Titanium alloy serves as a critical structural material for aircraft and engine components. During the manufacturing of these titanium parts, machining, particularly turning, is a fundamental process. However, continuous turning faces a significant bottleneck: severe tool wear caused by insufficient lubricant infiltration at the tool-workpiece interface and excessive cutting forces. The nanobiolubricant minimum quantity lubrication (NMQL) turning process of biomimetic textured cutting tools empowered by ultrasound is considered to have the potential to solve the problem of tool wear during titanium alloy cutting. Nevertheless, the lubricant infiltration dynamics mechanism and tribological properties under the new process are unclear. Based on this, the synergistic effect of ultrasonic vibration on lubricant infiltration and migration was first analyzed. Subsequently, research has been conducted on the frictional properties and surface damage characteristics of four working conditions: dry cutting, NMQL, textured tool assisted NMQL (T-NMQL), and ultrasonic vibration empowered T-NMQL (UVT-NMQL). Surface roughness, surface morphology, cutting specific energy, chip morphology, and tool wear analysis have also been carried out. Furthermore, wavelet analysis has been introduced to decompose surface roughness signals into high and low frequencies, enriching the quantitative evaluation system for surface damage of titanium alloy cutting workpieces. The average cutting specific energies under dry cutting, NMQL, T-NMQL, and UVT-NMQL conditions were determined to be 2.32, 2.18, 2.01, and 0.78 J/mm3, respectively. Based on the wavelet decomposition results of surface roughness signals, it was found that the surface damage energy of NMQL, T-NMQL, and UVT-NMQL conditions decreased by 28.84

With the widespread deployment and sharing of multi-source heterogeneous data in new power systems, power markets face cyber-attack risks while striving to improve resource allocation efficiency. Addressing the trade-off between privacy preservation and data utility, this paper proposes a dynamic protection and privacy-preserving computation method for power market data. The method constructs a multi-factor dynamic sensitivity classification model to achieve differentiated desensitization protection across business scenarios. Furthermore, a distributed privacy-preserving computation framework based on Haar wavelet transform and Paillier homomorphic encryption is proposed, complemented by a deep learning hardening mechanism based on loss function perturbation. Simulation results demonstrate that the proposed scheme ensures user privacy while maintaining a load sequence reconstruction fitness of over 97%, effectively balancing the security of power market data with operational efficiency.

High-temperature-resistant and chemically stable ceramic materials exhibit great adaptability across numerous industrial applications. Grinding is an essential component of the precision shaping and manufacturing processes for ceramic structural components. However, the low machining efficiency and high machining damage rate caused by hard and brittle material properties have been a challenge in both academia and industry. Grinding force is the most critical parameter reflecting the grinding system, and establishing an accurate prediction model is highly significant in reducing machining damage. However, a knowledge gap remains in the comprehensive review and evaluation of grinding force models for ceramic materials, which is undoubtedly not conducive to further theoretical advances. This review discusses the removal mechanism for polycrystalline ceramic materials. Subsequently, it comprehensively reviews and comparatively evaluates detailed grinding force modeling knowledge. Furthermore, it explores the specificities of the ultrasonic and laser energy-field-assisted grinding of ceramic materials in terms of their physical behavior and mechanical modeling. Finally, the theoretical value of grinding force modeling for predicting the damage to ceramic materials is explored. The current limitations of the grinding process, mechanical modeling of ceramic materials, corresponding potential research directions, and valuable research content are provided. The goal is to derive actionable low-damage grinding guidelines and establish a robust theoretical framework that enhances the quality of grinding processes for ceramics and other hard and brittle solids.

As large language models (LLMs) become increasingly embedded in diverse applications, from natural language processing to cybersecurity, the demand for robust privacy-preserving solutions has surged. This paper discusses privacy vulnerabilities in LLMs, identifying risks at system, application, and network levels. We categorize and evaluate current privacy-preserving methods, based on homomorphic encryption (HE), trusted execution environments (TEE), and secure multi-party computation (MPC), to assess their effectiveness in mitigating data exposure while supporting LLM performance. Building on these insights, we propose a novel privacy-preserving framework for fine-tuning and inference in LLMs. Our findings highlight existing gaps and propose future directions for developing secure, efficient, and scalable privacy-preserving LLM architectures.

