Applied Sciences

23 pages, 2817 KB

Open AccessArticle

Single-Scale EMA Latent Memory for Strictly Causal Dense Arm-Motion Recognition from Multi-Channel sEMG

by Xu Luo, Qianxiang Luo and Yan Zhang

Appl. Sci. 2026, 16(15), 7862; https://doi.org/10.3390/app16157862 - 6 Aug 2026

Strictly causal dense arm-motion recognition from surface electromyography (sEMG) requires past context without future samples. We evaluated single-scale exponential moving average (EMA) latent memory as a lightweight causal temporal mechanism under a controlled, matched design. Causal CNN, causal CNN–Transformer (CT), and causal CNN [...] Read more.

Strictly causal dense arm-motion recognition from surface electromyography (sEMG) requires past context without future samples. We evaluated single-scale exponential moving average (EMA) latent memory as a lightweight causal temporal mechanism under a controlled, matched design. Causal CNN, causal CNN–Transformer (CT), and causal CNN with EMA memory (CEMA) were tested on a 30-subject, eight-channel, seven-class dataset in five subject-independent folds. At 300, 500, and 800 ms, five deterministic seeds were averaged within-fold before inference. CEMA accuracy exceeded CNN accuracy by 9.105, 9.313, and 9.657 percentage points, respectively (all Holm-adjusted paired-t p < 0.001). CEMA–CT differences were +0.209, −0.240, and −0.715 points, respectively; confidence intervals included zero, and directions varied across seeds. With parameters frozen, resetting memory at the current-window start reduced CEMA accuracy by 10.928–17.009 points. CEMA–CT B100 recall differences were −4.97, −12.84, and −16.54 points, respectively, whereas absolute-delay differences were 6.79, 7.71, and 6.94 ms, respectively, all below 8 ms. At 300 ms, CEMA used 6968 parameters and 0.751 M MACs, 48.0% fewer MACs than CT; CT required 2.31× CEMA’s full-buffer host-wall time. CEMA offers a favorable average accuracy–efficiency trade-off, although CT remains more responsive at transitions. Timing reflects full-buffer host inference only, not incremental, embedded, or end-to-end deployment. Full article

► Show Figures

Figure 1

25 pages, 1132 KB

Open AccessArticle

Exploration of Convergence Point Characteristics in Near-Sea-Surface Wireless Power Transfer

by Jinyi Hui, Hongguang Wang, Chunliang Liu and Wen Ding

Appl. Sci. 2026, 16(15), 7861; https://doi.org/10.3390/app16157861 - 6 Aug 2026

Abstract

This study investigates the enhancement of long-distance microwave wireless power transfer (WPT) performance by utilizing naturally formed evaporation ducts near the sea surface. The parabolic equation method is applied to analyze how emission height, frequency, evaporation duct height and receiving aperture influence multiple [...] Read more.

This study investigates the enhancement of long-distance microwave wireless power transfer (WPT) performance by utilizing naturally formed evaporation ducts near the sea surface. The parabolic equation method is applied to analyze how emission height, frequency, evaporation duct height and receiving aperture influence multiple convergence points and power distribution. Results show that increasing emission height transforms the power pattern from a single- to a multi-layer structure, improving spatial coverage and providing additional transmission options while slightly reducing near-surface power concentration. Higher frequencies strengthen beam confinement, with the power ratio P_e at 12 GHz, about 2.7 times higher than that at 8 GHz. Furthermore, a higher EDH enhances power trapping, leading to a more complex vertical field structure and increased field strength. A larger receiving aperture also improves performance, with a 3 m aperture at 12 GHz achieving 2.5 times the power ratio of a 1 m aperture. Optimizing these parameters facilitates consistent and efficient power delivery over long distances. The findings provide theoretical support for developing sustainable wireless power transmission systems, particularly for offshore platforms, maritime vessels, and remote-area power networks. Full article

19 pages, 42112 KB

Open AccessArticle

A Novel TabPFN-Based Framework for Landslide Susceptibility Assessment: Multi-Model Comparison in Qingchuan County, Sichuan, China

by Guogen Shui, Chong Li, Yixiang Du, Wenjun Zhang, Quanhong Zheng, Yitong Yao, Rong Hu, Yijun Lu and Jialun Cai

Appl. Sci. 2026, 16(15), 7860; https://doi.org/10.3390/app16157860 - 6 Aug 2026

Abstract

Landslides are among the most common and destructive geological hazards in complex mountainous regions. Developing accurate and interpretable susceptibility assessment models is important for regional geological hazard prevention and spatial safety management. To address the limitations of traditional machine learning models in processing [...] Read more.

