In order to solve constrained optimization problems, a dingo optimization algorithm with ε constrained method and crisscross strategy (εCDOA) was proposed. The algorithm first introduces the crisscross strategy in dingo optimization algorithm after all individuals select one hunting strategy and update their positions, so as to improve the global and local search capabilities of the obtained algorithm, which also help the algorithm jump out of the local optimum. Then, according to ε constrained method, the equality constraints were transformed into the inequality constraints, and the ε level comparison method was used instead of fitness value comparison to evaluate the qualities of the dingoes. Finally, based on the individuals’ constraint violations, the population is divided into two subgroups according to adaptive ε values. The individuals’ survival rates were calculated using the survival strategy of each subgroup, and the individuals with low survival rates were updated. The results of numerical experiments on 19 standard constrained optimization problems in CEC 2006 show that algorithm εCDOA has better optimization performance than four comparative algorithms such as dingo optimization algorithm with ε constrained method. For three classical engineering design problems, the design schemes given by algorithm εCDOA are obviously better than those given by other algorithms.

The growing sophistication of deepfake speech poses significant security threats to ASV(automatic speaker verification) systems. Current anti-spoofing models based on CNNs(convolutional neural networks) are constrained by inadequate global feature extraction and limited generalization capability against unseen spoofing attacks. To address these challenges, a novel network architecture integrating CT-DSCNet(channel-temporal attention mechanisms with depthwise separable convolutions) was proposed. Building upon the RawNet2 framework, the developed model incorporates dual-domain attention modules to enhance discriminative feature representation while suppressing irrelevant acoustic artifacts. Furthermore, depthwise separable convolutional residual blocks were strategically implemented to optimize computational efficiency and real-time processing capabilities. Comprehensive evaluations were conducted across three benchmark datasets: ASVspoof2019 LA, ASVspoof2021 DF, and FMFCC-A. Experimental results demonstrate state-of-the-art performance with EER(equal error rate) of 1.53% on ASVspoof2019 LA, representing a 70.58% relative improvement over baseline systems. Notably, the proposed architecture exhibits superior cross-dataset generalization, achieving a 25.35% lower EER on the FMFCC-A evaluation set compared with conventional approaches. These findings validate the effectiveness of the hybrid attention-convolution design in advancing spoofing detection robustness and domain adaptability.

In order to advance the analysis and mining of TCM(traditional Chinese medicine) text data and achieve intelligent extraction and processing of knowledge, the BIO(begin, inside, outside) sequence labeling method, the BiLSTM-CRF model, and manually defined rules were adopted to complete the knowledge extraction task. Utilizing the Py2neo library in Python 3.6 and the Neo4j database, a spleen and stomach disease knowledge graph was constructed based on Neo4j, and a TCM spleen and stomach disease named entity recognition system was developed using the Flask framework. The results show that the BiLSTM-CRF model achieves high performance and good generalization ability on the test set, with accuracy, precision, recall, and F₁ scores of 96.19%, 86.64%, 88.82%, and 87.71%, respectively. The constructed knowledge graph includes eight types of node labels, such as prescriptions or patent medicines, Chinese medicines, and clinical manifestations, as well as ten types of relationships. It supports the querying and discovery of nodes and relationships among Western medical diagnosis, TCM syndromes, and TCM treatment principles for spleen and stomach diseases. It is concluded that the BiLSTM-CRF model demonstrates excellent generalizability in named entity recognition of TCM spleen and stomach disease. It exhibits outstanding performance in handling complex text structures and domain-specific terminology, providing strong support for the research on knowledge extraction and knowledge graph construction in Traditional Chinese Medicine for spleen and stomach diseases.

