As a coreauxiliary equipment in power plants, unplanned shutdowns of centrifugal pumps caused by vibration will result in huge economic losses to the plants. This paper aims to systematically elaborate on the generation mechanism, key influencing factors, and suppression methods of flow-induced vibration in centrifugal pumps. Pump body vibration is mainly derived from pressure pulsation generated by internal fluid flow, which acts on the pump body to induce vibration. The main influencing factors include unreasonable setting of pump body geometric parameters, unstable flow conditions, and cavitation. To reduce the intensity of flow-induced vibration, scholars mainly adopt numerical simulation, experimental testing, and a combination of both to formulate optimizationstrategies, minimizing pressure pulsation inside the pump by optimizing the design of pump body geometric parameters. Research shows that impeller parameter optimization can reduce the amplitude of pressure pulsation, increase pump efficiency by 1.6%~6.71%, and decrease vibration intensity by more than 30%; the multi-volute structure can maximize the reduction of radial thrust by 72% and improve efficiency by 6%~10%; the optimization method combining artificial intelligence/computational intelligence technology with CFD can increase pump operating efficiency by about 27% and shorten model training time by 180s. By sorting out the influence laws of pump body design and operating conditions on vibration, this paper summarizes the mainstream suppression technologies of flow-induced vibration, and clarifies that the application of multi-parameter collaborative optimization, advanced numerical simulation models (DES/LES), and adaptive intelligent algorithms is the key direction to reduce vibration and bridge the gap between theoretical and engineering applications in the future, providing theoretical support and technical reference for the stable operation and optimal design of centrifugal pumps.
Every built structure will endure a period of transformation over time, either due to natural deterioration or through intervention by renovation and preservation works. It is essential to record the building's current state by determining each component integrated with it, and this can be done by remodeling, referring to the dimensions of the building itself in the form of digital documentation. In common practice, the dimensions and data of the building are obtained by traditional measuring methods, where the process is time-consuming and may lead to errors. The objectives of these studies are to develop the workflow of built buildings' data documentation through 3D laser scanning with the aid of georeferencing to increase data accuracies. In the context of this study, the Student Centre located at Kampus Bandar, Universiti Tun Hussein Onn Malaysia, was chosen as the study area. The point clouds of the study area were documented and integrated with the georeferenced data, including three-dimensional coordinates (X, Y, and Z axis). As a result, it is ascertained that with the integration of georeferenced data on top of the point clouds, the outcome of the digital documentation will be more precise and accurate for future reference, specifically for design and construction works. The results demonstrate that incorporating georeferencing data into to the point cloud data significantly improves the spatial accuracy and reliability of the resulting as-built data. Such enhanced datasets provide a dependable foundation for engineering applications, including design refinement, construction verification, and detailed structural assessment.
The marine propeller in operation undergoes local defects on the surface and edges of the blades, thus causing vibrations of the propulsion system and a reduction of the propulsion efficiency of the propeller. These local defects develop gradually under the effect of cyclic stresses and lead to the failure of the marine propeller blade by fatigue. In this article, a unique non-destructive approach for structural defect identification in marine propeller blades is presented. The defect identification methods defined by modal strains, such as strain mode shapes changes, strain flexibility changes, and defect index, were used to find and identify single and multiple defects in the marine propeller blade's structure. The modal strain-related defect indicators were calculated from the strain mode shapes. The strain mode shapes of the marine propeller blade are tri-axial and are calculated along the x, y, and z directions of vibration. In this article, three defect types were considered, with different sizes along the length of the three guide curves starting from the blade base to the free end. The results proved that the proposed defect identification methods defined by modal strains can accurately locate single and multiple defects up to four appearing on the leading edge, the trailing edge, and the propeller blade surface. The magnitude of the defect index at nodes of defective elements is higher (order of 10-2) than the magnitude of strain mode shapes change (order of 10-3), and much higher than the magnitude of strain flexibility change, which is of the order of 10-9. The defect severity quantification on a marine propeller blade is feasible as the magnitudes of the three proposed defect indicators at the nodes of the defective elements are proportional to the severity of the defects.