Landslides are among the most common and destructive geological hazards in complex mountainous regions. Developing accurate and interpretable susceptibility assessment models is important for regional geological hazard prevention and spatial safety management. To address the limitations of traditional machine learning models in processing tabular geoscience data with limited samples and complex nonlinear relationships, this study took Qingchuan County, Sichuan Province, as the study area and introduced the Tabular Prior-data Fitted Network (TabPFN) for landslide susceptibility assessment. Its performance was systematically compared with four commonly used models: Logistic Regression (LR), Support Vector Machine (SVM), Random Forest (RF), and Extreme Gradient Boosting (XGBoost). An evaluation system was built using 12 conditioning factors, including elevation, lithology, and distance to roads (DTR). The SHapley Additive exPlanations (SHAP) method was further used to identify the contributions and effects of the main controlling factors. The results showed that TabPFN achieved the highest observed predictive performance among the evaluated models, with an AUC of 0.7834, showing higher observed performance than traditional models such as LR, SVM, RF, and XGBoost, indicating its strong ability to capture nonlinear relationships and good applicability under limited sample conditions. SHAP analysis further indicated that elevation, lithology, and distance to roads were the main controlling factors influencing the model predictions. This study demonstrates the applicability of TabPFN for landslide susceptibility assessment in complex mountainous regions and provides a new methodological reference for intelligent geological hazard prediction and refined regional risk management under limited data conditions. Full article

(This article belongs to the Special Issue AI-Enhanced Slope Stability and Landslide Risk in Transport Infrastructure)

► Show Figures

Figure 1

24 pages, 4735 KB

Open AccessArticle

Accurate and Interpretable Prediction of Exploration Input–Output Matching Under Data Scarcity: An Ensemble Learning Framework

by Xiao Chen, Hui Liu, Weiyun Zhan, Haitao Li, Yu Cao and Yuan Liang

Appl. Sci. 2026, 16(15), 7859; https://doi.org/10.3390/app16157859 - 6 Aug 2026

Abstract

Accurate prediction of input–output relationships in natural gas exploration is essential for improving exploration efficiency and optimizing investment allocation. However, this task is severely hindered by data sparsity and strong nonlinear characteristics inherent in oil and gas exploration systems, rendering conventional statistical methods [...] Read more.

Accurate prediction of input–output relationships in natural gas exploration is essential for improving exploration efficiency and optimizing investment allocation. However, this task is severely hindered by data sparsity and strong nonlinear characteristics inherent in oil and gas exploration systems, rendering conventional statistical methods and single machine learning models ineffective. This study develops a novel integrated framework combining data augmentation, nonlinear feature engineering, and ensemble learning to achieve accurate and interpretable prediction of exploration input–output matching under limited data constraints. Taking four core exploration indicators—including the number of exploration wells, total drilling depth, reserve abundance, and proven reserves—as input variables, adaptive prediction models were constructed for seven typical hydrocarbon basin exploration systems. To ensure comprehensive algorithmic exploration, nine advanced algorithms, including mainstream ensemble methods (RandomForest), few-shot neural networks (FewShot_NN), and kernel-based regressions, were systematically benchmarked. Furthermore, SHapley Additive exPlanations (SHAPs) was adopted to enhance model interpretability, and non-parametric Wilcoxon signed-rank tests were introduced to rigorously validate statistical significance. The results demonstrate that the optimal predictive pathway varies across different geological systems. Specifically, RandomForest and GBDT exhibit superior performance in systems with moderate heterogeneity (e.g., Jialingjiang and Changxing–Feixianguan Formations), whereas FewShot_NN and Kernel Ridge achieve the highest accuracy under extreme data sparsity and volatility (e.g., Xujiahe Formation and Lower Permian). The established framework yields a coefficient of determination (R²) greater than 0.96 for the majority of study cases, with overall absolute percentage errors heavily minimized. SHAP analysis further verifies that drilling depth and reserve abundance are the dominant controlling factors. This data-driven framework provides a robust and interpretable technical tool for the intelligent management of energy resources. Full article

(This article belongs to the Topic Advanced Technology for Oil and Nature Gas Exploration)

18 pages, 2068 KB

Open AccessArticle

Intelligent Grinding Path Planning for Surface Crack Removal in Special Steel Bars

by Ziliang Li, Ruolei Wang, Hongjian Tan, Jinzhu Zhang and Tao Wang

Appl. Sci. 2026, 16(15), 7858; https://doi.org/10.3390/app16157858 - 6 Aug 2026

Abstract

To address the demand for efficient and automated removal of surface crack defects in special steel bars, this paper proposes a grinding path planning method based on hybrid ACO–SA optimization. A geometric model of surface cracks is established, and the key factors influencing [...] Read more.

To address the demand for efficient and automated removal of surface crack defects in special steel bars, this paper proposes a grinding path planning method based on hybrid ACO–SA optimization. A geometric model of surface cracks is established, and the key factors influencing grinding efficiency are analyzed. Crack defects are represented as line segments with defined start and end coordinates, which effectively preserves defect continuity and overcomes the limitations of scattered point-based representations. To enhance optimization performance and avoid premature convergence, a task-oriented grinding path planning framework based on a hybrid ant colony optimization (ACO) and Simulated Annealing (SA) strategy is developed, combining the global search capability of ACO with the local optimization strength of SA. Experimental results for 10 m long special steel bars containing 100 randomly distributed cracks demonstrate that the proposed method achieves an average planning time of 9.4 s and a maximum path fluctuation of 324 mm, ensuring both computational efficiency and path stability. Validation using real surface defect data further confirms the effectiveness and practicality of the proposed approach for industrial grinding applications. The proposed method provides an effective solution for automated grinding of surface cracks in special steel bars and offers practical potential for automated defect repair in the steel manufacturing industry. Full article

(This article belongs to the Special Issue Modern Processing Routes for Metallic Alloys)

31 pages, 2403 KB

Open AccessArticle

Salinity-Driven Modulation of Growth and FAME Composition in Auxenochlorella protothecoides for Industrial Applications

by Thomas Morra, Samaneh Mohammadnejad, Veronica Lolli, Francesco Sansone, Ali Parsaeimehr, Giovanni Antonio Lutzu and Alessandro Concas

Appl. Sci. 2026, 16(15), 7857; https://doi.org/10.3390/app16157857 - 6 Aug 2026

Abstract

Microalgal lipid production requires cultivation strategies that reduce freshwater demand while maintaining biomass productivity and improving the quality of fatty acid methyl ester (FAME) profiles for downstream applications. In this context, salinity is a relevant but still insufficiently characterized factor, particularly for Auxenochlorella [...] Read more.