Bolts are the key to the stable connection of high-altitude equipment, but they are prone to abnormalities such as loosening under the influence of various factors, threatening the safety of the equipment. Currently, bolt detection methods based on deep learning are faced with the problems of class imbalance and label missing. Existing deep-learning-based bolt detection methods suffer from class imbalance and missing labels. A HDWL(historical dynamic weighted loss) model based on semi-supervised pseudo-label learning was proposed. By dynamic weighted orthogonality and class-adaptive fair punishment, the model classification was evaluated with historical data. Adaptive punishment was introduced to prevent overfitting and focus more on hard-to-classify samples, boosting model performance. Experiments showed that the HDWL model achieved significantly higher accuracy than other methods, with advantages in minority-class training and feature focus.

Recently, SSVEP(steady-state visual evoked potential-based BCI(brain-computer interface) researches have achieved significant development. However, the practical application of BCIs are still limited by several factors, one of which is the visual stimulus source. Most SSVEP-BCI systems rely on monitors, which are not portable and thus restrict the practical use in daily life. VR glasses, as wearable and portable devices, can provide realistic and immersive stimulus sources, which do not rely on monitors. Thus, they offer significant potentials for BCI applications. The VR(virtual reality) technology was introduced to display VR-SSVEP visual stimuli in 3D environment and enables subjects to immersively engage in BCI. The performance of 3D and 2D visual stimuli based on VR-SSVEP were compared in this study. The experimental results demonstrate that the performance of 3D visual stimuli is better than that of 2D visual stimuli. The average classification accuracy of 3D stimuli reaches 90.10%, which is 7.08% higher than 2D stimuli. Additionally, a 2-second stimulation duration achieves an optimal information transfer rate. This study confirms that 3D visual stimuli can effectively enhance SSVEP recognition performance, which indicates a practical use of the system and provides a novel approach for applying VR devices to the SSVEP paradigm.

To address the issues of distortion and poor segmentation results in weld defect image segmentation, the crack and porosity welding defect images in the rim production process were taken as the research object. An improved particle swarm optimization algorithm based on simulated annealing is proposed for the three-threshold image segmentation of welding defects. First, a three-dimensional Otsu model is constructed using the grayscale value, average grayscale value, and median grayscale value of the image. Next, an adaptive inertia weight and asymmetric learning factor were introduced and integrated into the SA strategy to enhance the algorithm’s solving efficiency and ability to escape local optima. Finally, the SA-IPSO algorithm was used to optimize the three-dimensional Otsu model to obtain the optimal threshold and corresponding defect segmentation image. Various algorithms and models are employed to segment welding defect images. The results show that for crack and porosity defect images, the proposed improved algorithm outperforms the comparison algorithms in terms of peak signal-to-noise ratio and structural similarity evaluation metrics. The proposed method accelerates algorithm convergence while preventing distortion in segmentation results, thereby improving segmentation accuracy.

Microwave assisted cutter breaking has a strong application prospect, in order to deeply analyse the auxiliary effect of microwave irradiation on cutter breaking, the TBM scale cutter breaking test after microwave irradiation was carried out. Firstly, the rock was irradiated by microwave using different parameters, and the influence law of different microwave parameters on the surface temperature of the rock was studied. After the rock is back to room temperature, the damaged rock was taken as the basis to carry out the TBM scale cutter rotary rock breaking experiment, to study the influence of microwave irradiation time and power on the cutter thrust, cutter torque, rock ballast weight, cutter wear and the specific energy of rock breaking. The results show that: with the increase of microwave irradiation time and irradiation power, the rock surface temperature increases, the rock heating rate increases, the highest temperature of the rock surface is 172.6 ℃. The cutter thrust is fluctuating in the breaking rock, the rock below the cutter is crushed to powder, the cutter side produces block ballast. With the increase of microwave irradiation time and microwave power, the cutter thrust decreases, the disc torque decreases, the weight of the rock ballast increases, the amount of cutter abrasion decreases, and the specific energy of rock breaking decreases. The best microwave irradiation time should be more than 40 s and the microwave power should be more than 7 kW for the test of Chifeng basalt, which provides a certain experimental basis for the microwave-assisted cutter rock-breaking.

