1. Introduction to the Special Issue
Ensuring the safety of hydraulic structure engineering is of paramount importance, as these infrastructures play a critical role in water management, flood control, and the provision of clean water for various human and ecological needs [
1,
2]. It is becoming increasingly difficult to ignore the impact of dam structural safety and operational reliability on public safety, environmental sustainability, and economic stability [
3,
4]. Failure to adequately safeguard these structures can lead to catastrophic consequences, including loss of life, environmental degradation, and significant financial losses [
5]. Therefore, the continuous safety assessment, maintenance, and improvement of hydraulic structure engineering are essential to mitigate risks, adapt to changing climatic conditions, and sustain the socio-economic benefits they provide [
6]. Robust safety monitoring methods and innovative health diagnostic technologies are crucial to ensuring that hydraulic structure engineering remains resilient and effective in the face of evolving challenges [
7].
Consequently, it is critical to propose safety monitoring methods for hydraulic engineering, as well as guarantee their secure, stable, and effective operation. With the gradual transformation of hydraulic engineering from digitization and intelligence to WISDOM, remote sensing technology, artificial intelligence and deep learning methods have been widely used for automatic perception, processing, storage and analysis of hydraulic structure engineering monitoring data [
8]. These novel methods and technologies, to some extent, compensate for the limitations of traditional safety monitoring approaches. The advent of remote sensing technologies such as three-dimensional tilt photography offers the opportunity to build an integrated hydraulic engineering monitoring and acquisition system capable of capturing all the details of hydraulic engineering. With the introduction of artificial intelligence and deep learning methods, the hydraulic engineering information was analyzed and exploited efficiently. Combined with the traditional hydraulic structure behavior analysis methods, such as geotechnical testing and numerical simulation, artificial intelligence and deep learning methods can help solve more complex hydraulic engineering problems by providing more accurate and professional intelligent analysis and ubiquitous hydraulic engineering services of great theoretical importance and application value in order to achieve the general improvement of safety monitoring of hydraulic structures [
9].
The aim of this Special Issue is to focus on artificial intelligence, deep learning methods and remote sensing technologies in the safety monitoring of hydraulic structure engineering. Since the announcement of the call for papers in April 2023, a total of 15 original papers have been accepted for publication following a thorough peer review procedure (Manuscripts 1–15); the contents cover the analysis methods of hydraulic engineering structural behavior, technologies of dam crack detection, gross error identification, artificial forecast of streamflow, application of UAVs in morphology measurement of rivers, temperature control index, and stability study of engineering structure. To enhance comprehension of this particular edition, we have succinctly outlined the key points of the published articles below.
2. Summary of the Contributions in This Special Issue
In the field of hydraulic structure engineering research, scholars have endeavored to ensure the long-term safety and health of projects through a variety of research methods and technological approaches. These studies pay attention to the applications of artificial intelligence, deep learning methods and remote sensing technologies in the safety monitoring of the entire system.
Gross error detection plays a vital role in ensuring accuracy of structural characteristic analysis [
10]. In order to identify gross errors in dam monitoring data, Contribution 1 introduced a novel method that integrates the Fuzzy C-Means clustering algorithm (FCM), Ordering Points To Identify the Clustering Structure (OPTICS) and Local Outlier Factor (LOF) algorithm. The proposed approach operates in several stages. Initially, the
FCM algorithm is employed to achieve the division of measurement point areas. Subsequently, the OPTICS and LOF algorithms are jointly utilized to identify the gross errors. The final step involves the confirmation of actual gross errors by analyzing the temporal coincidence of detected anomalies across measurement points within the same area. A case study demonstrated the efficacy of this method in accurately identifying spurious gross errors attributable to environmental anomalies. The findings highlight a significant enhancement in gross error detection accuracy and a notable reduction in the misjudgment rate of such errors.