Microalgal lipid production requires cultivation strategies that reduce freshwater demand while maintaining biomass productivity and improving the quality of fatty acid methyl ester (FAME) profiles for downstream applications. In this context, salinity is a relevant but still insufficiently characterized factor, particularly for Auxenochlorella protothecoides, whose response to salt stress in terms of growth dynamics, lipid productivity, and FAME composition remains poorly understood. This study investigated the effect of NaCl concentrations ranging from 0 to 50 g L⁻¹ on the growth, lipid accumulation, FAME profile, and predicted biodiesel-related properties of A. protothecoides. Growth kinetics were described using logistic and Gaussian models. The highest carrying capacity was observed at 10 g L⁻¹ NaCl, whereas the modeled optimum for specific growth rate occurred under moderate salinity, close to 20 g L⁻¹ NaCl. Lipid analyses showed that C16–C18 fatty acids (FAs) dominated across treatments, accounting for more than 90% of total FAMEs, and that increasing salinity shifted the profile from a more saturated C16:0-rich composition toward higher proportions of unsaturated C18 FAs, particularly oleic and linoleic acids. Biodiesel property estimations indicated that mild salinity (5 g L⁻¹) improved some fuel-relevant parameters relative to the other salinity treatments, including cetane number, iodine value, cold-flow-related properties, and oxidative stability. Nevertheless, the predicted viscosity values were below the EN 14214 specification range for all treatments, and oxidative stability at 5 g L⁻¹ only marginally met the European minimum requirement, indicating partial rather than full compliance with biodiesel standards. These findings indicate that salinity can be used as a practical tool to tune biomass production and lipid quality in A. protothecoides, supporting its potential use in microalgal biorefineries and oleochemical applications. Furthermore, the ability of the strain to tolerate saline conditions may support the future use of brackish water, seawater, or saline waste streams, potentially reducing freshwater demand; however, this environmental benefit was not quantified in the present study and should be validated through dedicated water-footprint or life-cycle assessment. Direct biodiesel application also remains limited by incomplete compliance with fuel standards, and further optimization through cultivation in real saline or wastewater-based media, process scale-up, blending, or downstream upgrading will be required to improve industrial feasibility. Full article

(This article belongs to the Section Environmental Sciences)

► Show Figures

Figure 1

25 pages, 4114 KB

Open AccessArticle

Data-Driven Intelligent Analysis System for Monitoring and Anomaly Detection in Hydrogen Refueling Station

by Minsu Kim, Seongseop Kim, Seungwoo Lee, Youngmin Kwon, Kai Oehring and Robert Bock

Appl. Sci. 2026, 16(15), 7856; https://doi.org/10.3390/app16157856 - 6 Aug 2026

Abstract

Hydrogen refueling station (HRS) requires continuous safety monitoring, yet conventional management relies largely on periodic inspection and manual oversight, limiting proactive risk mitigation. This study presents a data-driven intelligent analysis platform for real-time monitoring and anomaly detection of HRS safety data, including pressure, [...] Read more.

Hydrogen refueling station (HRS) requires continuous safety monitoring, yet conventional management relies largely on periodic inspection and manual oversight, limiting proactive risk mitigation. This study presents a data-driven intelligent analysis platform for real-time monitoring and anomaly detection of HRS safety data, including pressure, temperature, and flow-rate measurements from compressors, storage tanks, and dispensers. The platform integrates data collection adapters, a time-series database, and machine learning-based diagnostic modules (regression, clustering, and classification) into a unified reference software framework. For anomaly detection, an unsupervised LSTM-Variational Autoencoder trained on normal operating data is combined with DBSCAN-based clustering and a Mann–Kendall trend test to jointly identify point anomalies and pattern-level drifts, addressing the scarcity of labeled abnormal data in HRS environments. A continual learning mechanism further adapts detection thresholds to gradual and abrupt pattern changes without full retraining. The system was deployed and validated at BAM’s demonstration hydrogen refueling station in Germany, integrated with a remote safety-monitoring system and confirmed through performance testing, demonstrating reliable, proactive hydrogen safety management. Full article

(This article belongs to the Section Computing and Artificial Intelligence)

16 pages, 6276 KB

Open AccessArticle

Research on Electric Field Distribution and Shielding Measures for Houses near a 1000 kV UHV AC Transmission Line

by Haosheng Dai, Weifang Yao, Xueying Hua, Jian Chen, Jizhong Xi, Fangmin Liu, Jing Yu, Chao Ji, Longxu Tan and Wangling He

Appl. Sci. 2026, 16(15), 7855; https://doi.org/10.3390/app16157855 - 6 Aug 2026

Abstract

In recent years, UHV AC/DC transmission projects in China have developed rapidly, and transmission line corridors have become increasingly limited. As a result, UHV AC transmission lines are increasingly located near areas where residents live and work, and the electric field distribution around [...] Read more.

