The construction cost rate serves as a core quantitative indicator of safety investment in engineering projects. However, traditional prediction methods struggle to address issues such as high-dimensional data redundancy and nonlinear coupling among multiple factors. Therefore, this paper builds a prediction model for building construction cost rates by integrating data dimensionality reduction and parameter optimization. The model first applies principal component analysis and linear discriminant analysis to reduce data dimensionality and enhance feature discrimination. It then uses an improved gray wolf optimizer to optimize the parameters of support vector regression, allowing accurate capture of the complex relationships between multiple factors and construction cost rates. Experimental results show that the relative errors of the model are 1.02%, 1.25%, and 1.49% in residential buildings, commercial complexes, and industrial plants, respectively, whereas the mean absolute errors are 0.92%, 1.13%, and 1.39%. The model demonstrates strong performance across different scenarios, with the relative error reduced to as low as 1.26%, which is significantly better than that of the comparison models. The proposed model effectively overcomes the performance limitations of traditional methods and achieves accurate and efficient prediction of construction cost rates in high-dimensional and complex scenarios. It provides a practical technical solution for full life cycle safety investment planning and refined cost management in building projects.
Serpentine liquid-cooled plates require a coordinated balance among heat-removal capability, temperature uniformity, and hydraulic energy demand. This study develops a three-dimensional transient computational fluid dynamics model for a bottom-cooled module comprising five prismatic LFP cells and systematically evaluates six geometric and operating variables under a 2C discharge condition: flow-path branch configuration, coolant flow direction, channel width, channel depth, inlet volumetric flow rate, and stamping corner radius. Coolant flow direction and channel width exert the strongest influence on the battery temperature field. Relative to flow direction II, flow direction I reduces the maximum and average temperatures by 2.16°C and 2.15°C, respectively, whereas flow direction II improves the maximum temperature difference by an average of 0.87°C. Increasing the channel width from 10 to 16 mm reduces the maximum and average temperatures by 2.07°C and 2.33°C and lowers the pressure drop by 47.3%. Increasing the inlet flow rate from 2 to 10 L/min yields only 0.38°C of maximum-temperature reduction, whereas the pressure drop rises from 10,921 to 110,722 Pa and the hydraulic power increases from approximately 0.36–18.45 W. Increasing the channel depth and flow-path branching primarily reduces hydraulic resistance with only a minor thermal penalty, whereas the stamping corner radius has negligible thermal influence within the investigated range. A channel width of approximately 14 mm represents a practical compromise for the present module when thermal performance, temperature uniformity, hydraulic loss, and stamping requirements are considered together. The results provide parameter-specific design guidance for bottom-cooled prismatic LFP modules within the investigated geometry and operating conditions.
Machinery and equipment play crucial roles in industrial productivity and processes, enabling the rapid and efficient completion of projects. This study introduces a hybrid predictive maintenance model for industrial induction motors, which overcomes the main limitations of current predictive models in the oil and gas industry, as well as other developed models, such as the reliance on original equipment manufacturer (OEM)–specific solutions and the inability to efficiently process noisy and nonlinear industrial data. Predictive maintenance systems are based on sensor data obtained from industrial equipment; unfortunately, these sensor data are often inconsistent due to environmental and operational variability, especially in complex industrial environments. To overcome these challenges, this study proposes an integrated feature engineering method that integrates principal component analysis (PCA) to reduce dimensionality and fuzzy logic to handle uncertainty and nonlinear relationships of sensor data. The framework uses various machine learning algorithms, such as random forest, support vector machines (SVMs), artificial neural networks (ANNs), and linear regression, to create predictive models for motor fault detection and classification. The sensor data (temperature, vibration, rotational speed [RPM], and electrical parameters) were preprocessed, transformed, and utilized for the training and validation of the model. The measures of standard performance, including accuracy, precision, recall, F1-score, and receiver operating characteristic–area under curve (ROC-AUC), are used to evaluate the model performance in a strong way. The model's remarkable 94.0% accuracy in the table indicates that it is highly accurate at predicting and categorizing the electric motor's health status; 94% of the time, the model was accurate, which is a very excellent indication of its consistency. The model identified 94% of the real positive instances with a recall of 0.9408. The model's AUC score of 0.9960 indicates a dependable overall performance and a great capacity to differentiate between positive and negative classifications. The findings indicate that the combination of PCA and fuzzy logic increases the relevance of the features and enhances the robustness of the model when dealing with uncertain and noisy data. In addition, the proposed framework is OEM-agnostic and can communicate with industrial communication protocols, including Modbus, allowing it to seamlessly integrate with existing SCADA and distributed control systems (DCS). This study has advanced predictive maintenance by providing a scalable, interoperable, and robust framework that can be used in real-life industrial applications.