In the realm of structure stability research, Contribution 2 systematically provided a comprehensive review of the advancements and achievements related to the foundational theories concerning the stability of coal pillar dams and artificial dams in CMURs. This review encompasses the mechanical properties under hydraulic coupling, the evolution mechanisms and impact relationships of the “three fields” (stress field, fracture field, and seepage field), and the optimized design of dam dimensions. Clarifying the instability and failure patterns of CMUR dam structures is crucial to ensuring dam structure stability. Contribution 3 delved into the stability assessment of a double-row steel sheet pile cofferdam structure under soft ground conditions. The stability of the cofferdam design solution was evaluated by using a model that incorporates factors such as the coordination of independent pile top displacement, as well as the m-value for backfilled sand and the thrown rock body. The internal force and displacement results of the cofferdam under different working conditions are obtained. And the entire construction process was analyzed using the finite element method. The findings assist in determining whether the overall and overturning stability of the cofferdam meet the pertinent safety requirements.
In the field of dam crack detection, Contribution 4 proposed an improved residual neural network (ResNet)-based algorithm for concrete dam crack detection through dynamic knowledge distillation, which was aimed at achieving higher accuracy for small models. Preliminary experiments on the mini-ImageNet dataset demonstrated the effectiveness of this approach. ResNet18 was trained by incorporating additional tasks to match the soft targets generated by ResNet50 under dynamically high temperatures. These pre-trained teacher and student models were then applied to concrete crack detection, achieving an accuracy of 99.85%, which is an improvement of 4.92%. Contribution 5 presented a pure Vision Transformer (ViT)-based dam crack segmentation network (DCST-net). The DCST-net employs an improved Swin Transformer (SwinT) block to enhance long-range dependencies within a SegNet-like encoder–decoder architecture. The Swin Transformer (SwinT) is a type of deep learning model based on the Transformer architecture, specifically designed for computer vision tasks. Developed by Microsoft Research Asia. Additionally, a weighted attention block to facilitate side fusion between the symmetric pair of encoders and decoders in each stage to sharpen the edge of crack is utilized. The DCST-net was tested on a self-built dam crack dataset and two publicly available datasets, outperforming mainstream methods in both visualization and most evaluation metrics, demonstrating its potential for practical application in dam safety inspection and maintenance.
In the construction of a structural behavior prediction model, Contribution 6 introduced an improved long short-term memory (LSTM) model and weighted Markov model to predict sluice deformation. In the method, the seagull optimization algorithm (SOA) is utilized to optimize the hyper-parameters of the LSTM, enhancing the model’s accuracy. Subsequently, the relevant error sequences of the fitting results of SOA-LSTM model are classified and the Markovity of the state sequence is examined. Then, the autocorrelation coefficients and weights of each order are calculated to predict future random states of sluice deformation, resulting in a proposed SOA-LSTM-weighted Markov model. This model demonstrated superior prediction ability and accuracy compared to the SOA-LSTM and stepwise regression models, as evidenced by analysis of an actual sluice project in China. Contribution 7 established the MHHO-BiLSTM statistical prediction model of sluice seepage. This model incorporates water pressure, rainfall, and aging effects as input data. The parameters of the BiLSTM neural network are optimized using the improved Harris Hawks optimization algorithm. Empirical analysis using sluice seepage data indicated that the proposed optimization algorithm effectively searches for optimal parameters, enhancing the prediction accuracy and robustness of the BiLSTM model. Contribution 8 applied the random coefficient model of panel data into the analysis of sluice deformation, addressing the unobservable overall and individual effects. Moreover, the maximum entropy principle is used to approximate the probability distribution function of individual effect extreme values and to determine the early warning indicators, facilitating the assessment and analysis of non-uniform deformation states. A case study demonstrated the method’s effectiveness in identifying overall deformation trends and deviations at individual measurement points. Contribution 9 aimed to propose a settlement prediction model for CFRDs combining the variational mode decomposition (VMD) algorithm, long short-term memory (LSTM) network, and support vector regression algorithm (SVR). Primarily, VMD is applied in the decomposition of dam settlement monitoring data to reduce its complexity. Furthermore, feature information on settlement time series is extracted. Secondly, the LSTM and SVR are optimized by the Harris Hawks optimization (HHO) algorithm and modified least square (PLS) method to mine the major influencing factors and establish the prediction model with higher precision. Applied to the Yixing CFRD, the model demonstrated superior predictive performance compared to other models, highlighting its potential for evaluating settlement trends and safety states of CFRDs. Contribution 10 proposed an artificial intelligence-based process for predicting concrete dam deformation. Initially, using the principles of feature engineering, the preprocessing of deformation safety monitoring data is conducted. Subsequently, employing a stacking model fusion method, a novel prediction process embedded with multiple artificial intelligence algorithms is developed. Moreover, three new performance indicators are introduced to provide a comprehensive assessment of the model’s effectiveness. An engineering case study confirmed that the ensemble AI method outperformed traditional statistical models and single machine learning models in both fitting and predictive accuracy, providing a robust foundation for concrete dam deformation prediction and safety monitoring.