In recent years, UHV AC/DC transmission projects in China have developed rapidly, and transmission line corridors have become increasingly limited. As a result, UHV AC transmission lines are increasingly located near areas where residents live and work, and the electric field distribution around houses near transmission lines has become a major public concern. To further investigate the electric field around houses near UHV transmission lines, a full-scale house platform was constructed near an actual operating 1000 kV AC transmission line. The electric field distribution above the house platform and around the house was systematically measured and analyzed. The effects of house height and the distance between the house and the transmission line on the electric field distribution were discussed, with emphasis on the electric field distribution on the two-story platform. In addition, the shielding effect of shielding wires installed near the house on the electric field of the house platform was analyzed. The results show that the electric field on the two-story platform is significantly higher than that on the single-story platform, and the electric field decreases approximately linearly with distance. Installing shielding wires can effectively reduce the power-frequency electric field intensity on the house platform. A relatively optimal balance between shielding performance and installation economy can be achieved when the shielding wire is installed along the edge of the house, with a length 3 m longer than the house edge and a height 1.5 m higher than the position to be shielded. A reasonable combination of multiple shielding wires provides a much better shielding effect than a single shielding wire. Full article

(This article belongs to the Section Electrical, Electronics and Communications Engineering)

► Show Figures

Figure 1

14 pages, 874 KB

Open AccessArticle

Exploratory Analytical Characterization of Phenolic Profiles and Antioxidant Activity in Three Lavender Genotypes

by Mihaela Bogdan, Mihai Octavian Botea, Manuela Bianca Pașca, Mirela Delia Țiț, Anamaria Lavinia Purza, Tiberiu Sebastian Nemeth, Manuel Alexandru Gîtea, Gabriela S. Bungau and Daniela Gîtea

Appl. Sci. 2026, 16(15), 7854; https://doi.org/10.3390/app16157854 - 6 Aug 2026

Abstract

This exploratory study investigated the phenolic composition and antioxidant activity of dried inflorescence extracts prepared from three commercially cultivated lavender materials: Lavandula angustifolia ‘Moldoveanca 4’, Lavandula angustifolia ‘Sevtopolis’, and Lavandula × intermedia ‘George 90’. Using HPLC-DAD-ESI⁺ analysis, fourteen phenolic compounds predominantly belonging [...] Read more.

This exploratory study investigated the phenolic composition and antioxidant activity of dried inflorescence extracts prepared from three commercially cultivated lavender materials: Lavandula angustifolia ‘Moldoveanca 4’, Lavandula angustifolia ‘Sevtopolis’, and Lavandula × intermedia ‘George 90’. Using HPLC-DAD-ESI⁺ analysis, fourteen phenolic compounds predominantly belonging to the hydroxybenzoic acid, hydroxycinnamic acid, and flavone classes were tentatively annotated, and their levels were estimated as class-equivalent concentrations. Descriptive quantitative differences were observed among the three extracts for all fourteen compounds. The chromatographic sum of their class-equivalent concentrations showed the highest value in ‘Sevtopolis’ (1064.44 ± 1.03 µg/mL), followed by ‘George 90’ (924.03 ± 1.13 µg/mL) and ‘Moldoveanca 4’ (825.87 ± 2.3 µg/mL). The ‘Sevtopolis’ extract also showed the highest DPPH radical-scavenging activity, followed by ‘George 90’ and ‘Moldoveanca 4’. Taken together, these findings provide a preliminary analytical characterization of three specific lavender extracts and describe sample-specific phenolic and DPPH patterns under the conditions employed, without supporting genotype-level inference. Full article

(This article belongs to the Special Issue Biological Activities of Plant Extracts and Their Applications)

► Show Figures

Figure 1

18 pages, 3029 KB

Open AccessArticle

Low-Temperature EMF Stability of Mineral-Insulated Thermocouples for Pressurized Water Reactor Core Outlet Temperature Measurement

by Jie Chen, Xiaodong Peng, Min Liu, Anzhong Zhao, Meiliang Huang and Shuzhi Chen

Appl. Sci. 2026, 16(15), 7853; https://doi.org/10.3390/app16157853 - 6 Aug 2026

Abstract

The core outlet temperature is an important safety parameter for pressurized water reactors (PWRs), and its accurate determination depends on highly reliable thermocouple sensors. In this study, two typical commercial K-type thermocouple alloys (de1 and de2) were used to fabricate mineral-insulated metal-sheathed (MIMS) [...] Read more.

The core outlet temperature is an important safety parameter for pressurized water reactors (PWRs), and its accurate determination depends on highly reliable thermocouple sensors. In this study, two typical commercial K-type thermocouple alloys (de1 and de2) were used to fabricate mineral-insulated metal-sheathed (MIMS) thermocouples. A comparative severe accelerated aging test at 500 °C for 168 h was performed to investigate electromotive force (EMF) stability under high-temperature exposure conditions. After aging, both alloy types exhibited obvious positive EMF drift, and the maximum drift magnitude appeared at the calibration temperature of 400 °C. The measured irreversible EMF drift of de2 thermocouples was lower than that of de1 thermocouples. For the same de2 alloy, thermocouples fabricated with pre-oxidized thermoelement wires presented smaller irreversible drift than those made of bright wires. Reversible EMF drift and heating–cooling calibration hysteresis were more significant for de1 specimens. Stabilization treatment at 570 °C for 2 h followed by furnace cooling effectively reduced the thermal hysteresis of both alloys during temperature cycling. This study provides straightforward experimental data and practical processing references for optimizing the manufacturing route of nuclear-grade MIMS thermocouples with improved high-temperature EMF stability. Full article