The quest for optimal noise-reduction orifice plates in natural gas pressure regulating valves is often constrained by a fundamental dilemma: high-fidelity fluid-acoustic simulations, although accurate, impose prohibitive computational costs that render conventional optimization approaches impractical. To break this stalemate, this paper introduces a novel surrogate-assisted framework featuring a Query-by-Committee (QBC) mechanism that enhances the adaptive capabilities of kriging modeling. Unlike traditional single-criterion infill strategies, our framework constructs a diverse committee of kriging models using multiple kernel functions. By quantifying the level of “disagreement” among committee members regarding unobserved points, the algorithm intelligently selects the most informative samples, thereby maximizing global prediction accuracy while minimizing simulation calls. Extensive benchmarking against standard test functions demonstrates that the proposed QBC-kriging achieves a 6.13% higher comprehensive prediction accuracy and an 18.72% lower computational expenditure compared with the classical expected improvement (EI) strategy. In a multiobjective engineering case study focusing on a gas regulator's orifice plate, the new framework yields a 1.2% enhancement in optimization precision and a 39.6% boost in efficiency. Furthermore, a robustness analysis under varying inlet pressures reveals that the QBC-derived design maintains stable acoustic performance (4.8%–5.3% reduction, SD=0.21 dB), whereas the previously optimized design exhibits significantly wider fluctuations (4.1%–8.68%, SD=1.42 dB). This work offers a practical and scalable optimization pathway for complex fluid-acoustic systems burdened by high computational expenses.
The combined weir-gate systems are commonly used in water resources projects, but their hydraulic responses become significantly more complicated in the presence of downstream obstacles such as dikes, and the statistically based treatment of their hydraulics has received little attention. In the present study, the hydraulic and statistical characteristics of a rectangular combined weir-gate system with a single L-shaped super dike downstream are experimentally investigated under four different operating conditions. The experiments were conducted in a rectangular laboratory flume of 200×15×7.5 cm using 5 mm-thick beveled wooden structures, and 27 experiments were performed for each case by changing the pump discharge. The measured parameters were the actual discharge, downstream water depth, and water depth above the weir. The hydraulic investigation includes downstream velocity, specific energy, discharge coefficient, Froude number, and Reynolds number. The results were statistically treated using (a) descriptive statistics to provide a general overview of the seven variables, (b) the Pearson correlation coefficient to analyze the relationships between the different variables and to inspect the existence of multicollinearity, and (c) multiple linear regression (MLR) modeling of the discharge to predict its values using different variables, which was validated using R2, adjusted R2, F-statistics, p values, and residual scatter plots. Furthermore, the application of probability distribution functions (normal, Poisson, and binomial) to the measured data was investigated for each case. The MLR models performed very well, with R2 values ranging from 98.77% to 99.52%, and hence may be adopted for discharge prediction of the studied structure. Compared to the classical theoretical equations, the proposed MLR models produced significantly better results. The downstream velocity and the average downstream water depth were found to be the most important variables. The mean values of the discharge and discharge coefficient ranged from 0.000905 to 0.001102 m3/s and 0.733 to 0.886, respectively. It was found that the measured discharges followed the normal distribution for Cases 1–4, a Poisson-like pattern for Case 2, and a binomial-like pattern for Case 3. The results provide statistically based and empirically based relations for the discharge prediction of combined hydraulic structures.
