Spatial dependency enhanced dam safety evaluation: a digital-twin based monitoring platform integrating multi-sensor correlation analytics

The analysis of monitoring data plays a critical role in dam safety assessment. The data analysis process typically involves three stages: monitoring point information inspection, single-point analysis and early warning, and multi-point data fusion and evaluation. A key challenge in the multi-point evaluation stage is establishing an objective and reproducible standard that effectively links multi-point information with the dam’s structural behavior. For arch dams, the load-transfer system consists of horizontal arches and vertical girders. This study proposes an assessment system that utilizes three evaluation parameters: the alert level of individual monitoring points, the spatial correlation of alerted points, and the anomaly rate across all points. A deterministic correspondence is established between these parameters and the final score, thereby eliminating subjective judgment. The reliability of the evaluation results was verified using measured data and a typical case study. Furthermore, a digital twin (DT) platform was developed for managing, analyzing, and evaluating monitoring information. The proposed model was integrated into this platform. Both the monitoring data and the inferred dam behavior are visualized. This DT platform has been employed by a real arch dam in China. Operational results demonstrate its capability for integrated real-time analysis and presentation, significantly enhancing the intuitiveness of dam safety monitoring and the efficiency of decision-making.

1 Introduction

Monitoring data provide essential insights into the structural performance of a dam (Li B. et al., 2020). Spatially distributed sensors are employed to capture localized mechanical effects, while time-series data reveal dynamic evolutionary patterns. Modern monitoring systems enable the real-time collection of multi-source parameters, including critical metrics such as deformation, seepage, and stress–strain measurements. Within this context, an integrated indicator system has been established to support comprehensive dam safety assessment (Deng et al., 2025).

Single-point dam monitoring systems have established comprehensive frameworks that combine traditional statistical and deterministic models with machine learning techniques to enable intelligent data filtering, feature extraction, and trend prediction (Dong et al., 2022; Jiang et al., 2024). Statistical models decompose structural responses into hydrostatic, thermal, and time-dependent components (Liu et al., 2023), facilitating quantitative analysis of dam behavior. Deterministic models establish physical constitutive relationships by formulating functions that link environmental loads to structural responses (Yin and Wu, 2022). Meanwhile, deep learning approaches introduce non-linear fitting mechanisms that enhance prediction accuracy, particularly under complex operational conditions (Gu et al., 2022; Cao et al., 2025; Song et al., 2023).

Over the past decade, methods for dam safety assessment have evolved rapidly. The Analytic Hierarchy Process (AHP) creates a hierarchical structure with multi-dimensional indicators like deformation, seepage, and stress (Dai et al., 2022). It turns experts’ experience into calculable judgment matrices, which effectively solves the problem of assigning weights to multiple indicators in complex systems (Alabhya et al., 2025). Its main strengths are clear processes and good use of expert judgment. Yet it relies heavily on subjective input Bayesian Networks (Chen et al., 2024; Zhang et al., 2024) build probability graph models that link environmental loads, structural responses, and damage states. They combine prior knowledge with real-time monitoring data to enable dynamic probability reasoning and risk updates for safety status. This allows dynamic risk updates and handles uncertainty well. However, building the network requires deep expertise, and training it needs lots of data. The Cloud Model (Li W. et al., 2024; Sang et al., 2022; Xiong et al., 2025) uses three numerical features to bridge qualitative concepts and quantitative data. It can turn descriptions like “slight anomaly” into numbers, and also explain numerical results in simple terms. This makes conclusions easier to understand. Still, setting its parameters is complex, and it is sensitive to data distribution. However, little literature has been found on investigation into spatial correlation among multiple monitoring points. Their collective contributions to dam safety have not been adequately investigated. In practice, monitoring points often exhibit coupled mechanical interactions and data redundancy, especially in the case of arch dams. These structures are characterized by their spatially integrated load-transferring mechanisms, where stresses propagate through arch–beam systems across adjacent monitoring zones (Li Y. et al., 2024). Therefore, incorporating spatial correlation analysis among multiple monitoring points is essential for a more accurate dam safety assessment.

Conventional dam integrity assessment systems mainly rely on scheduled manual inspections and experience-based diagnostic protocols. However, these approaches are increasingly recognized as suboptimal in terms of both operational efficiency and evaluative precision, particularly given the demands of modern infrastructure. This technological gap underscores the urgent need to implement intelligent monitoring paradigms capable of delivering real-time, data-driven performance evaluation through automated anomaly detection mechanisms (Li et al., 2025; Li Y.P. et al., 2024). In this context, the digital twin (DT) concept has been introduced into the assessment of building assets due to its ability to provide seamless access to comprehensive and integrated information. Performance improvement of anomaly detection algorithms and lack of interoperability among open-source software are recognized as research gaps in the practical development and implementation of DTs applied for bridges (Jiménez et al., 2023). Some studies have utilized digital models as visual platforms to facilitate intuitive interpretation of structural health monitoring (SHM) data and simplified assessment of structural conditions. For example, Micheal and Ayan (Micheal and Ayan, 2023) explored the integration of Building Information Modeling (BIM) and the Internet of Things (IoT) through the use of an Arduino microprocessor to track and visualize data in both time and frequency domains. Davis (Davis, 2023) proposed an intelligent BIM-enabled DT framework that integrates wireless IoT sensing with communication, digital signal processing, and domain knowledge in structural analysis. This framework can be adapted to various structural elements, including beams, columns, trusses, slabs, arches, bracings, walls, footings, foundations, and girders.