22 pages, 3498 KB

Open AccessArticle

Recycling Versus Landfilling of Plastic Packaging: Comparative Life Cycle Energy Implications Under Circular Economy Conditions

by Patrycja Bałdowska-Witos, Izabela Piasecka, Zbigniew Kłos and Andrzej Tomporowski

Appl. Sci. 2026, 16(15), 7852; https://doi.org/10.3390/app16157852 - 6 Aug 2026

Abstract

Plastic packaging is indispensable in modern food and beverage supply chains. However, its production and end-of-life management require substantial primary energy inputs. Improving the energy efficiency of packaging systems is therefore a key objective of circular economy strategies. This study applies Life Cycle [...] Read more.

Plastic packaging is indispensable in modern food and beverage supply chains. However, its production and end-of-life management require substantial primary energy inputs. Improving the energy efficiency of packaging systems is therefore a key objective of circular economy strategies. This study applies Life Cycle Assessment (LCA) to evaluate the cumulative energy demand (CED) of a beverage packaging system comprising polyethylene terephthalate (PET) bottles, recycled polyethylene terephthalate (rPET) bottles, high-density polyethylene (HDPE) caps, and polypropylene (PP) labels under two end-of-life scenarios: recycling and landfilling. The assessment was performed using Polish production and waste-management conditions (2022–2024). A functional unit of 100,000 complete 1 L beverage packages was adopted, and the life cycle inventory combined primary industrial data with background datasets from the ecoinvent database. Recycling scenarios were modelled using a substitution-based allocation (system expansion) approach. The results indicate that non-renewable energy accounted for approximately 88–97% of the total cumulative energy demand of the analysed packaging components. Among the investigated materials, PP labels exhibited the highest component-specific cumulative energy demand, whereas rPET bottles showed the lowest values, corresponding to an approximately 13% lower energy demand than conventional PET bottles under the adopted modelling assumptions. Recycling consistently outperformed landfilling by reducing cumulative primary energy demand across all analysed impact categories, with the greatest energy-saving potential observed for PP labels. Negative CED values obtained for selected recycling scenarios represent avoided primary energy resulting from virgin material substitution rather than physically negative energy consumption. The findings demonstrate that component-level LCA provides a robust basis for identifying energy hotspots within plastic packaging systems and supports evidence-based decisions regarding eco-design, material selection, and recycling strategies. Although the absolute CED values are specific to Polish production and recycling conditions, the overall results confirm that material recycling is a more energy-efficient end-of-life strategy than landfilling and contributes to the development of more sustainable circular packaging systems. Full article

(This article belongs to the Special Issue Life Cycle Assessment (LCA) as a Tool for Evaluating Sustainable Packaging Systems)

► Show Figures

Figure 1

44 pages, 8074 KB

Open AccessArticle

Multiple Imputation of Missing Traffic Volume: An Advanced Framework and Multi-Domain Validation

by Zaid Abdulzahra Mahdi Mandalawi and Halit Özen

Appl. Sci. 2026, 16(15), 7851; https://doi.org/10.3390/app16157851 - 6 Aug 2026

Abstract

High-frequency traffic data from remote sensors often suffer from severe gaps and multi-day blackouts. Traditional deterministic imputation fails during these extended failures, artificially destroying natural traffic variance. To resolve this, this study develops an adaptive Multiple Imputation (MI) framework to reconstruct missing 2-min [...] Read more.

High-frequency traffic data from remote sensors often suffer from severe gaps and multi-day blackouts. Traditional deterministic imputation fails during these extended failures, artificially destroying natural traffic variance. To resolve this, this study develops an adaptive Multiple Imputation (MI) framework to reconstruct missing 2-min volumes. A novel multi-tier historical median predictor with adaptive expansion (up to ±30 min) serves as a variance-protected anchor for two stochastic engines: Stochastic Linear Regression and Predictive Mean Matching with Approximate Bayesian Bootstrap (PMM-ABB). PMM-ABB features dynamic

K

-neighbor autotuning, with simulation convergence governed by a dual-metric algorithm. Performance was evaluated against a Historical Average (HA) baseline via multi-domain validation, strictly assessing the models’ ability to recover hidden, real-world ground-truth counts rather than replicating the engineered input features. Under severe block-missingness, the stochastic models prevented collapse, reducing Temporal Cross-Validation MAE from 33.7 (HA) to 22.7 (Cohen’s

d

= −0.4). Power Spectral Density matching confirmed that both models preserved macro-periodic traffic waves, keeping spectral tracking errors under 4.8 dB. Ultimately, PMM-ABB slightly outperformed Stochastic Regression in sequential time dependency and point accuracy, confirming that the framework provides a highly reliable structural proxy for continuous highway flow modeling. Full article

(This article belongs to the Section Transportation and Future Mobility)

► Show Figures

Figure 1

42 pages, 29009 KB

Open AccessArticle

A Low-Cost Electronically Controlled Pneumatic Knee with Passive Four-Bar Stance Stability and Semi-Active Swing Damping: A Single-Case Feasibility Study

by Seung-Gi Kim, Jin-Kook Park, Bum-Ki Hong, Na-Yoen Park, Chil-Yong Kwon, Se-Hoon Park and Su-Hong Eom

Appl. Sci. 2026, 16(15), 7850; https://doi.org/10.3390/app16157850 - 6 Aug 2026

Abstract

Microprocessor-controlled knee prostheses (MPKs) face limited accessibility in resource-constrained environments due to high implementation costs and excessive power consumption associated with complex actuators. This study examines the technical feasibility of a low-cost electronically controlled pneumatic knee (ECPK) that combines structural mechanics with minimal [...] Read more.