The efficient energy management of hybrid renewable energy systems relies on a precise forecast of production and demand. In this study, the performance of MLP, LSTM, and BiLSTM models is evaluated for the prediction of photovoltaic and wind power generation, as well as energy demand. These models are evaluated using the RMSE, MAE, MAPE, and R 2 metrics. The results indicate that the BiLSTM model has the highest accuracy for load forecasting (RMSE = 0.115 kW, R 2 = 0.999), whereas the performance of all models is similar for forecasting photovoltaic and wind power generation (MAPE < 13 % ). These results confirm the efficiency of deep learning methods in improving the accuracy of forecasts and the energy management of hybrid systems. The results demonstrate the value of choosing the right neural network model to improve forecast precision and enhance the performance of the hybrid system. This study highlights the significance of advanced techniques in addressing energy management challenges and provides a groundwork for future research over extended timeframes and under more complex conditions.
The increasing dependence on global online networks has made cybersecurity a major concern for individuals and organizations. Understanding the dynamics of malware propagation is essential for formulating effective countermeasures. This paper proposes the SICRS (sensitive-infected-confined-recovery-sensitive) epidemic model to analyze malware propagation in heterogeneous and scale-free networks. Unlike traditional models, SICRS explicitly considers the impact of maintenance actions on infected and vulnerable nodes. To optimize defense strategies, we evaluate node importance using degree centrality and prioritize the protection of highly connected nodes. Simulation results show that this targeted control mechanism reduces the final epidemic size by more than 34% compared with models without such strategic intervention, while delaying the peak of infection sharply compared with other methods. Furthermore, we derive the epidemic threshold (R0), which shows that in scale-free networks, the threshold approaches zero, indicating high vulnerability. In the comparative analysis, the proposed SICRS model has more than 15% improvement in prediction accuracy compared with the traditional SIS, SIR, and SEIR models according to the corresponding graphs. These quantitative findings highlight the effectiveness of degree-based protection and provide practical insights for improving real-world cybersecurity protocols and reducing potential economic losses from large malware outbreaks.
This study investigates supply chain coordination for high-tenacity polyethylene terephthalate industrial yarns in omnichannel retailing environments, with a focus on the reserve online, pick-up in store (ROPS) strategy. A novel buyback and online profit-sharing (BOP) contract is developed. This contract addresses two main challenges: customer order cancellations and the gradual loss of yarn tenacity. The research compares decentralized, centralized, and coordinated supply chain structures, demonstrating that the proposed BOP contract significantly enhances overall supply chain performance. The Karush-Kuhn-Tucker (KKT) conditions are used to find optimal values for the decision variables. An optimization algorithm is also developed to solve the model efficiently. Our findings indicate that the coordinated approach increases total profit and optimizes pricing and deposit decisions, with the contract maintaining effectiveness across varying levels of customer loyalty to offline channels and rates of product quality deterioration. Higher customer loyalty to physical stores narrows the feasible range of profit-sharing parameters. Nevertheless, the BOP contract consistently outperforms decentralized operations. Furthermore, the study shows that while tenacity deterioration negatively impacts both retailers and suppliers, the coordinated structure provides better protection against these effects. Numerical results based on real data from a POY yarn producer demonstrate that the BOP contract can increase total supply chain profit by up to 42%, simultaneously reducing customer deposits by 93% and increasing fulfilled orders by 68% compared to a decentralized system. This research offers valuable managerial insights for implementing ROPS strategies in the retail of high-tenacity fibers, illustrating how effective contract design can align incentives across supply chain members while addressing operational challenges.