Real-time evaluation of dam performance is essential for the effective management of dam safety (Guru et al., 2022). This requirement has become increasingly urgent due to the rising frequency of extreme flood events driven by climate change, which elevates the risk levels of existing dams (Gregory and Gerald, 2021). In this context, Ding et al. (2025) recently proposed a DT framework for a core-wall rockfill dam, aiming to simulate and predict dam performance in real time. This framework employs a validated finite element method as its foundation. To enhance dam safety monitoring, Shao et al. (2025) developed a DT-based visualization system. The effectiveness of the system was demonstrated through its application in a real-world dam in China. Zhu et al. (2023) also proposed a DT-based approach for managing monitoring-related information and conducting comprehensive evaluation of dam safety. This approach utilizes a multi-rule cloud reasoning model to evaluate measurement data. All the aforementioned literature are related to the development of dam digital twin platforms. Some focus on the review of dam simulation information to evaluate dam safety, while others emphasize the research and development of visualization systems. However, most of the existing studies adopt a qualitative approach when evaluating monitoring information, lacking operational and quantifiable multi-source information evaluation criteria.

The development of a data integration framework capable of processing monitoring data from spatially distributed monitoring points within dam-targeted DT systems requires further investigation. While existing research has considered the correlations among multiple measurement points for data prediction, this aspect has not been sufficiently explored in the context of dam safety evaluation (Cao et al., 2024; Xu et al., 2011; Pi et al., 2024; Wu et al., 2024). Consequently, it is difficult to determine the extent and diffusion of warning signals during dam safety evaluation, which hinders the purpose of comprehensive evaluation and the automation of data processing.

The implementation of real-time predictive frameworks for dam integrity assessment continues to face persistent technical challenges. To address these challenges, this study proposes a DT-based monitoring information management system designed for the comprehensive evaluation of arch dam behavior. A BIM system was incorporated to enable the visualization of monitoring sensors. To characterize spatial correlations among monitoring data, the Manhattan Distance Algorithm (MDA) was employed to assess the similarity of data patterns, with distance metrics serving as a critical determinant. In addition, a non-linear interpolation method was introduced to grade evaluation indices more accurately.

2 The framework for dam safety assessment

A DT project for dam safety assessment generally includes five key components: the information infrastructure, DT platform, network security system, functional modules, and support system (Emina and Adnan, 2025), as illustrated in Figure 1. Among various applications in this framework, intelligent dam safety assessment is a key function, significantly enhancing the capabilities for operational safety assurance and improving the effectiveness of emergency responses during critical events.

Figure 1

Figure 1. The framework for a DT-based dam safety assessment system.

The complete procedure for dam safety assessment usually consists of five sequential stages:

a. Collection of lifecycle engineering information: This step focuses on acquiring monitoring sensor layouts, historical monitoring datasets, and existing dam analysis reports. These data are then integrated into a unified DT-based knowledge repository.

b. Development of a dam safety evaluation indicator framework: This framework is constructed to define the monitored objects, evaluation items, and sensor nodes, systematically addressing essential safety-related monitoring parameters.

c. Formulation of sensor-level evaluation methods: Field-derived monitoring data are compared with predefined safety thresholds to classify the alert levels associated with these data.

d. Establishment of hierarchical evaluation criteria: Multi-level assessment protocols are developed based on established engineering safety evaluation systems. These criteria provide decision-making benchmarks for diagnosing the safety of the dam comprehensively.

e. Comprehensive safety evaluation: Multi-criteria decision analysis is implemented by integrating the evaluation indicator framework, sensor-level evaluation methods, and hierarchical evaluation criteria. The structural integrity of the dam is then quantified using entropy-weighted aggregation algorithms.

It is important to note that this study builds upon existing single-point monitoring and early warning methods. Although early warning levels for individual monitoring points have been previously determined using physical methods, statistical models, and deep learning approaches, they serve in this framework solely as inputs for the proposed multi-point evaluation and early warning system.

3 Development of the dam safety evaluation indicator framework

The safety evaluation indicator framework serves as the foundation for dam safety assessment. Its development requires a comprehensive analysis of the structural characteristics of the dam, the configuration of the monitoring network, and operational performance data. The framework should emphasize the critical role of safety monitoring instrumentation as a core diagnostic tool.

In this study, the indicator system was constructed in accordance with the Guidelines for Safety Evaluation of Hydropower Station Dam Operation (China DL/T 5313-2014), adhering to the principles of scientificity, systematicness, operability, and representativeness. Scientificity is reflected in the selection of indicators based on theoretical analysis and engineering practice, ensuring that they accurately reflect the key factors influencing dam operational safety. Systematicness ensures that the indicator framework is comprehensive and hierarchical, enabling a multi-dimensional evaluation of dam safety conditions. Operability emphasizes the ease of data acquisition and analysis, enhancing the efficiency and feasibility of the evaluation process. Representativeness requires that the selected indicators highlight critical safety parameters and reflect core safety issues in dam operation. By adhering to these principles, the proposed indicator framework not only meets the practical requirements of dam safety evaluation but also provides reliable data support for subsequent analysis and decision-making processes.