Microprocessor-controlled knee prostheses (MPKs) face limited accessibility in resource-constrained environments due to high implementation costs and excessive power consumption associated with complex actuators. This study examines the technical feasibility of a low-cost electronically controlled pneumatic knee (ECPK) that combines structural mechanics with minimal electronic control. A functional decoupling strategy was implemented: stance-phase stability is provided by passive kinematic locking of a four-bar linkage over the near-extended stance range, while a lightweight feedforward controller driven by a single joint-axis Hall sensor segments the gait cycle continuously, updates its speed estimate once per step, and adjusts the valve only for swing-phase damping. From the stance duration of the preceding steps, this controller presets the pneumatic valve orifice to compensate for mechanical response delays, so that link rotation speed is regulated semi-actively without powered actuation. System integration and control viability were evaluated in a single-case feasibility study (N = 1), in which the ECPK was compared within subject with a commercial mechanical prosthesis after a 4-week adaptation period. Despite a 400 g distal mass penalty, the semi-active control algorithm was associated with a smaller increase in step-length asymmetry at the highest speed tested. Furthermore, net oxygen cost was lower with the ECPK during high-speed walking. Because the conditions were compared at unmatched self-selected speeds and the ECPK condition reached a respiratory exchange ratio (RER) of 1.13, this observation is hypothesis-generating. Coupling passive four-bar stance stability with minimal electronic swing regulation is therefore a viable engineering basis for accessible prostheses, and the present study establishes its technical feasibility rather than its clinical effectiveness. Full article

(This article belongs to the Special Issue Advanced Robotics, Mechatronics, and Automation)

► Show Figures

Figure 1

40 pages, 695 KB

Open AccessReview

Large Language Models: From Internal Architecture and Distributed Training Optimisation to Adaptation Strategies

by Martin Lukáč, František Duchoň, Jakub Ivan, Martin Dekan, Eduard Zelenay, Maroš Kocúr, Izabela Trepáčová, Tomáš Jurov and Radoslav Pšenka

Appl. Sci. 2026, 16(15), 7849; https://doi.org/10.3390/app16157849 - 6 Aug 2026

Abstract

This paper presents a unified technical survey of Large Language Models (LLMs), connecting three layers of the modelling pipeline that existing surveys address in isolation: internal architecture, distributed training optimisation, and downstream adaptation. Its organising principle is the dependency between these layers—how a [...] Read more.

This paper presents a unified technical survey of Large Language Models (LLMs), connecting three layers of the modelling pipeline that existing surveys address in isolation: internal architecture, distributed training optimisation, and downstream adaptation. Its organising principle is the dependency between these layers—how a choice at one constrains what remains feasible at the next. The survey examines fundamental mechanisms (tokenisation, scaled dot-product attention, activation functions, and normalisation, including RMSNorm and pre- versus post-normalisation placement), then the engineering of training at scale: data, tensor, and pipeline parallelism, hybrid schemes, mixed-precision training with BF16 and FP8, ZeRO-Offload memory management, activation checkpointing, and compute-optimal scaling laws together with the conditions under which they fail. The adaptation section covers supervised and instruction fine-tuning, a comparison of parameter-efficient methods (LoRA, QLoRA, adapters, prefix and prompt tuning), Reinforcement Learning from Human Feedback with its reward-hacking failure mode, alternatives including DPO, KTO and Constitutional AI, Retrieval-Augmented Generation beyond the basic pipeline, and decoding strategies. Practical configuration guidance is given for 7B, 70B and trillion-parameter regimes. Dedicated treatments of Mixture-of-Experts architectures, long-context modelling, and hardware-aware co-design close the survey, with open challenges classified by origin and severity. Full article

(This article belongs to the Special Issue Trends and Prospects in Large Language Models: Novel Methods and Applications)

31 pages, 4059 KB

Open AccessArticle

A Novel Lightweight Framework for Real-Time Pavement Crack Segmentation Based on Knowledge Distillation

by Ning Xu, Jinghui Qiao and Yunze Tang

Appl. Sci. 2026, 16(15), 7848; https://doi.org/10.3390/app16157848 - 6 Aug 2026

Abstract

Accurate pavement crack segmentation is essential for structural health monitoring, yet existing methods often face a trade-off between segmentation accuracy and computational efficiency. To address this issue, a novel teacher–student framework, termed RTCS-T and RTCS-S, is proposed. The teacher network RTCS-T is constructed [...] Read more.

