Generation and evaluation mechanism of digital media art pattern design scheme based on interactive genetic algorithm

High-performance chiral metasurface sensors optimized by a target-driven active learning framework

The distribution characteristics of the deviatoric stress tensor (DST) field in the surrounding rock of the end-mining retreat roadway (ERR) are intricate, exerting a substantial influence on the stability of the ERR's surrounding rock. Taking the fully mechanized working face and its retreat roadway in Wutong Coal Mine as the engineering backdrop, initially, a numerical model of the DST in the end-mining coal pillar (ECP) was established to analyze its evolutionary pattern. Based on the response surface methodology (RSM) model, the significance of each influencing factor was examined. Subsequently, the three dimensional stress expression above the ECP was deduced, and the invariants and distortion energy (DE) of the DST at any position within the ECP were ascertained. Then, the expressions for the cutting height and cutting angle were derived, and a novel time step control technology (TSCT) for the ERR's surrounding rock was put forward, based on the real-time evolution characteristics of DST, dynamic matching of support strategies and roof cutting parameters is achieved through two stages of passive reinforcement and active roof cutting to achieve collaborative control. Compared with traditional static design or single time step methods, this technology reduces the deformation rate of surrounding rock by about 84% through step-by-step and timely application of support and cutting. Ultimately, according to the field measured data, it was demonstrated that this technology can effectively mitigate the deformation of the ERR's surrounding rock.

Non-uniform rods are widely present in ultrasonic vibration systems, and the accuracy of their resonance design significantly impacts system performance. Traditional design methods often treat non-uniform rods as one-dimensional members, neglecting the influence of lateral inertia, which results in lower resonance design accuracy. This paper establishes a vibration model for non-uniform rods under ultrasonic excitation that accounts for lateral inertia effects. Subsequently, the natural frequencies and mode shapes of the rod are obtained using the transfer matrix method. The modal superposition method is then employed to derive the internal displacement distribution function and stress distribution function of the rod. The validation of the tapered rod and the ultrasonic fatigue specimen demonstrates that the proposed method is closer to the Finite Element Method (FEM) results than the traditional method. The results show that the relative deviation between the resonant length of the tapered rod and the FEM calculation is less than 0.16%, and the harmonic response analysis results of the two are also in good agreement; the relative deviation between the first-order natural frequency of the ultrasonic fatigue specimen and the FEM calculation is less than 0.315%, and the predicted maximum stress is highly consistent with the FEM. The research findings presented herein can serve as a universal methodology for resonance design and performance evaluation of longitudinal vibration components in ultrasonic vibration systems.

Driven by the demands of light weighting, multifunctionality has become increasingly important in the design of lattice metamaterials. While inverse design is crucial for developing such lattice structures, traditional inverse design methods such as topology optimization often fail to fully explore the design space. To overcome these limitations, this study introduces a generative AI framework that combines 3D Gaussian voxel generation with deep learning, enabling greater structural complexity and design freedom. As a proof of concept, we employ this bottom-up design approach to shell lattice structures for optimized energy absorption and broadband sound absorption capabilities. A hybrid architecture combining a 3D convolutional neural network and a conditional deep convolutional generative adversarial network enables accurate energy absorption prediction and performance-driven structural generation. In parallel, a genetic algorithm is employed to tune heterogeneous geometries for effective broadband sound absorption. Experimental validation through 3D-printed stainless-steel lattices demonstrates the superior multifunctionality of the designed structures -achieving 40%-200% greater energy absorption than conventional shell lattices, along with a high (average coefficient ∼0.7) and broad (α > 0.5 across 1000-5800 Hz) absorption bandwidth. Overall, our proposed framework overcomes the major drawbacks of existing inverse design approaches, offering enhanced voxel-level model generation informed by physical insights.