4 Multi-sensor spatial correlation analysis

4.1 Multi-dimensional distance metrics

Dam safety monitoring involves a spatially interdependent system composed of heterogeneous sensor networks (Gong et al., 2024). Spatial correlations among multiple monitoring nodes partly reflect the global structural integrity of the dam (Rong et al., 2024). Consequently, detecting spatial interdependencies among anomalous monitoring points plays a critical role in safety diagnostics, particularly for identifying latent systemic risks.

In this study, MDA was employed to quantify multi-dimensional data associations. The advantage of the MDA is considering multi-dimensional indicators to determine the correlation between two measurement points, including spatial distance indicators and data change similarity indicators. Compared with Euclidean distance and others, MDA can reflect the sum of multi-axis distances between two points, and is suitable for judging data correlation under the multi-dimensional indicator evaluation system. Widely applied in computer science, data mining, and transportation logistics (Huang et al., 2022), this algorithm is particularly effective for processing multi-dimensional engineering data.

The computational protocol adopted in this study followed the structured workflow described below:

a. Construction of four-dimensional spatial vectors: Four-dimensional spatial vectors were constructed to represent each monitoring point, incorporating three spatial coordinates (x, y, z) and a real-time monitoring parameter. The vectors for monitoring points L and M were formulated as Equation 1:

$V_L=\left[\begin{array}{c}\begin{array}{c}{L}_1\\ L_2\\ L_3\end{array}\\ L_4\end{array}\right],V_M=\left[\begin{array}{c}\begin{array}{c}{M}_1\\ M_2\\ M_3\end{array}\\ M_4\end{array}\right](1)$

where L₁, L₂, and L₃ denote the spatial coordinates of monitoring point L, while M₁, M₂, and M₃ represent the corresponding coordinates of monitoring point M. L₄ denotes the relative deviation in the monitored parameter for monitoring point L, calculated as Equation 2:

$L_4=\frac{\Delta L_t}{L_{avg}}\times 100\%(2)$

where ΔL_t signifies the increment of the monitored parameter over time interval t, and L_avg represents the annual average value from historical datasets. Similarly, M₄ denotes the relative deviation in the monitored parameter for monitoring point M.

b. Data normalization: To address the dimensional heterogeneity among different monitoring parameters, min-max normalization was performed in Equations 3, 4:

$L_i^{\prime }=\frac{L_i-L_{\mathit{min}}}{L_{\mathit{max}}-L_{\mathit{min}}}(3)$

$M_i^{\prime }=\frac{M_i-M_{\mathit{min}}}{M_{\mathit{max}}-M_{\mathit{min}}}(4)$

where

a. ${\mathrm{L}}_{\mathrm{i}}^{\prime }$ and ${\mathrm{M}}_{\mathrm{i}}^{\prime }$ are the normalized values for monitoring points L and M, respectively;

b. ${\mathrm{L}}_{\mathrm{i}}$ and ${\mathrm{M}}_{\mathrm{i}}$ are the raw sensor measurements;

c. ${\mathrm{L}}_{\min }$ and ${\mathrm{M}}_{\min }$ are the minimum values in the datasets for points L and M; and

d. ${\mathrm{L}}_{\max }$ and ${\mathrm{M}}_{\max }$ are the maximum values in the datasets for points L and M.

e. Computation of inter-node correlations: The spatial correlation between monitoring points L and M was quantified using the Manhattan distance (MD) in Equation 5:

$D_M\left(L,M\right)={\displaystyle \sum _{k=1}^4}\left|L_k^{\prime }-M_k^{\prime }\right|(5)$

The resulting MD was bounded within the interval [0,4], due to the normalization of all vector components ($L_k^{\prime }$ and $M_k^{\prime }$) within the range $\left[0,1\right]$ across the four dimensions (three spatial coordinates and one monitoring parameter).

4.2 Assessment of spatial correlations among monitoring points

4.2.1 Detection of intra-monolith anomalies

Abnormal monitoring points located within the same dam monolith were presumed to exhibit intrinsic spatial correlations due to their shared structural subsystem and common load-transfer mechanisms.

4.2.2 Assessment of inter-monolith correlations

For monitoring points distributed across different dam monoliths, the determination of spatial correlations required a comprehensive evaluation that integrated the geospatial proximity of the alert points and the magnitude of their parameter deviations.

This assessment was conducted through a three-step process. Firstly, the previously calculated MD was normalized to a unitized range of [0, 1]. Second, the normalized MD was inverted to reflect correlation strength. Third, the Pauta criterion (also known as the 3σ rule) was applied to identify statistically significant correlation anomalies. According to this criterion, 99.7% of values in a normally distributed dataset fall within the μ ± 3σ range, where μ represents the mean and σ is the standard deviation derived from historical operational data (Li X. et al., 2020). Therefore, monitoring points with values falling outside this interval were classified as spatially correlated anomalies and prioritized for further inspection.

5 Establishment of hierarchical evaluation criteria

5.1 Establishment of the monitoring information scoring framework

The monitoring information scoring framework proposed in this study incorporates single-point alert levels and multi-point spatial correlations as its primary inputs. By analyzing (a) the spatial distribution patterns of abnormal monitoring points and (b) the temporal evolution trends of abnormal data, the framework determines whether anomalies represent isolated events or spatially clustered patterns. Based on the diagnostic outcomes, the safety status of the dam is classified into five graded tiers, as shown in Table 1.