Accurate pavement crack segmentation is essential for structural health monitoring, yet existing methods often face a trade-off between segmentation accuracy and computational efficiency. To address this issue, a novel teacher–student framework, termed RTCS-T and RTCS-S, is proposed. The teacher network RTCS-T is constructed based on the Swin Transformer to capture long-range dependencies and multi-scale contextual information. To further enhance crack representation, a strip refinement module is introduced to model directional structural features, while a cascaded atrous spatial pyramid pooling module is employed to improve multi-scale feature aggregation. Based on the teacher network, a lightweight student model RTCS-S is developed by using depthwise separable convolutions to achieve efficient inference. In addition, a foreground-aware and boundary-aware knowledge distillation strategy is introduced to guide the transfer of structural and contextual information from the teacher to the student. Experiments on the Crack500, DeepCrack, and CFD datasets demonstrated competitive performance against representative segmentation models. On CFD, RTCS-S achieved an F1 Score of 0.7514 and an mIoU of 0.7962. Notably, RTCS-S required only 1.82 M parameters and 1.13 GFLOPs and achieved a model inference speed of 680 FPS on an RTX 4090 GPU. When deployed on an RDK X5 edge-computing platform, the complete pipeline achieved an end-to-end throughput of 34 FPS, with an average latency of approximately 29.4 ms and peak memory consumption of 1.8 GB. These results demonstrate that the proposed framework provides an efficient solution for automated pavement crack detection and shows strong potential for practical road inspection applications. Full article

(This article belongs to the Section Computing and Artificial Intelligence)

26 pages, 4429 KB

Open AccessArticle

Coordinated Cyber–Physical Attack Strategies in Power Systems Considering Defense Resource Allocation and Emergency Dispatch Responses

by Hongyu Fang, Hongfei Yu, Jinhua Huang, Yuhao Liu, Jikai Bi, Xudong Song, Yulin Chen, Li Yang and Zhenzhi Lin

Appl. Sci. 2026, 16(15), 7847; https://doi.org/10.3390/app16157847 - 6 Aug 2026

Abstract

Deliberate coordinated cyber–physical attacks, which combine cyber intrusions with physical disruptions, pose growing risks to the secure and reliable operation of power systems. To identify highly disruptive coordinated cyber–physical attack strategies under pre-attack defense allocation and post-attack emergency dispatch responses, a tri-level defender–attacker–defender [...] Read more.

Deliberate coordinated cyber–physical attacks, which combine cyber intrusions with physical disruptions, pose growing risks to the secure and reliable operation of power systems. To identify highly disruptive coordinated cyber–physical attack strategies under pre-attack defense allocation and post-attack emergency dispatch responses, a tri-level defender–attacker–defender (DAD)-based attack strategy optimization model is proposed. First, based on the association between substation automation control systems and transmission line operation, the mechanism of breaker-tripping attacks (BTAs) through compromised digital relays in substations is investigated. Second, a tri-level DAD model is developed to optimize coordinated BTA and physical line attack strategies while accounting for pre-attack cyber and physical defense, cascading failure propagation, and post-attack emergency dispatch responses. Then, based on duality theory, network flow models, and the column-and-constraint generation (C&CG) algorithm, an iterative solution framework consisting of a defense master problem and an attack-dispatch subproblem is constructed to capture post-attack topology updates, cascading line outages, generator redispatch and load-shedding responses. Finally, the proposed method is validated using the IEEE 39 bus and IEEE 118 bus power systems. Case study results demonstrate that the proposed model identifies the most damaging coordinated cyber–physical attack strategies and that BTA-induced topology changes and cascading failure propagation significantly affect attack target selection. The proposed solution method obtains the accurate optimal objective value while reducing computational time by 92.91% for IEEE 39 bus power system, and it successfully obtains a converged solution for IEEE 118 bus power system. Full article

(This article belongs to the Section Electrical, Electronics and Communications Engineering)

► Show Figures

Figure 1

54 pages, 5185 KB

Open AccessReview

Advances in Multi-Agent Deep Reinforcement Learning: Methods with Applications and Challenges

by Abdur Rakib, Khoa Phung, Marco Perez Hernandez and Mehmet Emin Aydin

Appl. Sci. 2026, 16(15), 7846; https://doi.org/10.3390/app16157846 - 6 Aug 2026

Abstract

Multi-agent deep reinforcement learning (MARL) extends deep reinforcement learning (DRL) to environments involving multiple interacting agents and has enabled applications in domains such as autonomous vehicles, robotics, unmanned aerial vehicles (UAVs), and multi-player games. Compared with single-agent learning, MARL introduces additional challenges, including [...] Read more.

Multi-agent deep reinforcement learning (MARL) extends deep reinforcement learning (DRL) to environments involving multiple interacting agents and has enabled applications in domains such as autonomous vehicles, robotics, unmanned aerial vehicles (UAVs), and multi-player games. Compared with single-agent learning, MARL introduces additional challenges, including non-stationarity, partial observability, multi-agent credit assignment, and scalability. This paper presents a narrative survey of recent developments in MARL and discusses major approaches proposed to address these challenges. In particular, we examine research directions centred on centralised training with decentralised execution (CTDE), value decomposition, learned communication, graph-based methods, and model-based learning. We further discuss commonly used benchmark environments and evaluation practices, highlighting considerations related to reproducibility, robustness, and generalisation. Finally, we outline open research challenges and future directions concerning theoretical understanding, sample efficiency, scalable coordination, and deployment in real-world settings. Rather than providing an exhaustive systematic review, this survey aims to offer an organised and up-to-date synthesis of recent progress in MARL. Full article