CFRP precision grinding is a critical process in aeronautical industry. To enhance surface integrity, ultrasonic-assisted MQL grinding with nanolubricants was developed. The effects of ultrasonic on permeation mechanism and fluid dynamic pressure lubrication were investigated. The mechanical behavior of droplet anisotropic spreading, contact line dynamics, and capillary action was analyzed. Additionally, tribological properties of CFRP grinding surfaces with various fiber orientation angles were evaluated. The findings reveal contact angles in vertical grinding direction are higher than those in parallel direction. The application of 2D ultrasonic vibration decreased stable aspect ratio and increased spreading speed of droplets. Nanolubricants demonstrated significant improvements in reducing friction coefficient, grinding force ratio, wear area and volume, with a maximum reduction in friction coefficient of 34.87%.

Cloud computing has emerged as a critical solution for handling large-scale computations, offering powerful resources and scalability. However, outsourcing computations to the cloud introduces significant privacy concerns, as sensitive data might be exposed during processing. To address these challenges, secure outsourcing protocols have been developed, ensuring that data privacy is maintained during remote computations. Despite their potential, existing secure outsourcing protocols have limitations in the variety of operations they support, restricting their applicability in real-world scenarios. In this paper, we propose SOSQRT, an efficient and secure outsourcing protocol for square root computation. We employ an innovative area approximation technique to compute square roots accurately while minimizing computational overhead. To protect user privacy throughout the process, we integrate three secure protocols-SCMP, SMUL, and SDIV-to ensure data confidentiality during remote execution. Additionally, our design optimizes the protocol's performance, achieving reduced time and communication costs compared to existing solutions. Experimental results demonstrate the effectiveness of SOSQRT, showing significant improvements in computational efficiency and privacy protection.

The proliferation of sophisticated neural architectures, particularly Vision Transformers (ViTs), has been paralleled by the evolution of advanced adversarial exploits, among which backdoor attacks remain a critical threat. These attacks instill hidden malicious behaviors, activated by specific triggers, into an otherwise high-performing model. While defenses have been proposed for convolutional networks by analyzing neuron-level statistics, these approaches do not directly translate to the component-specific architecture of Transformers. Concurrently, work in model interpretability has revealed that distinct functionalities, such as safety mechanisms, can be localized to individual attention heads. This paper bridges these two domains. We hypothesize that in a backdoored ViT, the malicious functionality is not diffuse but is sparsely encoded within a small subset of attention heads. We propose that these backdoor heads exhibit unique statistical signatures in their activation distributions. Building on this, we introduce two novel defense strategies: Entropy-Based Head Pruning (EHP), which identifies heads with low distributional entropy using a poisoned dataset, and Statistical Divergence Head Pruning (SDHP), a more practical method that identifies anomalous heads by comparing clean-sample statistics to the model's embedded Batch Normalization statistics. Our methods excise these malicious heads, effectively neutralizing the attack. Experiments on CIFAR-10 and Tiny-ImageNet demonstrate that our approach can reduce attack success rates to near-zero while preserving over 99% of the model's original clean accuracy, significantly outperforming existing neuron-based pruning defenses.

With the increasing integration of artificial intelligence systems across various domains, safeguarding model robustness against adversarial attacks has become a critical research imperative. The intrinsic lack of interpretability in machine learning models renders them particularly vulnerable to adversarial perturbations-subtle input alterations that can significantly degrade model performance. Although existing defense strategies, such as adversarial training with FGSM and PGD, enhance model resilience, they are constrained by computational inefficiency and incomplete coverage of the input space. In this paper, we propose DASA (Discriminator Adversarial Sample Attack), a novel defense framework that integrates adversarial sample generation with enhanced attack detection mechanisms. The method employs a generator-discriminator architecture that simultaneously produces realistic adversarial samples and improves attack detection through adversarial training. Through iterative optimization, the generator dynamically refines adversarial attack patterns based on discriminator feedback, progressively enhancing model robustness. Our framework incorporates accuracy monitoring and an early stopping mechanism to optimize the trade-off between defensive efficacy and computational efficiency, while integrated visualization tools facilitate an intuitive assessment of defense performance. DASA presents a unified framework for (1)evaluating and enhancing model security, providing an efficient and scalable solution to advance the field of adversarial defense.