As a reliability analysis method for evaluating systems, failure mode and effects analysis (FMEA) is widely used to assess potential failure modes and their impacts on systems. However, FMEA faces limitations: the handling of expert evaluation information and their weights during the assessment and aggregation of risk evaluation information can significantly influence the results. Meanwhile, when ranking failure modes, over-reliance on experts' subjective assessments of risk factors (RFs) leads to lower credibility of the outcomes. To address these shortcomings, this paper proposes a novel FMEA method integrating fuzzy probabilistic linguistic term sets (FPLTS) with system-derived objective data. Unlike existing approaches that primarily focus on subjective expert weighting, this study introduces a quantitative mechanism to derive objective severity (S) values based on the physical topology (series vs. parallel connections) of the system components. First, FPLTS are employed to explicitly model the cognitive uncertainty and hesitation in expert probability assignments using triangular fuzzy sets. Second, a Mahalanobis distance (MD) weighting scheme is utilized to objectively measure expert consensus without assuming prior weight distributions. Crucially, the proposed method bridges the gap between subjective risk perception and physical system reality by using the structural load-sharing characteristics to correct subjective severity estimates. These elements are seamlessly integrated into a fuzzy analytic hierarchy process (FAHP) and fuzzy TOPSIS framework to robustly rank failure modes.
Aiming at the difficulties of topology reconstruction lag and multiobjective resource allocation conflict accumulation in the dynamic scenario of digital twin manufacturing system, the study constructs an automated deployment optimization model based on the synergy of topology optimization and multiscale modeling. Through the topology robustness extension and hierarchical reinforcement learning mechanism, the accuracy of real-virtual mapping and the efficiency of cross-level resource balancing are improved. The experimental results indicated that the model reached a hypervolume index of 0.97 in the simulation scenario, which was 18.3% higher than that of the traditional method. The value of solution set spacing was 0.05, and the distribution uniformity was optimized by 54.5%. In the dynamic topology reconfiguration test, the peak task conflict rate was 5.9%, the mean value of resource allocation was 80.7%, the standard deviation was 5.3%, and the degree of balance was 0.93. Actual production line verification revealed that the system throughput extreme value reached 1050 tasks/minute, and the average energy consumption per unit capacity was 89.7kWh. The deviation fluctuation range was +/- 11.4 kWh, which was 55.7% lower than that of the baseline scenario. The average resource utilization at device level was 86.1% (standard deviation 4.3%), and the cross-tier load balancing degree was improved to 0.95. The extreme value of deployment time under high concurrency scenarios was stabilized at 8.7-14.5 ms. It verified its real-time performance and robustness in dynamic heterogeneous environments. In summary, the proposed model can enhance the global adaptivity of virtual-real collaboration in smart factories and provide solutions for dynamic resource allocation and multiobjective optimization.
This study presents a systematic CFD investigation of forced convection in a long circular tube representative of heat exchanger channels, with particular emphasis on the coupled effects of mesh topology, turbulence modeling, and near-wall treatment. The primary objective is to identify the most suitable combination of mesh type, turbulence model, and near-wall treatment for accurately predicting heat transfer and pressure drop over a wide Reynolds number range (Re = 1573 - 23,592). This assessment is carried out through a comprehensive analysis involving five mesh types, 10 turbulence models, and four wall treatments. Validation against experimental data yielded heat transfer predictions within +/- 10%, with grid independence achieved at approximately 1.7 million cells. The O-grid mesh provided the highest accuracy but at the highest computational cost, while hexahedral and prism meshes offered optimal accuracy-efficiency trade-offs. Turbulence models were systematically evaluated using a MATLAB-Fluent framework. The linear pressure-strain RSM with EWT performed best for wall-dominated quantities, while k-omega GEKO with standard wall functions and k-& varepsilon; standard with ML treatment excelled at high and transitional Reynolds numbers, respectively. Velocity-profile analysis at Re = 23,592 showed weak sensitivity to mesh density and axial resolution, enabling accurate bulk-flow predictions on relatively coarse structured meshes.
