Predictive assessment of ageing concrete gravity dams using feedback-driven modelling

1 Introduction

Ageing concrete dams pose an increasing challenge to infrastructure safety and resilience worldwide. Among the nearly 59,000 large dams, about 10%–15% are concrete gravity dams, and more than half of them are older than 50 years (International Commission on Large Dams, 2023). The ageing of dams is governed by time-dependent degradation mechanisms that progressively alter their material and structural characteristics. These include creep, thermal effects, chemical deterioration, and Alkali–Aggregate Reaction (AAR), which can lead to reductions in stiffness, cracking, and long-term deformation. Therefore, dam safety assessment cannot rely only on the original design condition or on periodic inspection of isolated response quantities. Cases such as Maich Fishery Dam, Sayano–Shushenskaya Dam, Blue Pond Dam, Bedford Weir, Kosi Barrage, and Derna & Mansour Dams highlight the serious consequences of inadequate assessment and maintenance (Nasser et al., 2025). These events have resulted in significant loss of life and property. Apart from failures, there are decommissioned dams (Custódio et al., 2025) and dams approaching decommissioning owing to ageing-related concerns (Perera et al., 2021).

Among the degradation mechanisms affecting concrete dams, AAR has received particular attention for its progressive, irreversible, and spatially non-uniform nature, since temperature, moisture availability, restraint, and stress state all vary within the dam body (Léger et al., 1996; Comi et al., 2009; Pan et al., 2013a). Such time-dependent, spatially variable behaviour makes ageing-dam assessment a continuously evolving problem rather than a static evaluation task.

Current dam assessment practices rely mainly on monitoring systems and numerical modelling approaches. Instrumentation systems measure parameters such as displacement, stress, seepage, and temperature. However, the collected data are discrete, spatially limited, and often interpreted empirically (Sharan et al., 2025). Numerical models have advanced significantly in simulating degradation processes and structural response under complex loading conditions. In line with this, structural health monitoring (Oliveira and Alegre, 2020; Sevieri and De Falco, 2020; Giahi et al., 2023), digital-twin modelling (Conde López et al., 2021), Bayesian calibration (Sevieri et al., 2021), and data assimilation (Qu et al., 2023) approaches have improved the interpretation of infrastructure behaviour. Recent studies on concrete gravity dams have further shown that seismic sequences, material and reservoir-level uncertainty, dam–water–sediment–foundation interaction, foundation flexibility, and response-based damage indices can significantly influence predicted dam performance, damage demand, and reliability (Wang et al., 2017; Pouraminian et al., 2020; Srivastava and Sahoo, 2023; Hussein et al., 2024a; b; Hussein et al., 2025a; b; Hussein et al., 2026). But for AAR-affected concrete gravity dams, these developments are still not commonly organised into a degradation-specific workflow that continuously links environmental exposure, material deterioration, measured response, parameter updating, and forward prediction.

This paper argues that the lack of systematic interaction between AAR-degradation modelling and monitoring limits the reliable tracking of degradation evolution in ageing dams. It proposes an AAR-specific, feedback-driven framework rather than a new updating algorithm. The framework provides a domain-specific organisation of existing modelling, monitoring, and updating techniques tailored to the particular physics and observability constraints of AAR. By linking physics-based modelling, instrumentation data, residual evaluation, and parameter updating, it aims to improve the assessment of structural performance, degradation progression, and risk in concrete gravity dams.

2 Current practices and their limitations

Existing modelling approaches for ageing concrete dams have largely focused on representing degradation mechanisms. Many studies focus on AAR-induced expansion and damage and evaluate their effects on material properties and structural response (Huang and Pietruszczak, 1999; Pan et al., 2013a; Mohammadpour, 2017; Saouma et al., 2020; Gallyamov et al., 2023). Several numerical frameworks have been developed for deterioration modelling, including chemo-mechanical, thermo-chemo-mechanical, damage mechanics-based, and meso-scale particle modelling approaches (Ulm et al., 2000; Comi et al., 2009; Pan et al., 2013b). These approaches have improved the understanding of ageing behaviour and degradation-induced response in concrete dams. They are generally implemented as periodically applied predictive tools and are rarely updated systematically using field observations. Therefore, they have limited capability to represent the evolving structural condition of real dams over time.