This research focuses on the need to clearly define the lateral design force distribution shape for the seismic design of reinforced concrete (RC) buildings. It specifically looks at the effects of soil-structure interaction (SSI) and inelastic structural behavior. Traditional seismic design methods usually assume fixed-base conditions and overlook SSI. This can result in inaccurate predictions of how much the structure will deform. To create a more reliable, data-driven seismic design framework, the study conducted extensive nonlinear time-history analyses on various regular RC buildings using OpenSeesPy both for generating a building database based on current code provisions and for assessing their dynamic response. These models vary across important factors, including fundamental periods, slenderness ratios, , and soil conditions. The spectral acceleration at the first mode period served as the measure of ground motion intensity (IM). The collected data were used to derive classical regression equations and to train machine learning models, including Neural Network Regression and Gradient-Boosting Regression Trees, to predict the optimal lateral force profile shape. These methods are essential for understanding the complex, nonlinear relationships between seismic input factors and the dynamic characteristics of a building, as well as engineering demand parameters (EDPs) such as storey drift. The study found that the key factors affecting nonlinear lateral displacement and force-shape profile include the structure's slenderness ratio , the fixed-base and flexible-base fundamental periods, and , and the IM, . By providing a probabilistic assessment, this methodology seeks to improve the outcomes of seismic design codes and enhance performance-based design for RC buildings.

Under the background of global sustainable development, this study focuses on the integrated application of green design concept in indoor environment art decoration and systematically discusses the core application principles such as economic rationality, health and environmental protection, and overall aesthetics. In order to solve the limitations of traditional design methods, an improved cultural algorithm based on pattern learning is proposed, which improves the innovation of environmental art design and enhances the convergence efficiency of the algorithm. In addition, a C- Support Vector Machine (C-SVM) model is constructed to evaluate the visual comfort of indoor lighting with color temperature and brightness as key variables. The model achieves high prediction accuracy and strong generalization ability. By integrating the concept of green design, interdisciplinary technical tools, and intelligent evaluation methods, this study provides a scientific basis for building an energy-saving, environmentally-friendly, and people-oriented comfortable indoor space.

The design of terahertz (THz) metamaterial absorbers has evolved from simple structures to complex composites to meet the demands for multi-frequency absorption, broadband absorption, and polarization independence. However, the nonlinear interactions and expanded design spaces of composite structures pose significant challenges, making traditional design methods time-consuming and labor-intensive. To address these issues, this study proposes a data-driven framework that integrates a hybrid variational autoencoder-transformer-long short-term memory (VTL) model. The architecture is specifically designed to capture structural-spectral relationships, where the transformer models global dependencies among structural parameters, the long short-term memory (LSTM) network enhances the modeling of sequential spectral features in the terahertz frequency range, and the variational autoencoder (VAE) improves feature representation by learning implicit latent distributions. This integrated design enables effective characterization of the complex electromagnetic responses of composite metamaterials. The proposed approach achieves high prediction accuracy, with a mean squared error (MSE) of 0.0009, a coefficient of determination (R2) of 0.9725, and a mean absolute error (MAE) of 0.0161. It predicts a perfect absorption rate of 99.9% and optimally adjusts structural parameters to achieve targeted frequency responses. By addressing the limitations of traditional methods, this framework not only shortens the design cycle and reduces experimental costs but also offers a robust solution for the efficient design of high-performance THz metamaterial absorbers.

Nuclear magnetic resonance (NMR) co-magnetometers enable high-precision magnetic field measurement and inertial measurement with great application potential, whose optical components need to adjust beam quality and polarization to ensure detection performance. However, existing devices are bulky with low integration, and traditional design methods also restrict their development. This study uses a metasurface to integrate polarization and collimation functions, and proposes an inverse design method combining transfer learning and optimization algorithms. Experiments show that the metasurface achieves 82.63% transmittance, with collimation and polarization performance deviating from theoretical values by less than 3%, effectively reducing system volume while guaranteeing co-magnetometer performance, thereby providing a feasible method for micro-nano atomic magnetometers.

