CNC Technology (Czechia)

Entity disambiguation is an important task in natural language processing that addresses situations where entity names mentioned in text may correspond to multiple real-world entities, with the goal of disambiguation by mapping entity names mentioned in text to unique and unambiguous entity identifiers. At present, most researches focus on entity disambiguation based on the concept level, which mainly uses contextual information to achieve entity disambiguation. However, the above methods still have some shortcomings: the word vector representation of the traditional model can not solve the polysemy problem, and the disambiguation terms can not consider the context globally to achieve entity disambiguation. In order to solve the above problems, this paper proposes a short text entity disambiguation method based on BERT model and shortest path algorithm. In this method, BiLSTM-CRF model is used in the ambiguous entity recognition stage, cosine similarity is used to perform optimal segmentation of short texts in the text segmentation stage, BERT pre-trained model is used to construct word vectors in the semantic vector construction stage, and the shortest path algorithm (SPA) is used to screen out the correct meaning of ambiguous entities in the current context in the entity disambiguation stage. The experimental results show that the three evaluation indicators of the proposed method, namely accuracy rate, recall rate and F1 score, have increased by an average of 23.37%, 26.47% and 24.98% respectively compared with the other three entity disambiguation methods. In conclusion, the proposed method can further improve the performance of entity disambiguation.

Although Li-ion conductivity has been the primary focus during decades of solid-electrolyte research, the mechanical compliance is equally important. For most state-of-the-art solid electrolytes, the mechanical compliance is characterized by the hardness above 1 GPa and Young's modulus above 15 GPa. Here, we report a particularly compliant solid electrolyte, 1.4Li2O-0.75ZrCl4-0.25AlCl3, whose hardness and Young's modulus reach 0.22 and 1.41 GPa, respectively. Meanwhile, it shows an ionic conductivity of 2.55 mS cm-1 at 25 °C and an estimated cost of $43.70 L-1, considerably lower than that of the Li2ZrCl6 solid electrolyte known for cost-effectiveness ($140.01 L-1). The improved mechanical compliance and fast Li-ion transport in 1.4Li2O-0.75ZrCl4-0.25AlCl3 enable decent cell performance. With high positive electrode active material loading above 20 mg cm-2, these two types of cells achieve areal capacities of 3.62 mAh cm-2 (85.78% capacity retention) and 3.92 mAh cm-2 (90.11% capacity retention), respectively, after 100 cycles under 0.1 C at 25 °C. The simultaneous achievement of highly competitive mechanical compliance, Li-ion conductivity, and cost-effectiveness in 1.4Li2O-0.75ZrCl4-0.25AlCl3 have the potential to pave the way for the realization of commercial, practical all-solid-state Li batteries.

Abstract Structural health monitoring of offshore jacket platforms is crucial to ensure the safety of offshore oil and gas development. At this stage, the judgment of platform structural safety state based on monitoring data mainly focuses on deterministic prediction, which often neglects the uncertainty and trend of safety state changes. So, a state detection model for offshore jacket platforms based on signal trend feature extraction is proposed in this article. First, the variational modal decomposition, along with Harris hawk optimization, was combined in this model, which was used to decompose the initial data into an intrinsic mode function (IMF) with clearer trends. Subsequently, the Holt–Winters algorithm is utilized to extract trend information from the historical data to predict the possible future changes of the IMF. Further, the Holt–Winters projection results under two different trend parameters are input into the regularized extreme learning machine to obtain the prediction intervals for the corresponding moments. Finally, a bi-objective optimization model with identification accuracy and interval width as the dual objectives is constructed to determine the reasonable interval of the platform's state change during operation, in order to monitor its safety state. The analysis results show that the proposed method achieves excellent recognition accuracy while the interval width is greatly reduced, which significantly improves the credibility of the model and can provide theoretical support for the state detection of offshore jacket platforms.

Hydroformylation, a catalytic transformation of alkenes into aldehydes using CO/H2, represents a fundamental route for the valorization of olefins in both industrial and academic settings. To address the safety and sustainability concerns associated with the direct use of carbon monoxide, recent research has increasingly focused on the development of C1-based carbonyl surrogates. In particular, formaldehyde, formic acid, and carbon dioxide have been investigated as viable alternatives capable of releasing CO in situ under appropriate catalytic conditions. These strategies have led to the emergence of “syngas-free” hydroformylation processes, which offer new possibilities for reaction design and system integration. This review provides a comprehensive overview of recent developments in the use of formaldehyde, formic acid, and CO2 as carbonyl surrogates in transition metal-catalyzed hydroformylation, highlighting advances in homogeneous and heterogeneous catalytic systems, substrate scope, and reaction mechanism.