Table 1

Table 1. The scoring framework for dam monitoring data.

5.2 Establishment of the non-linear interpolation method

Based on the score intervals defined in Table 1, the safety scores for various types of monitoring data were determined through interpolation according to the anomaly rate of each monitoring point. It should be particularly noted that, in real-world situations, when the anomaly rate of monitoring points is low, the safety status of an arch dam may not exhibit significant changes. Conversely, once the anomaly rate exceeds a specific threshold, its impact on overall dam safety may increase significantly. To reflect these non-linear responses in structural safety, the scoring framework proposed in this study adopted a non-linear interpolation method, applying different scoring criteria for each safety level of monitoring items.

An evaluation model for calculating the safety score of a monitoring item is expressed in Equation 6, with the corresponding process curve illustrated in Figure 2. For example, within the “Normal A” safety grade, where the safety scores ranged from 7 to 10 points, the score was interpolated based on the anomaly rate of early warning monitoring points. At low anomaly rates, a linear relationship between the anomaly rate and the safety score was maintained. However, once the anomaly rate exceeded 40%, the score rapidly approached the lower limit of the grade. This non-linear interpolation strategy enables the sensitive detection of critical state transitions and reduces the risk of score distortions associated with linear interpolation, thus improving the accuracy and practical utility of the scoring results.

$\left\{\begin{array}{cc}S=S_{\mathit{max}}-\left(S_{\mathit{max}}-S_{\mathit{min}}\right)\alpha /{\alpha }^{*}& \alpha >{\alpha }^{*}\\ S=S_{\mathit{min}}& \alpha \le {\alpha }^{*}\end{array}\right.(6)$

where

S_max is the maximum score for the safety level

S_min is the minimum score for the safety level

α∗ is the anomaly rate threshold; and

α is the actual anomaly rate of the monitoring point

Figure 2

Figure 2. Process curves showing non-linear responses in structural safety.

In this study, the anomaly rate thresholds adopted for different safety levels of monitoring items were defined as 40%, 30%, and 20%, in descending order of safety status. It should be noted that the selection of critical thresholds is tentative and involves some subjective experience, following the principle that higher safety grades correspond to lower critical thresholds for abnormal conditions.

The software functions proposed in this paper are applicable to the construction of digital twin systems for the safety of arch dam, gravity dam, and earth-rock dam projects, including the establishment of a data base, a model platform, and a knowledge platform. Turning to the dam safety assessment method, it is characterized by considering spatial correlation analysis among multiple monitoring points. Given this characteristic, for gravity dams, the safety assessment method proposed in this paper may be more suitable for the analysis of a specific dam section. However, different from concrete dams such as gravity dams and arch dams, earth-rock dams are prone to safety hazards related to seepage. Therefore, it may be more appropriate to propose a targeted safety assessment method that is compatible with the anti-seepage system.

6 Construction of the DT-based monitoring platform

The technical framework of the DT-driven monitoring platform was built upon the comprehensive safety evaluation module developed in this study. This platform integrates three core DT components: (a) data infrastructure layers, (b) model simulation platforms, and (c) knowledge platforms. Moreover, it incorporates visualization models to support real-time analysis and decision-making processes. The flowchart of the DT-based monitoring platform is shown in Figure 3.

Figure 3

Figure 3. Flowchart of the DT-based monitoring platform.

The data infrastructure layer consolidates multi-domain datasets, including geospatial information, foundational parameters, monitoring metrics, and operational records. It provides unified management of heterogeneous data streams originating from critical acquisition systems such as automated hydrological telemetry networks, structural health monitoring arrays, and routine inspection platforms. The model simulation platform enables unified management, deployment, configuration, and execution of hydraulic engineering models. The dam safety evaluation module is registered within this platform through standardized interfaces that specify its input parameters and output formats. The platform is also interconnected with other critical models, including engineering safety analysis and prediction models, monitoring and early warning models, and structural calculation models. The knowledge platform comprises repositories such as the forecast and dispatch scheme repository and the engineering safety knowledge base. It systematically consolidates multi-source heterogeneous knowledge resources, including sensor evaluation criteria, hierarchical assessment rules, and structural analysis results.

This work is based on a dam safety evaluation model. The model has its own algorithmic logic and assessment process. Our approach follows the overall technical framework for digital twin water conservancy projects. It uses core digital twin engines, including the data base-plate, model platform, and knowledge platform. Visualization models and relevant technical systems are integrated. A dam evaluation module was developed within the intelligent analysis and early warning system for engineering safety. The proposed dam safety evaluation model establishes a five-tier classification system (A+, A, A-, B, C) with corresponding scoring ranges from 0 to 10. The model integrates three critical input parameters: single-point threshold violations, monitoring point anomaly rate, and spatial correlation among monitoring points. The model outputs both quantitative scores and qualitative safety classifications, enabling systematic safety assessment through synthesized analysis of anomaly distribution patterns and their spatial characteristics.