Extensive monitoring programmes report measurable deformations such as crest displacement, joint opening, and cracking near galleries (Plusquellec et al., 2018; Arora et al., 2019; Ftima and Yildiz, 2023), typically obtained through periodic inspection, geodetic survey, and embedded instrumentation measuring displacement, stress, seepage, temperature, and pore pressure. These measurements remain spatially discrete and temporally discontinuous, and many dams rely on legacy systems with incomplete coverage and inconsistent acquisition frequency (Adamo et al., 2021), so monitoring largely remains observational and is interpreted independently of predictive models.

Recent advancements in health monitoring of dams incorporate seismic and structural health monitoring systems, ambient vibration-based modal analysis (Oliveira and Alegre, 2020), Bayesian inference frameworks (Sevieri and De Falco, 2020), machine/deep learning models (Kang et al., 2017; Qu et al., 2019), and data-driven risk assessment approaches using multivariate statistical models (Shao et al., 2019). High computational demands, limited deployment of dynamic monitoring in practice, sensitivity to environmental variability, and reliance on indirect response indicators often constrain these methods. No existing monitoring framework explicitly captures the evolution of time-dependent material degradation across the dam body and over time.

The current assessment practices operate through disconnected modelling and monitoring approaches. Numerical models provide predictive capability but lack continuous validation, whereas monitoring systems capture structural response but do not reliably forecast future behaviour. This separation limits the ability to track the evolution of degradation and to assess the long-term risk and reliability of ageing concrete dams.

3 Proposed framework

Dam behaviour should be interpreted as a continuously evolving degradation state governed by coupled environmental, operational, and mechanical effects: temperature variation, moisture transport, reservoir fluctuation, uplift pressure, and mechanical loading influence both deterioration mechanisms and structural response over time. Assessment should therefore move beyond isolated evaluation of individual parameters toward sequentially updated tracking of degradation evolution across the service life of the structure.

A feedback-driven framework is proposed in which modelling, monitoring, and updating operate as interacting domains (Figure 1). In the modelling domain, geometry, spatial material properties, environmental actions, reservoir loading, uplift, and boundary/foundation-interaction conditions are assigned for numerical simulation, generating displacement, stress, strain, crack-related response, and modal outputs comparable with monitoring quantities. In the monitoring domain, data from existing and planned instruments are acquired, filtered, quality-checked, and converted into response parameters after environmental and operational normalisation; reservoir level, temperature, uplift pressure, seepage, and vibration response are transferred to the modelling domain as loading, boundary-condition, or calibration data. Because AAR is not directly measured by a single sensor, the instrumentation supporting this loop combines thermal/moisture sensing (thermometers, humidity probes), hydraulic monitoring (reservoir gauges, piezometers, uplift cells, seepage weirs), global-deformation survey (pendulums, total stations), local strain and stiffness sensing (vibrating-wire and fibre-optic strain gauges, extensometers), crack/joint meters, and vibration sensing (accelerometers) supporting regional stiffness identification.

The feedback loop itself is a residual-based inverse problem. The modelling and monitoring domains are compared through a residual vector built from paired response quantities such as crest-displacement, joint-opening, and strain histories. When the residual exceeds a prescribed acceptance tolerance, the updater module revises the uncertain parameter set; elastic modulus, AAR-induced expansion strain, joint stiffness, damping, boundary conditions, and foundation-interaction parameters by minimising a weighted residual norm, with the specific algorithm (deterministic least-squares minimisation, Bayesian updating, Kalman-filter-type data assimilation, or surrogate-assisted inverse modelling) selected according to data volume, noise characteristics, and available computational resources. The revised model is re-simulated until residual acceptance is achieved, and the converged model represents the updated degradation state, usable for risk assessment, inspection planning, rehabilitation prioritisation, reliability evaluation, and forward prediction.

Direct stress measurement is not treated as the primary monitoring requirement; stress redistribution is instead inferred from the calibrated model, since small stress variations are difficult to detect reliably with direct stress meters in mass concrete. Vibration monitoring is likewise not treated as a standalone AAR indicator, because natural frequencies reflect global dynamic behaviour and may be insensitive to local degradation; mode shapes, modal strain energy, damping, and transmissibility indices are used instead to support regional stiffness identification and model updating.