Accurate registration of multipolarized petrographic thin-section images is essential for quantitative characterization of rock microstructures under plane-polarized and cross-polarized illumination. However, conventional diffusion probabilistic models struggle with the pronounced uncertainty arising from birefringence, textural heterogeneity, illumination variations, and blurred mineral boundaries. To address this, we propose an intuitionistic fuzzy set-guided diffusion probabilistic model for robust registration of multipolarized thin-section images. We derive an IFS-based fuzzy similarity field to drive a directional diffusion process and enable semantically consistent propagation of correspondence probabilities. In parallel, a membership-adaptive diffusion coefficient modulates the diffusion strength according to local correspondence confidence, enhancing propagation in reliable regions while suppressing it in uncertain areas, and an IFS-weighted loss emphasizes high-confidence pixels while penalizing high-hesitancy regions to stabilize optimization in structurally ambiguous zones. Experiments on a benchmark dataset of 2634 rock thin-section images spanning sedimentary, metamorphic, and igneous lithologies demonstrate that IFS-DPM preserves mineral grain boundaries, fracture networks, and pore structures under complex textures and multipolarized conditions, providing a robust registration framework for petrographic analysis.
This study evaluates the ability of Fire Dynamics Simulator (FDS) to represent fine-particle dynamics relevant to industrial and environmental applications. Simulations with 4 mu m silica particles were used to examine how particle number, grid resolution, travel distance, initial time step, and particle grouping affect predicted settling behavior in quiescent and flow-driven environments. Under quiescent air conditions, coarser meshes showed weak sensitivity to particle number, whereas finer meshes showed a systematic increase in apparent settling velocity with increasing particle loading. A central finding is that slip-based settling estimates are substantially less sensitive to particle loading than apparent cloud-settling velocities, indicating that deviations from the analytical Stokes reference increasingly reflect cloud-induced gas motion rather than particle slip alone. Changes associated with the initial time step remained minor within the tested range. Increasing travel distance reduced the influence of initial transients and improved agreement with the analytical baseline, while the imposed horizontal-flow case yielded a Stokes number of 4.5 & times; 10-7, indicating minimal inertial lag. Particle grouping preserved the global settling metric while substantially reducing computational cost on the finest tested grid. Overall, the results show that FDS can provide physically useful and computationally efficient predictions of fine dust transport when numerical settings are interpreted with respect to coupling effects and metric definition.
Structural electronics is an emerging field of technology in the electronics industry. As a result of technological development, electrical functionality can be hermetically embedded into various structural elements including smart glass laminates. Such glass laminates with LEDs can be utilized in the food industry for disinfection purposes, as ultraviolet light has an antimicrobial influence. Design and fabrication process optimization of structural glass elements is very demanding due to the lack of established design rules and limited experience in fabrication. To overcome this problem, a digital model was designed, and system-level multiphysics simulations were conducted. In this study, we focus on the electro-optical functionality of glass laminate, combining ray trace simulations with spectrum analyzer measurements to analyze light-induced disinfection characteristics of the glass element. Based on experimental verification, the simulation model gives reasonably accurate results to predict the functional performance of the glass element. The obtained results suggest that the concept of LED glass laminate is feasible and produces an irradiance profile suitable for disinfection.
Existing integrated guidance and control (IGC) schemes often face a trade-off between chattering suppression and rapid adaptability, particularly when relying on single-parameter tuning or discontinuous switching laws. To bridge this gap, this paper proposes a novel adaptive continuous higher order sliding mode controller (ACHOSMC) designed for autonomous aerial systems operating in highly contested, GPS-denied environments. The primary innovation lies in the integration of a fourth-order continuous sliding manifold with a dual-parameter neural network (NN) adaptation mechanism. Unlike conventional adaptive approaches that tune a single gain, the proposed architecture utilizes two lightweight NNs to simultaneously optimize both the integral and shaping gains in real-time. This dual-adaptation strategy enables the controller to aggressively reject compound disturbances—including wind gusts, sensor noise, and aerodynamic uncertainties—while maintaining a smooth, chattering-free control signal. Lyapunov analysis confirms global ultimate boundedness, and comparative simulations in a fully coupled 3D engagement scenario demonstrate the method's superiority. Quantitative results reveal that ACHOSMC reduces time-to-intercept by 89.7% relative to PID and 48.0% relative to nonadaptive CHOSMC, whereas increasing terminal intercept altitude by over 619%. These findings validate ACHOSMC as a robust, combat-ready solution that merges the precision of higher order dynamics with the flexibility of online neural learning.