7 Real-world application: a case study on an arch dam

7.1 Development of a DT platform for arch dam safety monitoring

In this study, a concrete arch dam was selected as the research object, and a DT platform featuring a 3D visualization environment was constructed (Figure 4). The arch dam is a concrete double-curvature arch structure with a crest elevation of 988 m and a foundation surface elevation of 718 m at the riverbed. The dam body features relatively thin sections and was constructed without longitudinal joints using full-width concrete placement. Fourteen transverse joints divide the dam into 15 monoliths from the left to the right bank. The dam incorporates five surface spillways and six mid-level outlets for flood discharge. Both the spillways, outlets, and their piers exhibit complex structural configurations. The surface spillways have a weir crest elevation of 959 m, with each orifice measuring 12 m by 16 m.

Figure 4

Figure 4. Digital models of the arch dam used in this study: (a) overall model, (b) dam interior corridor, and (c) observation station.

The dam is equipped with over 1000 monitoring instruments, including direct and inverted pendulums, precision leveling systems, piezometers, strain gauges, and flow measurement weirs. All internal monitoring points are equipped with automated monitoring systems operating at a frequency of four measurements per day, while external monitoring points are typically measured every 3–5 days. These instruments are used to monitor critical safety-related parameters such as deformation, seepage, and stress–strain responses. By integrating BIM results into the 3D visualization environment, the platform intuitively displays the spatial layout of the monitoring instruments, as illustrated in Figure 5. Moreover, as shown in Figure 6, the platform enables interactive data queries linked to real-time monitoring datasets, allowing users to filter data by monitoring item types and access cross-sectional views for enhanced analytical clarity.

Figure 5

Figure 5. Visualization of monitoring point locations within the DT platform.

Figure 6

Figure 6. Visualization of real-time monitoring datasets and the filtering interface.

7.2 Smart safety management based on real-time monitoring information

The smart monitoring information management module offers a wide range of functionalities to support comprehensive dam safety monitoring. It includes sensor type management, allowing users to define, add, modify, or delete different types of sensors (Figure 7a). It also enables sensor information management, where users can configure sensor properties and locations to facilitate rapid data retrieval. For monitoring data management, the module handles both historical and real-time monitoring datasets (Figure 7b). Templates are provided for one-time batch imports, and temporal changes in data can be visualized through interactive charts. Additionally, the module possesses model analysis capabilities, enabling modeling based on historical sensor data, time variables, and environmental conditions (Figure 7c). These inputs are used to generate key computational factors, which are integrated into evaluation models. This process facilitates the validation of sensor functionality and the detection of anomalies.

Figure 7

Figure 7. The interface of the monitoring information management module: (a) sensor type management, (b) monitoring data management, and (c) monitoring model management.

7.3 Development of safety evaluation indicators for arch dams

The dam safety evaluation indicator framework developed in this study is designed specifically for arch dams and encompasses four key aspects: dam deformation, dam stress, seepage pressure, and foundation deformation (Figure 8). This indicator framework has been integrated into the DT platform (Figure 9), and users can customize it according to their project-specific requirements.

Figure 8

Figure 8. The indicator framework for arch dam safety evaluation.

Figure 9

Figure 9. The management interface for dam safety evaluation indicators.

Dam deformation is an important indicator of dam structural stability. It includes four parameters: radial displacement, tangential displacement, vertical displacement, and transverse joint opening. Radial and tangential displacements monitor the horizontal deformation trends of the dam, while vertical displacement reflects settlement or uplift in the vertical direction. Transverse joint opening evaluates changes in the structural joints of the dam. Together, these parameters provide a comprehensive basis for analyzing the overall stability of the dam.

Dam stress focuses on monitoring stress within concrete components. By monitoring the distribution of stress in key sections of the dam, this indicator evaluates the load-bearing capacity and crack resistance of dam materials, providing critical references for assessing the structural safety of the dam.

Seepage pressure is a key factor affecting dam safety and is represented by two parameters: foundation seepage pressure and seepage quantity. Foundation seepage pressure reflects hydraulic changes within the foundation rock mass, while seepage quantity monitors internal seepage within the dam. Together, these parameters provide essential data for analyzing the impact of seepage on dam safety.

Foundation deformation is another key indicator for evaluating foundation stability. It includes three components: deformation of the resistant body, deformation of the abutment rock mass, and foundation deformation in the riverbed section. Resistant body deformation reflects the stability of the main load-resisting structure at the foundation level. Abutment rock mass deformation monitors the deformation trend of the surrounding rock anchoring the dam, while deformation in the riverbed section reflects settlement or uplift in the base of the dam.

Table 2 lists the aforementioned monitoring data type and their corresponding measured sensors.

Table 2

Table 2. The scoring framework for dam monitoring data.

7.4 Multi-sensor spatial correlation analysis

Based on the dam safety evaluation criteria described in Section 5, the dam safety evaluation model employs three primary input parameters: the early warning status of individual monitoring points, the spatial relationships among these points, and the overall anomaly rate. The DT platform automatically accesses real-time monitoring data, identifies early warning levels, and calculates the anomaly rates for 1135 measuring points installed on the dam. Using the method described in Section 4, the platform also determines the presence of spatial correlations. The spatial layout of pendulums and piezometers used in this analysis is illustrated in Figure 10. As an example, monitoring data related to dam deformation and foundation seepage pressure were selected. Monitoring data collected from April 2023 to April 2024 were used to calculate the MD between two early warning points (ZC04-04 and ZC02-12), and the results were visualized using a process curve shown in Table 3; Figure 11. Although both monitoring points triggered early warnings, no significant spatial correlation was observed between them.