In knowledge graph construction, traditional NLP models suffer from poor cross-domain adaptability, while large language models (LLMs) face issues of hallucinations and data shift in information extraction. To address these challenges, we propose a collaborative framework for LLM-based triple extraction that integrates dynamic prompt generation and BERT assistance, effectively combining the entity recognition strengths of traditional models with the end-to-end reasoning capabilities of LLMs. On one hand, the framework uses the K-nearest neighbor (KNN) algorithm to vectorize text entity types. It matches highly relevant examples and generates dynamic prompts with progressive reasoning chain-of-thought (CoT) templates, enhancing LLMs' ability to adapt to task features and perform logical reasoning. On the other hand, it fine-tunes pre-trained BERT models to achieve accurate recognition and type annotation of text entities, alleviating LLM hallucinations by enhancing the structured representation of input text. Experimental validation on 3 cross-domain and cross-lingual datasets shows that when few-shot learning is combined with the CoT strategy, the triple extraction F1 score of Qwen2.5-14B increases from 0.1718 to 0.5856 (a 240.9 % relative improvement), demonstrating a performance breakthrough from strategy integration. After introducing BERT assistance, the F1 score of DeepSeek-V3 reaches 0.7426, and Qwen2.5-14B achieves a 277.3 % relative improvement compared to the zero-shot baseline, which verifies the effectiveness of cross-model collaboration. Compared with the traditional NLP model, it significantly enhances LLMs' knowledge transfer capability and domain adaptability. It provides an innovative paradigm for intelligent knowledge graph construction through the collaboration of traditional NLP techniques and prompt engineering, effectively compensating for the inherent limitations of LLMs.

Conventional electronic skin (e-skin) that combines non-contact sensing with contact sensing relies on the combination of sensors, greatly hindering its application in human-machine interface (HMI). Herein, a triboelectric nanogenerator based on Ecoflex/PEDOT:PSS (EP-TENG) is proposed to achieve both contact and noncontact sensing. The EP-TENG comprises an Ecoflex/PEDOT:PSS triboelectric layer with a cicada winginspired microstructure and MXene electrode with superior conductivity. Benefit from the high specific surface area brought by cicada wing micropillars and the charge retention ability of PEDOT:PSS, The EP-TENG has good non-contact sensing capability within a range of 80 mm and also has excellent performance in contact mode, with an open circuit voltage (Voc) of 124 V and a power density of 75.3 mW/m2. Importantly, our work develops a dual-mode control strategy that combines non-contact gesture recognition and contact-based interactive control. We demonstrate the EP-TENG dual-mode control through a Tetris unlocking system, which makes up for the lack of precise control in the non-contact mode and shows its important application potential in multimode human-computer interaction in complex scenarios in the future.

Secure outsourced computation is crucial for data confidentiality protection in cloud environments, while enabling efficient and flexible computations across various fields. Most of the existing secure computation protocols focus on integer operations, which limits their potential to handle practical realworld applications requiring precise floating-point arithmetic. In this paper, we propose SOCIE, a protocol to extend the domain of secure computation to floating-point operations by incorporating a novel encoding scheme. Our design is based on a suite of secure arithmetic protocols including secure comparison (SCMP), secure multiplication (SMUL) and secure sign bit-acquisition (SSBA), yet optimized for floating-point data. Further, we utilise the threshold Paillier encryption for efficient encryption, decryption and secure multi-party computation to ensure strong preservation of privacy. Theoretical analysis justifies the correctness and security of SOCIE; Experimental analyses show promising results on computational efficiency and acceptable communication overhead. Allowing SOCIE to be a feasible solution for practical secure floating-point computations in a cloud environment.

Sorting over encrypted data is a fundamental building block for privacy-preserving data processing but remains a significant performance challenge. While solutions based on Fully Homomorphic Encryption (FHE) provide powerful capabilities, they often incur substantial computational overhead. To address this, we present an efficient two-party protocol for privacypreserving sorting that operates in a lightweight twin-server model. Our design leverages the data-oblivious properties of the Bitonic sorting network, built upon the (2,2)-Threshold Paillier cryptosystem. The core of our protocol is a novel, highperformance secure compare-and-swap primitive. We further introduce a communication-optimized variant that significantly reduces network latency by employing an alternating initiator model and batch processing of comparison tasks. Our experimental evaluation demonstrates that our optimized protocol substantially outperforms a state-of-the-art FHE framework, achieving speedups of up to $28 \times$ in total runtime and $1500 \times$ in amortized runtime for sorting 128 elements. Our work provides a practical and highly efficient solution for the fundamental task of secure sorting.