This study presents a next-generation, resilient integrated guidance and control (IGC) framework for long-range, high-speed aerospace vehicles operating in the most contested and navigation-degraded environments, including complete GNSS denial, high-intensity sensor noise, and deliberate fault-injection conditions. The vehicle dynamics are modeled using an industrial-fidelity, nonlinear six-degree-of-freedom (6-DOF) formulation that captures time-varying mass and inertia due to fuel consumption and configuration changes, Mach- and flight phase-dependent aerodynamic coefficients, realistic actuator dynamics with saturation and rate limits, and inertial navigation channels affected by bias, drift, and electromagnetic interference. At the core of the design is a long short-term memory (LSTM) enhanced adaptive continuous higher-order sliding mode controller (ACHOSMC). The LSTM module enables real-time temporal learning to proactively compensate for navigation errors and inertial drift in the absence of GNSS, whereas the ACHOSMC law ensures robust disturbance rejection, fault accommodation, and stability preservation throughout all flight phases from initial boost to terminal approach. High-fidelity simulations, configured to reflect rigorous industrial qualification standards, demonstrate that the proposed approach achieves a landmark improvement in precision, delivering a reduction in circular error probable (CEP) ranging from 27% to over 75% compared with established PID, SMC, and H infinity controllers across all tested scenarios. These improvements confirm the framework ' s robustness, adaptability, and suitability for advanced aerospace guidance applications in GNSS-denied and dynamically uncertain operational domains.
Helical compression springs are widely utilized in numerous applications, such as automotive suspension systems, due to their remarkable features; the designers focus on three parameters: the wire diameter, mean coil diameter, and number of active coils due to their impact on spring performance. This research proposes a novel approach offering unused optimization algorithms. In contrast, previous works focusing on solely one or multiobjectives, such as minimizing weight and enhancing the fatigue life using traditional algorithms, besides ANSYS, inventor, and experimental validation, the present work highlights a multiobjectives approach for unexplored targets utilizing SolidWorks simulation for validation. Therefore, it is aimed at minimizing the spring weight of a Sedan vehicle, besides enhancing fatigue life and coil clearance using MATLAB algorithms for multiobjectives optimization including grey wolf (GWO), red fox (RFO), modified grey wolf, and hybrid-modified grey wolf-red fox algorithms along with their validation to confirm the stress and deformation properties besides the required objectives under nonlinear loading constraints. The results indicate that the reduction in weight increased from 37% to 47%, besides achieving 1.556 as a safety factor for hybrid MGWO with 1.735 in the FEM study; furthermore, coil clearance improved by 103%-236% for RFO and hybrid MGWO, respectively. The optimized parameters improved and were within the design range. Other results, such as buckling risk and spring index, were validated, and they found that stable and within the acceptable range, respectively. The proposed design methodology is successful in terms of integration between simulation and optimization study.