Relying on a deep foundation pit project of a metro station in a dual soil-rock stratum, the deformation patterns of the supporting structure and key sections of the foundation pit during different construction stages were obtained through on-site measured data. Based on the presence of adjacent buildings, a 3D numerical model for dual soil-rock deep foundation pits was established. After the excavation and dewatering construction of the deep foundation pit, the deformation and stress of the retaining structure on both the side adjacent to and away from the buildings, surface subsidence, and changes in the axial force of the supports were analyzed. On this basis, a sensitivity analysis was conducted on the influencing parameters of foundation pit deformation and stress, and the variation patterns were summarized and fitted. The study results indicate that the excavation of dual-element deep foundation pits in soil and rock exhibits significant spatial and temporal effects. In the time dimension, this is manifested as rapid development in the lateral displacement of pile bodies and surface settlement in soil layers, while the development rate slows in rock layers. In the spatial dimension, this is manifested as the corner effect of the pit. Changes in spacing and pile diameter essentially alter the overall stiffness of the retaining structure and the magnitude of external soil and water pressure borne by an individual pile. When the pile diameter is less than 1.0 m, the force sensitivity of deformation in the adjacent building side of the foundation pit significantly increases. The properties of rock and soil masses vary greatly at the soil-rock interface in a dual soil-rock formation, where over-excavation, support spacing, and changes in prestress are more pronounced. The research findings can provide valuable insights for similar soil-rock dual-element deep foundation pit engineering projects.

Pipeline leakage is a major cause of urban road collapse accidents. Understanding the evolution process and catastrophic mechanisms of road subsidence is crucial for preventing such safety incidents. Focusing on sewage pipelines in Beijing municipal roads, this study employs DEM-CFD(discrete element method-computational fluid dynamics) coupled flow-solid approach. Microscopic model parameters were calibrated based on laboratory experiments to simulate deformation and cavity evolution in sandy soil layers under various pipeline leakage locations and burial depths. Key parameters, including particle displacement, soil compactness, and medium flow, were analyzed during cavity formation. The results indicate that leakage at the top and middle of the pipeline leads to the formation of a funnel-shaped cavity as water and soil are lost. Without traffic load, the road surface exhibits negligible settlement. By analyzing particle displacement and compactness variations, the soil deformation was divided into stable, loose, and cavity zones, and an elliptical partition model was established for the loose zone. Based on the particle loss rate, the progressive failure process of the soil was classified into three stages: particle migration, rapid loss, and gradual convergence. In terms of cavity formation time, subsidence extent, particle loss rate, and total particle loss, leakage at the pipeline’s middle section yielded the highest values, followed by the top section, with the lowest at the bottom section. However, bottom leakage resulted in the largest loose zone. These findings provide theoretical support for detecting and identifying underground risks associated with urban road collapse disasters.

As a key equipment and a major source of energy consumption in a building, chiller plant, if it fails, it will not only affect the normal operation of the system, but also cause serious energy waste. In order to improve the reliability of chiller system operation. A multi-strategy IDBO(improved dung beetle optimization algorithm) combined with a HKELM(hybrid kernel extreme learning machine) fusion fault diagnosis model was constructed to achieve accurate diagnosis of early faults in chiller systems. The model firstly employs hybrid kernel functions to improve the learning ability and generalization of KELM(kernel-extreme learning machine). Secondly, Bernoulli mapping, adaptive inertia factor, and Levy flight fusion dynamic weight coefficients strategies were used to improve the DBO(dung beetle optimization) algorithm in order to balance the global exploration performance of the DBO algorithm. Finally, the effectiveness of the IDBO algorithm was verified by benchmark functions, and the HKELM hyperparameters are optimized using the IDBO algorithm to construct a data-driven model for early fault diagnosis of chiller units. Through relevant training simulations and experimental validation, the accuracy of the proposed IDBO-HKELM model for early fault diagnosis of chillers is improved to 99.71%, which is an obvious advantage over other algorithms.