Figure 10

Figure 10. The spatial layout of pendulums and piezometers installed on the arch dam (1:500).

Table 3

Table 3. The scoring framework for dam monitoring data.

Figure 11

Figure 11. Spatial correlation results among dam deformation and foundation seepage pressure gauges.

Based on the monitoring data of the case in this paper, there are no measuring points with spatial correlation. However, to illustrate the impact of spatial correlation, two hypothetical cases are presented using horizontal radial displacement as an example:

Case 1: The data of vertical measuring points at the top of Dam Segments 4# and 8# exceed the monitoring indicators, but there is no spatial correlation between them. According to the scoring criteria in Table 1, the safety level is A with a score of 9.91.

Case 2: The two aforementioned measuring points have spatial correlation. In this scenario, the safety level is A-with a score of 6.88.

If the impact of spatial correlation is ignored, and only the single-point early warning level and anomaly rate are used for judgment, the two cases would yield the same score. Nevertheless, their reflected operational performance differs significantly. If the early warning points are not correlated, it indicates the early warning phenomenon may be an isolated case and has not spread. If the early warning points are correlated or geographically close, it suggests the abnormal phenomenon may have spread locally, requiring the attention of management personnel. Therefore, it is necessary to incorporate the spatial correlation of early warning measuring points into the evaluation system.

7.5 Integrated dam safety assessment

The interface of the comprehensive safety evaluation module is shown in Figure 12. This module has been seamlessly integrated into the DT platform and performs key functions such as the construction of the evaluation model, configuration of decision rules, and visualization of evaluation results. Through comprehensive reasoning, the module enables a systematic assessment of dam safety performance. Based on the monitoring data, the dam safety assessment model proposed in this study evaluates the current safety status as “Normal Dam (Grade A)”. This evaluation is consistent with the actual normal operational conditions of the dam, thereby validating the model’s effectiveness. This module provides decision-making support for dam operation and risk management.

Figure 12

Figure 12. The interface of the comprehensive dam safety evaluation module.

8 Conclusion

This study presents a comprehensive method for dam safety evaluation, integrating a novel alert-level classification approach and a non-linear scoring system that considers both single-point anomalies and spatial relationships among multiple monitoring points. A DT platform was developed to manage multi-source monitoring data effectively. The model was applied to an arch dam. Tests using both measured data and hypothetical cases show that the model’s evaluation results align with the dam’s actual operational conditions. The model’s reliability has been verified.

The DT platform supported the intelligent integration of monitoring data. By incorporating BIM results into a 3D visualization environment, it enabled the intuitive display of monitoring instrument layouts and interactive data queries linked to real-time monitoring datasets. The platform dynamically visualized monitoring data, alert levels, and spatial relationships, thus providing a smart “data-knowledge-decision” support system for effective dam safety management.

However, this study has several limitations. Firstly, the dam safety evaluation model relies on a single data source. Monitoring equipment may fail and be difficult to repair, potentially affecting the model’s accuracy and reliability over long-term operation. Simulation data and inspection information have not been fully utilized, and a data-physics dual-driven evaluation model needs to be developed. Secondly, the judgment of spatial correlations depends on the Pauta criterion. Using fixed thresholds may lead to misjudgments, and an adaptive threshold model should be developed.

Data availability statement

The original contributions presented in the study are included in the article/supplementary material, further inquiries can be directed to the corresponding author.

Author contributions

YM: Writing – original draft, Writing – review and editing. LH: Validation, Writing – review and editing, Writing – original draft. ZQ: Investigation, Writing – review and editing. ZT: Investigation, Methodology, Writing – review and editing. JY: Data curation, Writing – review and editing. XD: Formal Analysis, Writing – review and editing. ZD: Data curation, Writing – review and editing. HF: Software, Writing – review and editing.

Funding

The authors declare that financial support was received for the research and/or publication of this article. The authors declare that this study received funding from China Yangtze Power Co., Ltd. (Z152302046) and the National Key R&D Program of China (2024YFC3210704). The funder had the following involvement in the study design, collection, analysis, the writing of this article and the decision to submit it for publication.

Conflict of interest

Authors YM, ZQ, ZT, JY, XD and ZD were employed by China Yangtze Power Co., Ltd.

The remaining authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Generative AI statement

The authors declare that Generative AI was used in the creation of this manuscript. During the preparation of this work, the authors utilized DeepSeek-R1 solely for the purpose of linguistic refinement and polishing of specific text passages (lines 248-276, and 287-295). The AI tool was not employed in any aspect of experimental design, data collection, analysis, interpretation of results, or substantive content generation. The final content remains entirely the responsibility of the authors.

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.