3D point cloud models are being utilized more and more in computer vision, virtual reality, intelligent manufacturing, and cultural heritage preservation as a result of the quick advancements in computer graphics and 3D modeling technologies. Therefore, protecting the copyright and data integrity of 3D models has become an important issue. The study enhances the robustness of the model to rotation, translation, and scaling attacks by affine invariant processing. Moreover, the digital watermark is embedded into the processed 2D image by combining techniques such as principal component analysis, coordinate projection, redundant discrete wavelet transform, and singular value decomposition. The experimental results show that the peak signal-to-noise ratios of the four 3D models after embedding the watermark are all higher than 47 dB, among which the Happy_recon model reaches 56.83 dB. The incremental 3D point cloud information hiding algorithm achieves an embedding capacity of 1643 bits in the Happy_recon model, which is 57.22% higher than that of the label clustering method. In the watermark robustness test, the bit error rate of this method against geometric attacks is zero, and the normalized cross-relation number is 1. Under a 40% sheet-cutting attack, the bit accuracy rate still remained above 0.85. The peak signal-to-noise ratio after embedding the watermark is on average approximately 18.2% higher than that of the method based on geometric features and approximately 9.5% higher than that of the method based on the transform domain. The information hiding algorithm based on incremental point cloud improves the embedding capacity by 57.22% compared with the label clustering method and by 29.98% compared with the distance feature method. In terms of robustness, the resistance to geometric attacks is completely immune, and the bit accuracy rate against noise attacks is approximately 5.9% higher than that of deep learning methods. In conclusion, the proposed method effectively enhances the performance of 3D point cloud models in watermarking and information hiding tasks.
This examination studies the two-dimensional flow and heat transmission of a Jeffrey fluid over a nonlinear stretching sheet saturated in a permeable medium, accounting for the effects of viscosity variation and magnetic field. The prevailing nonlinear boundary layer equations are turned into an arrangement of ordinary differential equations by the practice of similarity adaptations and solved numerically via the bvp4c solver. The numerical process is indorsed against existing outcomes from the literature, establishing wonderful agreement and approving the precision of the present methodology. The effects of important factors, including the nonlinearity factor of stretching sheet eta, the Prandtl number Pr, the porosity parameter epsilon, the Jeffrey parameter delta, the magnetic field parameter M, and the viscosity variation parameter beta, on the velocity, temperature, skin friction coefficient, and rate of heat transfer are explored. The results indicate that increasing the nonlinearity factor of stretching sheet eta enriches both skin friction and heat transfer rate, whereas greater porosity parameter epsilon, the viscosity variation parameter beta, the Jeffrey parameter delta, and the magnetic field parameter M lead to their reduction. Moreover, an escalation in the Prandtl number Pr increases the heat transfer rate while it decreases the skin friction. This study is appropriate to processes such as polymer processing, cooling of electronic devices, and magnetohydrodynamic (MHD) flow control in industrial thermal structures, where non-Newtonian fluids and temperature-dependent viscosity play a major role.
Accurate deformation forecasting is essential for concrete dam safety monitoring, yet real-world deformation series are often nonlinear, nonstationary, and contaminated by noise, missing values, and outliers, which limits the performance of traditional prediction models. This paper proposes a high-precision forecasting framework that integrates variational mode decomposition (VMD), adaptive wavelet-threshold denoising, and the informer model. First, VMD decomposes complex deformation signals into intrinsic mode functions to mitigate mode mixing and enhance frequency-specific interpretability. Then, an adaptive wavelet-threshold strategy is introduced, where the threshold is dynamically adjusted using signal-to-noise ratio (SNR) guidance and Stein's unbiased risk estimate (SURE), enabling effective noise suppression while preserving informative deformation patterns under nonstationary conditions. Finally, a multichannel informer architecture is employed to fuse multiscale components and capture long-range dependencies efficiently via ProbSparse self-attention. Experiments on deformation monitoring datasets from multiple concrete dams demonstrate that the proposed VMD-AWT-informer consistently outperforms mainstream baselines across forecasting horizons. For instance, at a 30-day horizon, the proposed method achieves an RMSE of 0.55 mm, reducing errors by 23.6% compared with VMD-LSTM and by 55.6% compared with SVR, while maintaining strong goodness of fit (R2=0.887). Robustness tests further confirm improved stability under noisy, incomplete, and outlier-corrupted inputs. These results indicate that the proposed framework provides an effective and practical tool for long-horizon dam deformation prediction and early warning.