ABSTRACT Metamaterials, with their unique electromagnetic properties, offer a promising solution for enhancing the signal‐to‐noise ratio (SNR) in Magnetic Resonance Imaging (MRI). However, their customized design tailored to patient‐specific anatomical features remains a significant challenge. Strict geometric dimensional constraints lead to extreme sparsity within the parameter space, causing traditional inverse design methods to fail in generating valid structures that satisfy physical constraints due to interpolation errors. To address the bottleneck of high simulation costs in customized design, this paper proposes a generative data augmentation strategy. By leveraging an improved CWGAN‐GP to accurately fit high‐Q resonances and bridge data gaps, combined with an inverse network, we successfully achieve the precise design of metamaterials satisfying strict geometric constraints at zero post‐training simulation cost, establishing a novel, general, and efficient paradigm for personalized design. The results indicate that this “Generate‐Augment‐Inverse” closed‐loop strategy not only resolves the issue of interpolation failure under sparse sampling but also provides a generic data‐driven pathway for the efficient and low‐cost customization of personalized medical devices.

The Fluid Antenna System (FAS), which enables flexible Multiple-Input Multiple-Output (MIMO) communications, introduces new spatial degrees of freedom for next-generation wireless networks. Unlike traditional MIMO, FAS involves joint port selection and precoder design, a combinatorial NP-hard optimization problem. Moreover, fully leveraging FAS requires acquiring Channel State Information (CSI) across its ports, a challenge exacerbated by the system’s near-continuous reconfigurability. These factors make traditional system design methods impractical for FAS due to nonconvexity and prohibitive computational complexity. While deep learning (DL)-based approaches have been proposed for MIMO optimization, their limited generalization and fitting capabilities render them suboptimal for FAS. In contrast, Large Language Models (LLMs) extend DL’s capabilities by offering general-purpose adaptability, reasoning, and few-shot learning, thereby overcoming the limitations of task-specific, data-intensive models. This article presents a vision for LLM-driven FAS design, proposing a novel flexible communication framework. To demonstrate the potential, we examine LLM-enhanced FAS in multiuser scenarios, showcasing how LLMs can revolutionize FAS optimization.

The design of an aircraft engine nacelle directly affects aircraft drag, fuel consumption, and noise levels, and is a key step in aerodynamic shape optimization. Traditional optimization design methods based on computational fluid dynamics (CFD) rely heavily on high-precision simulation data. However, high-precision CFD calculations are expensive and time-consuming, severely limiting the exploration of the design space and the efficiency of optimization. With the development of artificial intelligence (AI), optimization methods based on surrogate models have become a research hotspot. However, existing methods often rely on data of a single precision and are unable to effectively utilize abundant and easily accessible low-precision data resources. This study aims to introduce a multi-fidelity deep learning (MFDNN) method for the aerodynamic optimization of aircraft engine nacelles. By fusing CFD data of different precisions, a high-precision and efficient aerodynamic performance prediction model is constructed. This method significantly reduces the number of expensive CFD calculations while maintaining optimization accuracy, providing a new technical approach for the rapid and efficient design of aircraft engines.

One of the key problems in the construction of wireless charging systems is to increase the efficiency of wireless energy transfer, especially with limited dimensions of devices and changing mutual position of coils. Traditional design methods are often insufficiently accurate for multilayer coils, which requires the development of new approaches. The study is based on a mathematical model of a wireless energy transfer system with a series LC topology, built on the basis of equivalent circuits, while the parameters of the transmitting and receiving coils are determined using the vector potential method. An approach is proposed that allows reducing the number of variable parameters that determine the efficiency of wireless energy transfer. A technique for optimizing the considered wireless charging systems at the design stage is developed, according to the criteria of wireless energy transfer efficiency, the value of the transmitted power and over-voltage on the capacitors of the resonant circuit. The technique is based on multi-criteria analysis using relative im-portance coefficients. For specific conditions, the optimal parameters of the transmitting and receiving coils are determined, namely the number of turns and layers of the winding under given dimensional constraints, ensuring maximum efficiency. Experimental testing on a model confirmed the correspondence of the calculated and real characteristics with sufficient reliability. It is shown that the proposed approach allows one to effectively take into account geometric constraints and minimize losses in the resonant circuit. The proposed optimization method demonstrates high efficiency in the design of wireless charging systems with multilayer coils and series LC topology under conditions of possible coil displacements.