34 pages, 3791 KB

Open AccessArticle

MambaACE-YOLO: Frequency-Decoupled State-Space Modeling and Compact Higher-Order Relational Reasoning for Lightweight Real-Time Object Detection

by Jiangxiao Li, Weijie Wu, Wengang Che, Shengxiang Gao and Yang Liu

Appl. Sci. 2026, 16(15), 7845; https://doi.org/10.3390/app16157845 - 6 Aug 2026

Abstract

Lightweight real-time detectors must balance long-range contextual modeling, crossscale relational reasoning, and deployment efficiency. We present MambaACE-YOLO, an integration-oriented framework that separates computation into intra-scale encoding, cross-scale relational reasoning, sparse multi-stage feature distribution, and multiscale detection decoding. Its primary contribution is the coordinated [...] Read more.

Lightweight real-time detectors must balance long-range contextual modeling, crossscale relational reasoning, and deployment efficiency. We present MambaACE-YOLO, an integration-oriented framework that separates computation into intra-scale encoding, cross-scale relational reasoning, sparse multi-stage feature distribution, and multiscale detection decoding. Its primary contribution is the coordinated integration of prior frequency-decoupled and state-space ideas with detection-specific interfaces, targetscale leave-one-source aggregation, and physically prunable feature distribution, rather than a new frequency-transform, state-space, or hypergraph mechanism class. Building on prior frequency-decoupled hybrid visual Mamba research, its detection-oriented DMobileMamba backbone applies bidirectional state-space scanning only to low-frequency components, while directional high-frequency correction and multi-kernel depthwise convolutions preserve boundaries and local shape. Compact Partial-Channel HyperACE (CP-HyperACE) models cross-scale higher-order relations in a semantic subspace and uses target-scale leave-one-source aggregation. Selective Additive FullPAD (SA-FullPAD) projects each cross-scale increment once and selects injection paths through static, physically prunable gates. On MS COCO 2017 val, the unpruned MambaACE-YOLO-N achieves 42.5 AP with 2.6 M parameters, and MambaACE-YOLO-S achieves 48.8 AP with 9.0 M parameters. Under a common documented RTX 5090 TensorRT FP16 setting, the unpruned Nano model records 1.19 ms network-forward latency at 42.5 AP, whereas its physically pruned counterpart retains 42.4 AP and records 1.02 ms. Accuracy and latency values are single-run or single-record point estimates without reported variance, and the 42.5-AP result uses the 600-epoch schedule without a matched 600-epoch YOLOv13-N control. We distinguish published cross-paper results from same-framework, same-device measurements and assess the individual design choices through controlled ablations and physical-pruning experiments. Full article

(This article belongs to the Special Issue Advanced Computer Vision Technologies and Applications)

25 pages, 7313 KB

Open AccessArticle

Efficient Polish-Language Keyword Spotting on Microcontrollers: Compact Neural Architectures, Quantization, and On-Device Validation on the Raspberry Pi Pico 2

by Jakub Sobczyk and Krzysztof Fonał

Appl. Sci. 2026, 16(15), 7844; https://doi.org/10.3390/app16157844 - 6 Aug 2026

Abstract

Keyword spotting (KWS) is the always-on front end of voice interfaces; running it directly on microcontrollers, rather than streaming audio to the cloud, is essential for low-latency, privacy-preserving, and energy-efficient operation, yet it must reconcile high accuracy with severe memory, compute, and energy [...] Read more.

Keyword spotting (KWS) is the always-on front end of voice interfaces; running it directly on microcontrollers, rather than streaming audio to the cloud, is essential for low-latency, privacy-preserving, and energy-efficient operation, yet it must reconcile high accuracy with severe memory, compute, and energy limits. Existing small-footprint KWS solutions are developed and benchmarked almost exclusively on English and are seldom validated on physical hardware or under quantization for other languages; their transferability to typologically different, consonant-rich languages such as Polish therefore remains largely unverified. We present a KWS pipeline for Polish, deployed and benchmarked on the RP2350 microcontroller (Raspberry Pi Pico 2, Raspberry Pi Ltd., Cambridge, Great Britain).We propose three compact architectures from the convolutional (CNN), convolutional recurrent (CRNN), and depthwise separable neural network (DS-CNN) families and benchmark them against the state-of-the-art BC-ResNet on a 25-keyword Polish vocabulary using MFCC features. All models are evaluated in full precision (Float32) and after eight-bit integer (INT8) post-training quantization, with inference latency measured directly on the target hardware. BC-ResNet attains the highest full-precision accuracy (97.81%) but is the most fragile under quantization, whereas the proposed CRNN is the most accurate quantized model (94.28%), the DS-CNN the smallest (46.15 KB), and the CNN the fastest (107.5 ms); all quantized models meet a one-second real-time budget. We further show that the accuracy ranking inverts after quantization, that memory savings are highly architecture-dependent, and that phonetically similar Polish words are the dominant source of error. These results offer practical guidance for deploying small-footprint KWS in Polish and other underrepresented languages. Full article

► Show Figures

Figure 1

Journal Description

Applied Sciences

Latest Articles

Journal Menu

Journal Browser

Highly Accessed Articles

Latest Books

E-Mail Alert

News

Topics

Conferences

Special Issues

Topical Collections

Further Information

Guidelines

MDPI Initiatives

Follow MDPI