References

Alabhya, S. S. D., Bharti, M. K., Shrimali, T. K., and Datta, T. (2025). Overall dam safety assessment using the analytical hierarchy process model. Structures 80, 109875. doi:10.1016/j.istruc.2025.109875

CrossRef Full Text | Google Scholar

Cao, W. H., Wen, Z. P., Feng, Y. M., Zhang, S., and Su, H. (2024). A multi-point joint prediction model for high-arch dam deformation considering spatial and temporal correlation. Water 16 (10), 1388. doi:10.3390/w16101388

CrossRef Full Text | Google Scholar

Cao, X., Sheng, J. B., Jiang, C., Yuan, D., and Zhang, H. (2025). Concrete dam deformation prediction model considering the time delay of monitoring variables. Sci. Rep. 15 (1), 8458. doi:10.1038/s41598-025-92806-y

PubMed Abstract | CrossRef Full Text | Google Scholar

Chen, Y., Wang, Y. J., Zheng, X. Z., Tian, D., and Jin, H. L. (2024). Risk analysis of dam break accident combining case mining and Bayesian network. J. Hohai Univ. Nat. Sci. 52 (4), 13–21. doi:10.3876/j.issn.1000-1980.2024.04.003

CrossRef Full Text | Google Scholar

Dai, X., Wu, X., Hong, Y., Xie, J., Lin, D., Chen, Y., et al. (2022). Safety and stability evaluation of the uranium tailings impoundment dam: based on the improved AHP-Cloud model. J. Radiation Research Applied Sciences 15 (1), 21–31. doi:10.1016/j.jrras.2022.01.020

CrossRef Full Text | Google Scholar

Davis, A. M. (2023). ICOLD 2023, gothenburg, Sweden: operational safety of dams and reservoirs. Dams Reservoirs 33 (4), 163–166. doi:10.1680/jdare.23.00301

CrossRef Full Text | Google Scholar

Deng, Z. H., Gao, Q., Huang, M. S., Wan, N., Zhang, J., and He, Z. (2025). From data processing to behavior monitoring: a comprehensive overview of dam health monitoring technology. Structures 71, 108094. doi:10.1016/j.istruc.2024.108094

CrossRef Full Text | Google Scholar

Ding, S. L., Pan, J. J., Wang, Y., Xu, H., and Liu, X. (2025). Developing a digital twin for dam safety management. Comput. Geotechnics 180, 107120. doi:10.1016/j.compgeo.2025.107120

CrossRef Full Text | Google Scholar

Dong, K., Yang, D. W., Yan, J. H., Sheng, J., Mi, Z., Lu, X., et al. (2022). Anomaly identification of monitoring data and safety evaluation method of tailings dam. Front. Earth Science 10, 1016458. doi:10.3389/feart.2022.1016458

CrossRef Full Text | Google Scholar

Emina, H., and Adnan, I. (2025). Quantifying durability and failure risk for concrete dam–reservoir system by using digital twin technology. Computation 13 (5), 118. doi:10.3390/computation13050118

CrossRef Full Text | Google Scholar

Gong, S. L., Sun, F. T., Huang, W., Chen, K., and Shen, H. Y. (2024). Intelligent monitoring and early warning method for dam safety based on knowledge graph. Adv. Eng. Sci. 56 (3), 32–40. doi:10.15961/j.jsuese.202300710

CrossRef Full Text | Google Scholar

Gregory, B. B., and Gerald, E. G. (2021). US flood risk management in changing times. Water Policy 23 (1), 202–215. doi:10.2166/wp.2021.269

CrossRef Full Text | Google Scholar

Gu, C. S., Wang, Y. B., Gu, H., Hu, Y., Yang, M., Cao, W., et al. (2022). A combined safety monitoring model for high concrete dams. Appl. Sci. 12, 12103. doi:10.3390/app122312103

CrossRef Full Text | Google Scholar

Guru, P., Revanth, D., Mohamed, B., and Sadhu, A. (2022). Recent advancement of concrete dam health monitoring technology: a systematic literature review. Structures 44, 766–784. doi:10.1016/j.istruc.2022.08.021

CrossRef Full Text | Google Scholar

Huang, X., Han, D., Weng, T. H., Wu, Z., Han, B., Wang, J., et al. (2022). A localization algorithm for DV-Hop wireless sensor networks based on manhattan distance. Telecommun. Syst. 81 (2), 207–224. doi:10.1007/s11235-022-00943-w

CrossRef Full Text | Google Scholar

Jiang, Z. H., Gu, H., Fang, Y., Shao, C., Lu, X., Cao, W., et al. (2024). Three optimization methods for preprocessing dam safety monitoring data using machine learning. Struct. Control Health Monit. 2024 (1), 4385464. doi:10.1155/stc/4385464

CrossRef Full Text | Google Scholar

Jiménez, R. A., Plevris, V., and Nogal, M. (2023). Bridge management through digital twin-based anomaly detection systems: a systematic review. Front. Built Environ. 9, 1176621. doi:10.3389/fbuil.2023.1176621

CrossRef Full Text | Google Scholar

Li, B., Yang, J., and Hu, D. X. (2020). Dam monitoring data analysis methods: a literature review. Struct. Control Health Monit. 27 (3), e2501. doi:10.1002/stc.2501

CrossRef Full Text | Google Scholar

Li, X., Li, Y. L., Lu, X., Wang, Y., Zhang, H., and Zhang, P. (2020). An online anomaly recognition and early warning model for dam safety monitoring data. Struct. Health Monit. 19 (3), 796–809. doi:10.1177/1475921719864265