Prestressed concrete (PC) girder bridges are essential in modern bridge engineering. Traditional design methods are time-consuming and labour-intensive, often failing to produce economically or structurally superior schemes. Optimisation techniques automate the design process, enabling optimal designs. However, challenges remain, including limited optimisation research on long-span girder bridges, high computational intensity of finite element analyses (FEAs) during optimisation, a tendency to get trapped in local optima, insufficient attention to structural performance, and a lack of interpretability for optimal designs. In this paper, a hybrid-heuristic machine learning-based framework is proposed, applied innovatively to the optimisation of long-span PC girder bridges. The framework incorporates optimisation objectives focused on cost and stress indicators. Within this framework, a support vector regression (SVR) model replaces traditional FEA, enhancing optimisation efficiency. Additionally, a hybrid-heuristic algorithm, combining simulated annealing and particle swarm optimisation (SA-PSO), is incorporated for the global optimum. The shapley additive explanations (SHAP) approach is integrated to clarify relationships between the objective and variables, providing interpretability for the optimal design. This framework is validated on a PC rigid-frame bridge. Results indicate it rapidly achieves optimisation, balancing structural performance and economic efficiency with strong interpretability. Furthermore, the framework demonstrates potential for future research and engineering practice beyond bridges.

Abstract Metamaterials are engineered materials composed of specially designed unit cells that exhibit extraordinary properties beyond those of natural materials. Complex engineering tasks often require heterogeneous unit cells to accommodate spatially varying property requirements. However, designing heterogeneous metamaterials poses significant challenges due to the enormous design space and strict compatibility requirements between neighboring cells. Traditional concurrent multiscale design methods require solving an expensive optimization problem for each unit cell and often suffer from discontinuities at cell boundaries. On the other hand, data-driven approaches that assemble structures from a fixed library of microstructures are limited by the dataset and require additional post-processing to ensure seamless connections. In this work, we propose a neural network-based metamaterial design framework that learns a continuous two-scale representation of the structure, thereby jointly addressing these challenges. Central to our framework is a multiscale neural representation in which the neural network takes both global (macroscale) and local (microscale) coordinates as inputs, outputting an implicit field that represents multiscale structures with compatible unit cell geometries across the domain, without the need for a predefined dataset. We use a compatibility loss term during training to enforce connectivity between adjacent unit cells. Once trained, the network can produce metamaterial designs at arbitrarily high resolution, hence enabling infinite upsampling for fabrication or simulation. We demonstrate the effectiveness of the proposed approach on mechanical metamaterial design, negative Poisson's ratio, and mechanical cloaking problems with potential applications in robotics, bioengineering, and aerospace.

Traditional visual design methods are no longer able to meet the diverse needs of users in the digital age in terms of information processing speed, visual recognition accuracy, and interactive experience. This article first summarizes the basic theories and mainstream models of computer vision in the field of visual communication, deeply analyzes the limitations of existing methods, and proposes an innovative computer vision fusion method for visual communication based on this. Through multidimensional analysis of model architecture, design process, and practical application cases, the significant advantages of this method in improving the level of visual communication intelligence and user experience have been demonstrated. Finally, this article also validates the effectiveness of relevant innovative practices, providing useful references for subsequent design innovation and theoretical development.

Prestressing is critical in prestressed concrete (PC) bridges, where improper design not only wastes material but may ultimately compromise serviceability or even cause structural collapse. Traditional design methods, often cumbersome and code-oriented, typically yield suboptimal solutions. While existing optimization studies mainly target cost minimization, they exhibit several limitations. These include insufficient attention to structural stress performance, resulting in suboptimal mechanical behavior, as well as a lack of research on long-span girder bridges, which involve inherent complexities in both structural behavior and construction processes. Moreover, applications of multi-objective evolutionary algorithms (MOEAs) remain scarce in bridge optimization. To address these gaps, this paper innovatively proposes a multi-objective prestressing optimization method for long-span PC girder bridges, simultaneously considering cost and stress performance. Focusing on long-span continuous girder bridges, the method aims to minimize tendon usage while achieving uniform distribution of stress safety degree. Constraints incorporate code-specified stress and configuration requirements. The Non-dominated Sorting Genetic Algorithm II (NSGA-II) is employed to handle multi-objective, discontinuous, and nonlinear optimization challenges. The proposed method is applied to an actual bridge, demonstrating its feasibility and effectiveness. Additionally, comparison with a single-objective optimization confirms that the proposed method provides a more balanced and effective scheme.

