combination (multi-objective optimization). The optimization relies on the execution of ensemble simulations for a defined forecast period (1–48 h), where each run uses different values for the (set of) of controller setting(s) of interest given pre-defined ranges. Optimal setting values are identified based on the optimization (e.g., minimization) of the defined objectives (i.e., KPIs). In the multi-objective optimization, different objectives are combined using adequate weighing factors based on the magnitude of the single objectives. Currently, an add-on module for prediction and minimization of N₂O emissions is under development and testing for Bjergmarken WRRF (Roskilde, Denmark).

Key aspects of connecting the two digital twins include data exchange, graphical interfaces, control strategies, and optimization strategies.

Data exchange is set up between FCF (using MIKE Operations) and TwinPlant (using DIMS.CORE) or other systems available at the collections systems or the WRRFs. Both tools have flexible features for data exchange. Depending on the circumstances, secure data exchange may be necessary between several control systems (one or more catchment systems, one or more WRRFs). The most suitable exchange configuration will likely minimize security risks. The amount of data to be exchanged is relatively moderate, so the transfer itself is not likely a limiting factor.

The respective graphical user interfaces for FCF and TwinPlant need to be complemented with pertinent information from the other digital twin. This should be customized for each application and would reflect the need for information within each organisation if the information is crossing organisational boundaries. Traditionally catchments and WWRFs are often operated by different organisations, internally or externally. Thus, analysis, planning, and setting goals can be useful for determining which information should be made available. This work is typically done in a workshop form, involving all personnel who benefit from the new information available.

Once the necessary data are available and the required control strategies defined, optimization strategies can be built and implemented. These can include:

• how to best utilize the catchment with respect to available storage, pumping capacities, combined sewer overflows, etc.

• how to best utilize the WRRF to produce a suitable effluent

• how to minimize the environmental impact on the recipient

• how to optimize energy and chemical usage in the collection system

• how to optimize energy and chemical usage at the WRRF

In the present study, the integration of collection systems–WRRF digital twins was tested for the Rya WRRF (Gothenburg, Sweden). The FCF collection system digital twin currently in operation (Lumley et al. 2024) was coupled with a TwinPlant implementation of the first stage of the WRRF (primary settling, anoxic-aerobic activated sludge biological treatment). A model of the first stage of Rya WRRF was implemented in WEST (DHI A/S, Hørhsolm, Denmark), using the modified ASM2d (Gernaey & Jørgensen 2004) as the underlying process model, and verified based on typical effluent quality from the section of interest. The integrated digital twin includes all the previously mentioned components of the two stand-alone solutions (models, automated data acquisition, tools for visualization, scenario analysis, and control), extended with automated data communication between them. As detailed in Figure 4, the integration between the digital twins is based on the use of inflow data generated by FCF as input to TwinPlant. Inflow time series are automatically generated and sent from FCF to TwinPlant with a 1-h schedule, including measurements (6-h hindcast) and forecasts (up to 48-h in the future) without and with RTC implementation in the catchment. Based on inflow forecasts, TwinPlant provides a prediction of the expected WRRF performance and effluent quality, allowing to (i) evaluate alternative operational decision through what-if scenarios and (ii) compare the performance of the WRRF with and without RTC in the catchment. Applications examples for cases (i) and (ii) are presented in the following section.

The applicability of the integrated digital twin for Rya WRRF was tested by considering two examples of relevance for daily operations. For both examples, predictions for the WRRF performance are based solely on simulation results, being expected to support utility staff in decision-making for the operation of the collection system and the WRRF in combination.

Step 1. FCF feedforwards forecasted flows, levels, and wastewater composition to TwinPlant.

The collection system operators evaluate the impact on the catchment.

Step 2. Based on the influent forecast by FCF, TwinPlant identifies optimal operational settings and setpoints (e.g., DO for aeration control; return sludge flow from secondary settlers). Furthermore, WRRF operators use TwinPlant to evaluate plant performance by considering variations on the initial influent forecast based on the possibilities in terms of catchment control. These variations may include, e.g., a delay in the peak inflow (due to utilization of storage volume in the catchment) or higher inflows for a period of time (in case the hydraulic capacity of the WRRF is sufficient). If any of these variations is shown to achieve further improvement in WRRF performance, TwinPlant feeds back desired conditions (flows, levels, etc.) to FCF.

Step 3. FCF evaluates the feasibility of controlling the collection system to achieve the desired influent conditions indicated by TwinPlant. The collection system operators re-evaluate the impact on the catchment. FCF then feedforwards a new customized forecast of flows, levels, and water quality to TwinPlant (which performs evaluations described in Step 2).

Step 4. FCF sends suggested optimized setpoints (Step 3) to the collection system control system. Collection system operators scrutinize the suggested setpoints and take appropriate action, i.e., acceptance or rejection of the recommendations and, in the first case, implementation of the setpoints in the control system of the catchment.

Step 5. TwinPlant sends suggested optimized setpoints (Step 2) to the WRRF control system.