On certain occasions, the finite element method proves invaluable for analyzing dam structural characteristics [
11]. Contribution 11 utilized a finite element model based on seepage–stress coupling theory to investigate the variations in the phreatic line, earth pressure, and deformation of a core wall rockfill dam due to rapid fluctuations in the reservoir’s water level. Additionally, the results of the finite element simulation were compared with and analyzed alongside safety monitoring data, providing a scientific basis for assessing the health and ensuring the long-term safety of core wall rockfill dams in pumped storage power facilities. Furthermore, artificial intelligence algorithms have been applied to integrated time-dependent analyses of hydraulic structures. Contribution 12 developed a comprehensive superstructure–foundation–backfill model to investigate the time-dependent displacement and stress of a lock head project on a soft foundation during construction. Finite element analyses incorporated a transient thermal creep model for concrete and an elasto-plastic consolidation model for the soil. The modified Cam-clay model was used to describe the soil’s elasto-plastic behavior. Subsequently, global sensitivity analyses were conducted to determine the relative importance of model parameters on the system’s response, utilizing Garson’s and partial derivative algorithms based on the backpropagation (BP) neural network.
Machine learning and statistical methods are increasingly utilized in streamflow forecasting. Contribution 13 conducted a rigorous investigation into the effectiveness of three machine learning techniques and two statistical approaches using streamflow data from the Göksu Stream in the Marmara Region of Turkey, covering the period from 1984 to 2022. This comparative analysis aimed to advance the methodologies used in streamflow prediction. The results revealed that the model generated using the XGBoost algorithm outperformed other machine learning and statistical techniques. Consequently, the models developed in this study demonstrate a high level of accuracy in predicting streamflow within the river basin system, offering significant contributions to the field of hydrological forecasting.
It is important to adopt scientific safety control criteria for dams in the construction period according to practical experience and theoretical calculation [
12]. Contribution 14 synthetically introduced a novel methodology combining information entropy and a cloud model to develop in situ observation data-based temperature control indexes from a spatial field perspective. The order degree and the disorder degree of observation values are defined according to the probability principle. Information entropy and weight parameters are combined to describe the distribution characteristics of the temperature field, with weight parameters optimized via projection pursuit analysis (PPA), leading to the construction of temperature field entropy (TFE). Based on this framework, multi-level temperature control indexes were established using a cloud model. Finally, a case study was conducted to verify the performance of the proposed method. The calculation results indicated that the change law of TFEs aligned well with actual conditions, demonstrating the reasonableness of the established TFE. The cloud model showed broader applicability compared to typical small probability methods, and the derived temperature control indexes significantly enhanced the safety management level of high concrete dams. The research outcomes provide a scientific reference and technical support for temperature control standards in similar projects.
In order to analyze the morphometry of drifting ice on the water surface, Contribution 15 employed UAV technology as a cost-effective and efficient method for describing ice phenomena in lowland rivers. This methodology focused on measuring the area, average size, perimeter, and circularity of frazil ice floes. Individual frames captured by a UAV were analyzed using statistical techniques to obtain these measurements. In prior research, the team successfully assessed ice velocity on a similar test sample. By deriving the average velocity, surface area, and fundamental morphological characteristics of frazil ice, the study facilitated the automated segmentation, classification, and prediction of potential risks related to ice blockages in water routes. These risks include waterway obstructions, infrastructure damage, and potential threats to human safety.