Demand for high-protein beverages is rising, yet the poor thermal stability of soy protein isolate (SPI) limits its application because processing induces aggregation and gelation. Here, a low-concentration pre-modification strategy combining preheating and laccase (LAC) was developed to improve SPI thermal stability. Preheating partially unfolded SPI and exposed buried aromatic residues, enhancing LAC-catalyzed oxidation and radical generation. This promoted intermolecular covalent crosslinking and produced more stable protein particles. LAC improved SPI thermal stability in a level- and time-dependent manner, as shown by weaker gelation and better flowability after reheating. Particle-size analysis showed that LAC promoted aggregation before reheating and that the resulting aggregates remained relatively stable afterward. Structural analyses further indicated that LAC-induced stabilization involved conformational rearrangement and strengthened non-covalent interactions. Overall, combining preheating with LAC provides an effective strategy for improving SPI thermal stability in high-protein systems.
Multimodal speech emotion recognition aims to identify speakers' emotional states by jointly utilizing linguistic semantic information and acoustic paralinguistic cues. However, existing methods generally suffer from promi nent cross-modal heterogeneity, insufficient utilization of inter-modal discrepancy information, and inadequate interaction between textual and speech representations. To address the above limitations, this paper proposes a multimodal speech emotion recognition model based on a dual cross-modal attention mechanism. The model adopts a dual-path intra-modal feature enhancement framework. For the textual path, a pre-trained language model is integrated with an emotion lexicon enhancement mechanism, and a bidirectional long short-term mem ory network is employed to capture contextual semantics and emotional prior knowledge. For the speech path, a dual-channel encoding architecture composed of temporal-frequency dual-scale Mel-frequency cepstral coeffi cients and a gated enhanced Mel spectrogram is constructed to fully extract multi-level acoustic emotional cues. In the modal fusion stage, a cross-modal differential attention module and a cross-modal matching attention module are devised. The former explicitly models inter-modal feature discrepancies to capture complementary cues under emotion-semantic inconsistency scenarios. The latter completes intra-modal semantic recalibration and comple mentary enhancement based on discrepancy information, realizing deep interactive fusion while preserving the semantic integrity of each modality. Experimental results on the public IEMOCAP and MELD datasets show that the proposed model achieves unweighted accuracy, weighted accuracy, and overall accuracy of 75.24%, 74.57%, and 74.38% on the IEMOCAP dataset, and 72.46%, 71.19%, and 72.62% on the MELD dataset, respectively, out performing multiple mainstream comparative methods. Ablation experiments and attention visualization analysis further verify the effectiveness and interpretability of the dual cross-modal attention mechanism. These results significantly outperform the comparison methods, validating the effectiveness and robustness of the proposed dual cross-modal attention mechanism.
In bearing looseness faults, the looseness between the inner ring (IR) of the bearing and the rotating shaft (RS) is usually induced by improper machining and assembly. To identify the vibration characteristics it induces, this study proposes an IR loosening–slip excitation force model that reflects actual operating conditions and faithfully reproduces the dynamic interaction of separation, collision, and reseparation between the IR and the RS. Using Hertz contact theory, and the lumped mass method, a nonlinear dynamic model of a rotor system with IR looseness is established, incorporating both IR loosening and slipping. A three-degree-of-freedom dynamic model is formulated to characterize the IR loosening–slip behavior, and its fault-free configuration is validated experimentally. On this basis, the influence of IR looseness clearance on the nonlinear vibration characteristics of the rotor system was investigated through simulation analysis. The results indicate that, as the looseness clearance increases, the amplitudes of higher-order harmonic components of the rotational frequency generally increase. There is a slip phenomenon between the IR and the rotor, which leads to a decrease in the characteristic vibration frequency of the bearing. Meanwhile, the loose fit between the rotor and the IR can also cause a decrease in the system’s critical rotational speed (CRS). Under the condition of bearing IR looseness-slip, the typical vibration characteristics of the system are correlated with the magnitude of bearing clearance. The research findings provide a valuable scientific basis and technical reference for bearing fault detection and early warning in mechanical equipment.
Knowledge graph (KG)-based recommendation has emerged as an effective solution to data sparsity by incorporating structured semantic information into user and item representations. However, existing KG-based recommendation approaches still face two core challenges: (i) over-reliance on explicit relations while insufficiently capturing latent connections between items, and (ii) feature degradation during high-order propagation, which leads to the dilution of informative signals in learned representations. To this end, we propose PLRA-KG, a pattern-derived latent relation augmentation framework for KG-based recommendation. PLRA-KG uncovers implicit item associations by generating syn-relations and syn-entities from global user preference patterns and relational structures. Specifically, PLRA-KG first performs relation-aware clustering via self-training to group items with strong association patterns. It then introduces a discernment-aware relation-pair selection mechanism to identify relation combinations that significantly influence user decisions. Based on the selected relation-pairs, a pattern-derived syn-relation generation strategy is designed to construct latent relations by jointly modeling global user preferences and relational value interactions. These generated syn-relations and syn-entities are subsequently integrated into the original KG, resulting in an augmented graph that better captures hidden semantic connectivity. Finally, PLRA-KG is optimized in a unified framework that jointly considers recommendation learning, KG embedding, and clustering objectives, enabling seamless integration with various KG-based recommendation models. Extensive experiments on multiple benchmark datasets demonstrate that PLRA-KG consistently improves both recommendation accuracy and diversity, achieving average improvements of approximately 6.64% in Recall, 7.83% in AD, 5.56% in Coverage, and 6.20% reduction in ARP. The source code is accessible at https://github.com/ZZP-RS/PLRA-KG.
Preservation packaging is essential for reducing fish spoilage and maintaining food safety and quality. In this study, fish skin gelatin (FSG), chitooligosaccharide (COS), and nisin were used to fabricate nanofiber-based hydrogel membranes with antibacterial and antioxidant functions. The membranes were prepared by electrospinning, followed by cross-linking via the Maillard reaction and subsequent absorption and swelling treatment. The cross-linked nanofiber membrane exhibited uniform fibers with smooth surfaces, homogeneous nisin distribution, and favorable water absorption and swelling behavior. Upon hydration and swelling, the resulting NCF-H-4 formed a porous hydrogel structure, exhibiting an optimal permeability profile and excellent rheological stability. Model fitting indicated that the water absorption behavior of the hydrogel membranes was well described by the pseudo-first-order model: initial water uptake followed diffusion-driven pseudo-first-order kinetics, while the contribution of pseudo-second-order kinetics increased over time. Moreover, the NCF-H-4 demonstrated excellent antibacterial and antioxidant activities and extended the shelf life of fish by 4 d at 4 degrees C. Overall, the developed hydrogel membranes show promising potential for application in aquatic product preservation.