WRRF operators scrutinize the suggested setpoints and take appropriate action, i.e., acceptance or rejection of the recommendations and, in the first case, implementation of the setpoints in the control system of the WRRF.

In this iterative process, the optimal operation of the collection system and the WRRF can be found to meet the given objectives, e.g., minimize impact on the recipient, minimize energy consumption, and minimize chemical consumption. The forecast time horizon for the optimization can range from 1 h ahead for small and/or fast responding systems (e.g., aeration control) up to 24–48 h ahead for large and/or slow responding systems (e.g., MLSS control). An upper limit of 48 h ahead can be considered as realistic in relation to the reliability of input forecast data (e.g., rainfall), allowing the optimization to rely on accurate model predictions.

Notably, the approach presented here represents a future vision of the integration of digital twins. While the presented application examples have shown that the integration can be achieved incrementally, it is undoubted that a number of challenges need to be addressed to ensure that a full-fledged integration is feasible and valuable. These challenges can be both technical and strategic and include (i) the need to ensure a continuous, stable, and reliable exchange of high-quality data between digital twins; (ii) the need for fast and reliable simulations and optimizations (possibly achievable by integrating mechanistic and data-driven approaches; Schneider et al. 2022); and (iii) the definition of a clear set of common objectives in the joint operation of catchments and WRRFs. The last challenge, in particular, appears critical for the use of connected digital twins in practice, depending on site-specific conditions and organizational setups. For example, recommendations provided by the digital system may lead to conflicting interests between operational teams (e.g., energy savings in WRRF operations may result in increased energy consumption in catchment operations), which may be overcome with the definition of holistic objectives at the utility level.

An alternative approach to the one presented in this study is based on the development of a unique digital twin platform covering catchments and WRRF simultaneously and allowing for a global optimization of the wastewater management system. This alternative approach is technically feasible (process simulation software like SIMBA#, SUMO, or WEST allows for modelling of the integrated urban water system in one platform) and can offer certain benefits (e.g., reduced data exchange). Nevertheless, the integration of different DTs allows to preserve a high level of fidelity in the description of the respective physical systems (hence, a higher controllability) and, perhaps most importantly, it is relevant in the frequent case when collection systems and WRRF operations are managed by different organizations.

Traditionally, catchments and WRRFs are often managed and operated by different organizations or different suborganisation within an organisation. All too often, managers and operators do not have a deeper understanding of their counterpart's system. Connected digital twins can provide a venue for improved collaboration and mutual understanding between organizations. This can be especially valuable when these systems are, for example, during storm events pressed to capacity and many decisions need to be made quickly and in parallel.

To gain the trust of operators, digital twins must be robust, i.e., predictable and plausible. This requires trust in the digital twins if they are to be used to their full potential, perhaps even more in a situation where two digital twins are connected, and two or more organisations need to develop this trust. Previous experience (Lumley et al. 2024) highlights the importance of operator confidence and including them early in the adoption of new technology is advised. Taking into account the factors affecting technology adaptation and how operators respond to and interact with it (Roberts et al. 2021) can be a key to a successful implementation.

Digital twins are powerful control and optimization tools that provide a sound base for modelling, simulating, forecasting, and controlling collection systems and WRRFs. Valuable insight into the collection system and WRRF dynamics can be gained while creating digital twins, and this insight can be used to quantify and optimize combined sewer overflow reductions, flows to the WRRF, WRRF discharge, energy and chemical consumption, etc. according to defined objectives. In this study, we have assessed the benefits and challenges of connecting digital twins for collections systems (FCF) and WRRFs (TwinPlant) through concrete application examples and a long-term vision. The key findings of the study are the following:

• Digital twins are increasingly becoming state-of-the-art digital technologies for the optimal operation, through decision support and RTC, of collection systems and WRRFs – still, in a siloed fashion.

• The integration of digital twins can help achieve proactive and optimized operations of the wastewater management cycle. Assessed use cases included (i) the use of inflow forecasts during extreme wet-weather events to activate emergency treatment steps in WRRFs and (ii) inflow regulation through storage volume utilization in collection systems to optimize WRRF performance.

• The range of possibilities for a more elaborated connection between digital twins is broad, and a future vision was presented. The benefits and the challenges to be addressed when approaching this vision were discussed, and a concrete example of the system-wide optimization of catchments and WRRFs was provided.

Connecting collection system and WRRF digital twins and sharing information can help to bridge the traditional silos between organisations by providing a venue for improved collaboration and mutual understanding between these organisations. Digital twins must be robust to gain operator trust which is essential for optimal operation of the collection system and the treatment plant. Connecting collection system and WRRF digital twins is one step to a more integrated handling of water in cities.

The FCF consortium was partially financed by Vinnova, the Swedish governmental agency for innovation systems, and which is hereby gratefully acknowledged. DHI also acknowledges funding from The Ministry for Education and Research of Denmark (area ‘Hav, Vand og Klimamål’).

Data cannot be made publicly available; readers should contact the corresponding author for details.

The authors declare there is no conflict.