Deep neural networks frequently degrade when distribution shifts occur. Test-Time Adaptation (TTA) offers a solution that tackles the distribution shift and prevents access to source data. At the interference, it adjusts a source model that has already been trained for the target data. One widely used method TSD-MSLC uses a feature revision technique to address TTA and suggests two strategies for feature uniformity and alignment: test time self-distillation and memorized spatial local clustering. However, noisy pseudo-labels affect its performance: (1) unreliable cluster prototypes, ignoring the impact of noise by calculating the cluster’s average of each sample’s features; (2) pseudo-label noise overfitting, assuming that all sample pseudo-labels are reliable in self-distillation; (3) neighbor noise overfitting, making the model vulnerable to labeling noise from the K nearest neighbors. To address these issues, we propose to improve TSD-MSLC from three aspects: first, for unreliable cluster prototypes, we propose Dynamic Weight Prototypes (DWP) to obtain reliable cluster prototypes by dynamic distribution weights. Second, for pseudo-label noise overfitting, we propose Confidence-based Self-Distillation (CSD) to promote reliable self-distillation learning by adding sample-level prediction uncertainty weights. Finally, for neighbor noise overfitting, we propose Nearest-neighbor Knowledge Aggregation (NKA) for more reliable alignment by forcing each target sample to converge towards the centroid of its neighbors. Extensive experiments show that these modules consistently enhance TSD-MLSC accuracy and exceed baselines on various datasets and backbones.

With the ongoing development of industrialization, the demand for inspection in industrial environments has significantly increased, leading to rising costs associated with manual inspection. In recent years, automated guided vehicles (AGVs) have matured and been increasingly applied in inspection tasks, effectively reducing the reliance on human labor and enabling continuous operation regardless of human working hours. However, the complexity of industrial terrains and the scattered distribution of inspection instruments limit the effectiveness of simple mobile platforms. Quadruped robots exhibit superior adaptability to unstructured environments, while robotic arms are capable of locating and recognizing dispersed instruments through search and tracking mechanisms. By integrating these two technologies, an intelligent inspection system combining a quadruped robot and a robotic arm is proposed. This paper explores the design and implementation of such a system and validates its feasibility and advantages through practical deployment in real industrial scenarios.

Summary Recently, the calcium oxide (CaO)-aluminum oxide (Al2O3)-silicon dioxide (SiO2)-water (H2O) hydroceramic system has demonstrated strong potential for use as a cementing material for deep wells due to its stability at high temperatures. However, the typically used aluminum source α-alumina (α-Al) has extremely low hydration reactivity. With this study, we aim to improve the participation of Al2O3 in the hydroceramic reaction by replacing α-Al with calcium aluminate (CA). CaO-Al2O3-SiO2-H2O hydroceramic systems with various Ca/Si/Al molar ratios were prepared to have a constant density of 1.65 g/cm3 and sufficiently retarded to meet the placement requirements in deep wells. The hydroceramic systems were then cured at 240°C for 2 days, 30 days, and 90 days, respectively, to study their long-term stability by various evaluation methods, including compressive strength test, water permeability test, mercury intrusion porosimetry (MIP), thermogravimetric analysis (TGA), and X-ray diffraction (XRD) analysis. Test results indicated that CA had significantly higher reactivity than α-Al, and the total hydration heat of the hydroceramic system showed a strong correlation with its compressive strength at 2 days. Overall, the high calcium systems had markedly superior performances compared with the low calcium systems, primarily due to increased silica reactivity. For the low calcium case, hydroceramic systems with a Ca/Si/Al ratio of 1:2:1 exhibited inferior performance than those with a Ca/Si/Al ratio of 2:4:1, primarily due to alumina’s competition with silica for calcium sources, which led to the formation of the wairakite (CaAl2Si4O12·2H2O) phase. For the high calcium case, hydroceramic systems with a Ca/Si/Al ratio of 2:2:1 exhibited similar performance as those with a Ca/Si/Al ratio of 4:4:1, suggesting that the content of reactive alumina had a relatively small effect in these systems. Although various systems showed complex strength retrogression behavior, the use of CA to replace α-Al improved the compressive strength by 88.4% on average in the high-calcium hydroceramic systems.