Ice shell structures have gained widespread attention in cold regions due to their unique aesthetic forms and efficient structural configurations. However, the poor mechanical properties of the materials and the adoption of traditional design and construction methods have constrained the appearance and construction techniques of ice shell structures. This necessitates the development of an integrated approach that combines design and construction. This study proposes a combined approach based on graphical statics theory and the finite element method (FEM) to generate free-form geometries for ice shell structures and conducts a comprehensive structural analysis of their mechanical performance. The construction strategy primarily involves the prefabrication of ice shell components and on-site assembly, the workflow that has been validated through a case study of the Harbin Ice and Snow World. Subsequently, a 16-day structural monitoring program was conducted to investigate the factors influencing the performance of ice shell structures. The monitoring results indicate that the primary factors affecting the structural behavior of ice shells were identified as air temperature and wind speed, with solar radiation as secondary influence. The results demonstrate that the proposed workflow enables the free-form design of ice shells, while the prefabrication and assembly methods are suitable for ice shell construction. Additionally, the monitoring method and analytical results can provide support for establishing safety assessment protocols for ice shell structures.

Proteolysis-targeting chimeras (PROTACs) represent a transformative strategy in drug discovery, enabling the selective degradation of target proteins rather than merely inhibiting their activity. However, their structural complexity and deviation from conventional drug-like properties present major challenges for traditional design and optimization methods. In this review, we provide a comprehensive overview of recent computational advances that facilitate PROTAC development, encompassing chemoinformatics, structural bioinformatics, molecular modeling and machine learning resources. We highlight computational tools for warhead and linker design, ternary complex modeling and the prediction of degradation efficiency and ADMET profiles. Finally, we discuss current limitations and future perspectives, emphasizing strategies to enhance design effectiveness and accelerate clinical translation.

This study aims to provide a deeper and more realistic understanding by conducting a systematic comparison between the two approaches (Reynolds number and volumetric flowrate). The analysis emphasizes the impact of internal channel design, using inclined winglets and surface corrugation. An experimental investigation was carried out to prepare and characterize a TiO₂/H2O nanofluid at 1% volume concentration, including accurate measurements of its thermophysical properties and stability validation. A numerical model was also developed using ANSYS Fluent to simulate the hydrothermal behavior of two channel configurations (straight and corrugated), in which the effects of both Reynolds number and flowrate were evaluated across key parameters such as heat transfer coefficient, pressure drop, performance evaluation criterion, and wall temperature distribution. By observing the flow patterns inside the corrugated channel, three distinct flow behaviors were identified: axial flow along the channel, transverse flow induced by winglets, and swirling flow within the corrugated grooves. This combination of flow modes enhanced fluid mixing and significantly improved heat transfer performance. The results show that TiO₂ nanofluid significantly enhances the thermal–hydraulic performance, with the relative friction factor (Γ) increasing from 6.9 to 7.6 and the thermal enhancement ratio (En) reaching 2.8 (PEC ≈ 1.5) when evaluated using Reynolds number, while volumetric flow rate assessment (7–9 L/min) yielded higher Γ (3.9–4.2) and En/PEC (2.5/1.6). The effects of the internal enhancement techniques were found to be more pronounced when using flowrate as the reference indicator. This work represents a valuable scientific contribution by integrating three advanced enhancement strategies (surface corrugation, inclined winglets, and nanofluid), and it highlights the need to reconsider traditional thermal system design methods based solely on Reynolds number.