Secure outsourced computation is a crucial technique for ensuring data security and privacy in cloud computing. While fully homomorphic encryption (FHE) enables computations on encrypted data, its ability to support an unlimited number of arithmetic operations comes at a high computational cost. In recent years, secure computation schemes based on interactions between multiple computation servers and partially ho-momorphic encryption (PHE) have been proposed. These schemes allow efficient execution of unlimited addition and multiplication operations on encrypted data without introducing noise. However, existing solutions often have functional limitations—for example, they may only support computations on natural numbers—or they risk leaking information about the underlying data. To address these challenges, this paper introduces BATCH, a secure outsourced batch computation scheme based on PHE and a dual-server architecture. Compared to existing solutions, BATCH supports batch multiplication and batch squaring on encrypted integers, rather than just natural numbers, while significantly enhancing both security and accuracy. Theoretical analysis and experimental results demonstrate that BATCH outperforms existing approaches in both computational and communication efficiency.

The application of artificial intelligence (AI) technologies in intelligent systems has become increasingly widespread. However, training high-performance AI models still relies on large-scale datasets, exposing them to data poisoning attacks. In this paper, we propose a reconstructionbased data poisoning detection technique to protect AI models from malicious manipulation, ensuring their security, robustness, and normal functionality. The method utilizes diffusion models to reconstruct input images and generate their “clean” counterparts. By incorporating an attention mechanism, the model focuses on critical features and patterns in the data. Due to the anomalous characteristics of poisoned samples, their attention weight distributions exhibit significant deviations. By comparing the differences between original and reconstructed data, the attention mechanism effectively identifies abnormal or maliciously altered samples. This approach integrates the reconstruction capability of diffusion models with the feature sensitivity of attention mechanisms, enabling reliable detection of poisoned samples and preserving the integrity of machine learning models during training. Experimental results demonstrate the method’s high detection accuracy and robustness across various attack scenarios.

As a key material in aerospace applications, carbon fiber reinforced polymer (CFRP) requires precision grinding to maintain the accuracy and integrity of assembly surfaces. However, machining damage remains a significant technical bottleneck that affects the service performance of CFRP components. To address this, ultrasonic vibration-assisted minimum quantity lubrication (MQL) grinding with nanolubricants was developed. Despite this advancement, the formation and evolution of damage under the inherent random properties of grinding grains remain unclear. In this study, the instantaneous dynamic generation, conduction, and accumulation of grinding heat, along with its thermal effects, were examined. Additionally, 3D microscale three-phase finite element numerical models were established to investigate the mechanisms of material removal and damage formation for different fiber orientation angles (FOAs) using single-grain grinding. Grinding experiments were conducted to evaluate the damage distribution and quantitatively compare the effects of different ultrasonic vibration and nanolubricant combinations. The results indicate that the grinding damage mechanisms vary with FOA, complementing the debris generation mechanisms. Damage is primarily distributed along the grinding direction, with the lowest damage observed at 0 degrees FOA and the highest at 45 degrees FOA. The 2D ultrasonic-assisted nanolubricant MQL with a 90 degrees coupled vibration angle (theta U) provided the optimal damage reduction, achieving a maximum decrease of 27.56% compared with conventional grinding. In contrast, a 45 degrees theta U showed excellent damage suppression only at 135 degrees FOA, with a reduction of 40.38%. This article provides theoretical support for the damage suppression strategy in CFRP grinding.Highlights Instantaneous dynamic thermal generation of CFRP grinding was investigated. 3D mesoscopic CFRP grinding mechanical-thermal FEM models were established. Mechanisms of damage generation, evolution, and debris formation were revealed. Distribution and comparison of damages were quantitatively characterized.