The framework also accounts for the coupled influence of temperature, moisture, and stress on AAR progression; since reservoir level, seasonal temperature, and moisture availability vary over time, the rate and distribution of AAR-induced expansion vary correspondingly, while compressive stress restrains or modifies expansion and cracking, altering structural response. Integrating these effects with monitoring data allows the updated model to identify zones of accelerated degradation, evaluate deformation and stiffness-loss trends, and support forward prediction under future environmental and operational scenarios.

4 Discussion

Practical implementation may begin by linking existing monitoring records with a physics-based numerical model and progressively increasing the level of updating as data quality and computational resources improve. The updating frequency may be governed by reservoir operation, environmental variability, degradation rate, and data availability. Published studies demonstrate that AAR can produce measurable response changes. For example, Pan et al. (2014) reported that, after 20 years of AAR deterioration, the Fontana gravity dam showed an increase in maximum horizontal crest displacement from 1.0 cm to 1.8 cm and vertical displacement from 0.24 cm to 0.35 cm under the 1967 Koyna earthquake. Such shifts can serve as response residuals for model updating and decision support.

The framework is presented as a conceptual and algorithmic perspective rather than as a validated case study. Formal validation of the framework is identified as an important direction for future research. This should include implementing the proposed updating algorithms, testing parameter identifiability, and benchmarking the framework using data from an instrumented AAR-affected dam. Future work should also extend the framework to three-dimensional and multi-hazard assessments, including combined seismic and AAR effects, and use high-fidelity AAR models to generate synthetic monitoring data when field records are restricted. In addition, ensemble or Bayesian model-averaging techniques could be used to quantify uncertainty in the estimated degradation state rather than relying on single-point estimates.

The framework’s applicability and its uncertainties are closely linked. Applicability is strongest where displacement, strain, or crack/joint instrumentation already exists with reasonable spatial coverage. It is also more suitable when degradation is at an early-to-moderate stage. Its applicability becomes weaker when instrumentation is sparse or when structural distress is already severe enough to require urgent intervention rather than iterative calibration. Uncertainty arises from several sources: inverse-problem non-uniqueness (equifinality), where different combinations of elastic modulus, expansion strain, and boundary stiffness can reproduce similar observed responses; measurement noise, sensor drift, and inconsistent acquisition frequency across heterogeneous instrument types; and the difficulty of isolating AAR’s contribution when multiple coupled degradation mechanisms act simultaneously (Bangert et al., 2003). Incomplete construction records and initial material data add further uncertainty to model calibration, and continuous multi-physics updating over long service lives carries significant computational demand and convergence-related difficulty. Practical implementation may additionally be constrained by economic considerations, institutional adoption, and the legal or operational protocols governing dam-safety data.

5 Conclusion

Ageing of concrete dams is a continuously evolving process governed by interacting degradation mechanisms that progressively alter structural behaviour. Although degradation modelling, monitoring, and instrumentation technologies have all advanced, they often remain disconnected in current dam-assessment practice, so the evolving condition of ageing dams is not continuously represented or systematically updated using field observations. This paper argues that dam degradation should be interpreted as an evolving structural state influenced by coupled environmental and mechanical effects. The proposed feedback-driven framework links physics-based modelling with monitoring and instrumentation data to support continuous tracking of degradation progression rather than periodic condition evaluation alone. Realising the full potential of the framework requires implementing the proposed algorithms, assessing whether the model parameters can be reliably identified, and validating the approach using either real case-study data or synthetic data. It can also assist risk-informed maintenance planning, rehabilitation prioritisation, inspection scheduling, and early warning of abnormal structural response. Despite remaining uncertainties and implementation challenges, the proposed perspective provides a basis for more reliable and continuously updated assessment of ageing concrete gravity dams.

Generative AI statement

The author(s) declared that generative AI was not used in the creation of this manuscript.

Any alternative text (alt text) provided alongside figures in this article has been generated by Frontiers with the support of artificial intelligence and reasonable efforts have been made to ensure accuracy, including review by the authors wherever possible. If you identify any issues, please contact us.

Publisher’s note

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.

Footnotes

Edited by: Rocio L. Segura, Polytechnique Montréal, Canada

Reviewed by: Thulfiqar Salim Hussein, The Islamic University of Al Diwaniyah, Iraq

Saket Rusia, Rajkiya Engineering College, Mainpuri, India