CrossRef Full Text | Google Scholar

Li, W., Gu, X. B., Yang, C., and Zhao, C. (2024). Level evaluation of concrete dam fractures based on game theory combination weighting-normal cloud model. Front. Mater. 11 (11), 1344760. doi:10.3389/fmats.2024.1344760

CrossRef Full Text | Google Scholar

Li, Y., Zhao, E. F., Hu, L. Z., Shao, C., Dong, Z., and Zhong, Q. (2024). Assessment on safety and performance of super-high arch dam with asymmetric foundation and its reinforcement effect identification. Structures 70, 107812. doi:10.1016/j.istruc.2024.107812

CrossRef Full Text | Google Scholar

Li, Y. P., Shen, D. B., Pei, L., Lu, X., and Huang, H. (2024). Online intelligent perception, detection, and application of safety monitoring data: taking the right bank slope of PB dam as an example. Struct. Health Monit. 14759217241295989. doi:10.1177/14759217241295989

CrossRef Full Text | Google Scholar

Li, S. P., Zhang, B., Tong, G. Q., Li, Y., Liu, Z., Shi, B., et al. (2025). Online intelligent monitoring system and key technologies for dam operation safety. Adv. Civ. Eng. 2025, 9983255. doi:10.1155/adce/9983255

CrossRef Full Text | Google Scholar

Liu, X., Li, Z., Sun, L., Khailah, E. Y., Wang, J., and Lu, W. (2023). A critical review of statistical model of dam monitoring data. J. Build. Eng. 80, 108106. doi:10.1016/j.jobe.2023.108106

CrossRef Full Text | Google Scholar

Micheal, S., and Ayan, S. (2023). Visualization of structural health monitoring information using internet-of-things integrated with building information modeling. J. Infrastructure Intell. Resil. 2 (3), 100053. doi:10.1016/j.iintel.2023.100053

CrossRef Full Text | Google Scholar

Pi, L. L., Yue, C. F., and Shi, J. C. (2024). A multi-point correlation model to predict and impute earth-rock dam displacement data for deformation monitoring. Buildings 14 (12), 3780. doi:10.3390/buildings14123780

CrossRef Full Text | Google Scholar

Rong, Z., Pang, R., Xu, B., and Zhou, Y. (2024). Dam safety monitoring data anomaly recognition using multiple-point model with local outlier factor. Automation Constr. 159, 105290. doi:10.1016/j.autcon.2024.105290

CrossRef Full Text | Google Scholar

Sang, L., Wang, J., Sui, J., and Dziedzic, M. (2022). A new approach for dam safety assessment using the extended cloud model. Water Resour. Manag. 36 (15), 5785–5798. doi:10.1007/s11269-022-03124-1

CrossRef Full Text | Google Scholar

Shao, C. F., Zheng, S., Xu, Y. X., Gu, H., Qin, X., and Hu, Y. (2025). A visualization system for dam safety monitoring with application of digital twin platform. Expert Syst. Appl. 271, 126740. doi:10.1016/j.eswa.2025.126740

CrossRef Full Text | Google Scholar

Song, J. T., Liu, Y. H., and Yang, J. (2023). Dam safety evaluation method after extreme load condition based on health monitoring and deep learning. Sensors 23 (9), 4480. doi:10.3390/s23094480

PubMed Abstract | CrossRef Full Text | Google Scholar

Wu, B. B., Niu, J. T., Deng, Z. P., Li, S., Jiang, X., Qian, W., et al. (2024). A multiple-point deformation monitoring model for ultrahigh arch dams using temperature lag and optimized gaussian process regression. Struct. Control Health Monit. 2024 (1), 2308876. doi:10.1155/2024/2308876

CrossRef Full Text | Google Scholar

Xiong, Y., Wang, K., Cheng, Y., Zhou, C., and Zhao, M. (2025). Research on dam health diagnosis methodology based on a cloud model and enhanced dempster-shafer evidence theory. Insight Non-Destructive Test. and Cond. Monit. 67 (6), 343–353. doi:10.1784/insi.2025.67.6.343

CrossRef Full Text | Google Scholar

Xu, B., Xia, H., and Yao, Y. (2011). Quantitative and comprehensive evaluation of qualitative indices in dam safety monitoring. Adv. Measurement Test, 301-303, 924–931. doi:10.4028/www.scientific.net/AMR.301-303.924

CrossRef Full Text | Google Scholar

Yin, C., and Wu, Z. Y. (2022). Separate modeling technique for deformation monitoring of concrete dams. Struct. Health Monit. 21 (6), 2968–2989. doi:10.1177/14759217221079013

CrossRef Full Text | Google Scholar

Zhang, H., Li, Z., Ge, W., Zhang, Y., Wang, T., Sun, H., et al. (2024). An extended Bayesian network model for calculating dam failure probability based on fuzzy sets and dynamic evidential reasoning. Energy 301, 131719. doi:10.1016/j.energy.2024.131719

CrossRef Full Text | Google Scholar

Zhu, X., Bao, T. F., Yeoh, J. K. W., Jia, N., and Li, H. (2023). Enhancing dam safety evaluation using dam digital twins. Struct. Infrastructure Eng. 19 (7), 904–920. doi:10.1080/15732479.2021.1991387

CrossRef Full Text | Google Scholar