Machining-induced damages encountered during the grinding of titanium alloys are a major setback for processing different components from these materials. Recent studies have shown that nanofluid (NF)-based minimum quantity lubrication (MQL) systems improved the machining lubrication and the titanium alloys’ machinability. In this work, the tribological characteristics of a palm oil-based tripartite hybrid NF (ZnO/Al 2O 3/Graphene Oxide, GO) are studied. The novel usage of the developed lubricants in MQL systems was examined during the grinding of Ti6-Al-4V (TC4) alloy. The NF was produced by mixing three weight percent mixtures (i.e., 0.1, 0.5, and 1 wt.%) of the nanoparticles in palm oil. A comprehensive tribological and physical investigation was conducted on different percentage compositions of the developed NF to determine the optimum mix ratio of the lubricant. The findings indicate that increasing the NF concentration caused an increment in the dynamic viscosity and frictional coefficient of the NFs. The tripartite hybrid NF exhibited superior tribological and physicochemical properties compared with the pure palm and monotype-based NFs. Moreover, the dynamic viscosity of the tripartite-hybrid-based NFs increased by 12%, 5%, and 11.5% for the Al 2O 3, GO, and ZnO hybrid NFs, respectively. In addition, the machining results indicate that the tripartite hybrid NF lowered the surface roughness, specific grinding, grinding force ratio, tangential, and normal grinding forces by 42%, 40%, 16.5%, 41.5%, and 30%, respectively. Hence, the tripartite hybrid NFs remarkably enhanced the tribology and machining performance of the eco-friendly lubricant.

Download This Paper Open PDF in Browser Add Paper to My Library Share: Permalink Using these links will ensure access to this page indefinitely Copy URL Copy DOI

With the integration of smart grid and digital twin technologies, federated learning becomes pivotal in ensuring privacy-preserving AI model training. In such scenarios, each user within the smart grid is geographically dispersed, prohibiting the sharing of raw data to safeguard privacy. However, this setup is susceptible to backdoor attacks. When an attacker initiates such an attack, the smart grid model is manipulated to produce the attacker's desired output, while maintaining normal functionality under standard conditions. The inherent distributed nature of federated learning complicates the detection of backdoor attacks. To address this, we model the federated learning process within a digital twin as a cooperative game. By employing the Shapley value to measure the contribution of nodes in each training round, we propose an outlier detection algorithm to identify malicious users. The effectiveness of the proposed algorithm has been evaluated via extensive experiments.

ObjectiveThis study aimed to investigate the effects of 6 wks of peroneal longus neuromuscular electrical stimulation on the balance control ability in young adults with chronic ankle instability. DesignThis study is a double-blind randomized controlled trial. Six weeks of neuromuscular electrical stimulation and placebo intervention were conducted in the neuromuscular electrical stimulation and control groups for 20 mins, three times a week, respectively. Thirty-eight participants successfully completed the whole intervention and single-leg standing tests. The kinetics data of the center of pressure trajectory during static single-leg stance were measured using a Kistler force platform. Two-way repeated measures analysis of variance was used to analyze the electrical stimulation effects. ResultsSignificant interactions were detected in Cumberland Ankle Instability Tool scores and all balance parameters including displacement X, displacement Y, 95% confidence ellipse area, root-mean-square, and center of pressure mean displacement velocity (P < 0.05, 0.103 <= eta 2 <= 0.201). Significant between-group differences were found in Cumberland Ankle Instability Tool scores (P = 0.003, Cohen's d = 0.215), displacement X (P = 0.045, Cohen's d = 0.107), root-mean-square ml (P = 0.019, Cohen's d = 0.143), and 95% confidence ellipse area (P = 0.031, Cohen's d = 0.123) after the 6-wk interventions. ConclusionsSix weeks of neuromuscular electrical stimulation on the peroneus longus can improve static balance control ability in young adults with chronic ankle instability, especially the stability of ankle frontal plane.

SiCp/Al composites, known for their outstanding properties, are widely used in aerospace, automotive, and other fields. Despite extensive research, a complete theoretical framework and evaluation system for processing these composites have not yet been established. To enhance their processability, it is crucial to understand how processing-induced characteristics affect their machinability and to establish a robust evaluation system. This study addresses this gap by first reviewing the influence of particle properties on the mechanical properties of SiCp/Al composites. Next, both conventional and advanced machining mechanisms are analyzed, and the processability based on the cutting/grinding forces and surface roughness is evaluated. The results show that advanced machining techniques significantly improve processability compared to conventional methods. The ultrasonic elliptical vibration-assisted turning of the 25 vol% SiCp/Al6061 composites resulted in an 82.4% reduction in the cutting force. Laser-assisted turning of 45 vol% SiCp/Al composites achieved an 89.5% improvement in surface roughness (Sa). Finally, we identify research gaps and future challenges in processing SiCp/Al composites, providing technical support and theoretical guidance for both industry and academia.