Mobile robot localization is essential for autonomous navigation, allowing robots to accurately determine their positions within an environment. Traditional scan matching algorithms primarily only provide an area estimation of the robot's location without the exact coordinates within its environment. This paper presents a work on improving the scan matching localization algorithm using K-nearest neighbour (KNN), that is able to provide 2D coordinate information of the mobile robot position (x,y). The mobile robot with an RP lidar and high capability of a computer system is used in this study. Typical algorithms often only encompass input preparation, scan matching, and area classification stages. The proposed localization algorithm introduces a fourth stage aimed at determining the coordinates of the mobile robot within its environment. Three distance parameters (Euclidean, Mahalanobis, and Manhattan), are investigated to determine the optimal choice for KNN. The proposed improved algorithm employing Euclidean distance achieved an accuracy of 93% when classifying 100 non-reference test samples within a range of 4cm area. Furthermore, the mobile robot continuously maintains a coordinate determination accuracy of less than 3cm in 30 samples across all local maps. These findings hold promise for applications requiring precise mobile robot localization.
Nutritional adequacy during pregnancy requires mothers to follow a well-balanced, personalized diet plan supporting maternal and fetal well-being. However, formulating such plans is challenging because of the combinatorial aspect of food selection and competing nutritional needs. This study offers an integrated framework of algorithms for intelligent menu planning specific to pregnant women. To measure the net optimizing value of meal recommendation programs within set nutrition limits, this study evaluated three AI programs, such as Q-Learning, SARSA, and Monte Carlo Tree Search (MCTS). Each algorithm was optimized through hyperparameter tuning to maximize a fitness function based on nutrient sufficiency, distribution, and adherence to predefined limits. Q-learning showed the fastest adaptation and performed best at elevated learning rates. In contrast, SARSA exhibited stable, more consistent behavior over extended training periods and performed best at alpha = 0.4 and epsilon = 0.35. MCTS provided better performance than both reinforcement learning methods by achieving the highest fitness score through a tree-based search concerning optimal exploration weight and simulation count. These findings support the claim that each algorithm performs best in specific areas, such as Q-Learning excels at learning new information, SARSA is proficient with fixed policies, and MCTS can easily solve complex combinations of variables and meet limits. The comparative results highlight that the choice of an algorithm must be aligned with the system's goals concerning efficiency, consistency, or comprehensiveness. This study develops AI-based dietary assistance systems and demonstrates the possibilities of customized algorithms designed for effective nutrition planning in maternal healthcare.
This paper presents the edge deployment of a behavior-based Smart Home Automation System (SHAS) designed for proactive electrical overload prevention. The term behavior-based refers to modeling appliance control decisions using historical usage patterns and contextual sensor data rather than predefined deterministic rules. The proposed system integrates IoT sensors and a Long Short-Term Memory (LSTM) model running on a Raspberry Pi 4 edge platform. The dataset was collected continuously over one week with a one-minute sampling interval, resulting in a real-world time-series dataset representing household energy usage behavior. Model performance was evaluated using standard classification metrics, including accuracy, precision, recall, and F1-score, along with system-level evaluation using cumulative overload duration. Experimental results show a significant 59.6% reduction in total overload duration compared to baseline operation without adaptive control. This demonstrates the effectiveness of the proposed approach compared to conventional rule-based or non-adaptive systems. The system also maintains low-latency communication and stable resource utilization. Overall, this work demonstrates the feasibility of deploying deep learning-based energy management entirely on edge devices, providing improved responsiveness, reduced internet dependency, and enhanced data privacy in smart home environments.
Power Quality (PQ) has become a critical requirement in modern power systems due to its impact on industrial automation, computer-based technologies, and controlled electrical systems. Accurate identification and mitigation of electrical disturbances are essential for improving system reliability and efficient power delivery. However, conventional approaches often face limitations in optimizing power flow at the end-user level. In this paper, a superior controller to a Distributed Static Compensator (DSTATCOM) was developed with intentions of optimizing power quality in a traditional bus system. A baseline DSTATCOM model of a basic bus setup had first been set up, followed by gathering real-time data over a range of disturbance conditions to power quality. This dataset was then trained in a Feed Forward Neural Network (FFNN) controller, creating a DSTATCOM pulse signal to be created in the analysis of bus voltage measurements. The proposed controller efficiency was estimated using case studies based on IEEE 33 and 13 bus systems. Performance validation was conducted at a number of disturbances such as interruption, swell, harmonic content, and sag. In the case of the IEEE 13-bus system, the model proposed meets the total harmonic distortion (THD) of 2.93, 1.81, and 0.02, under sag, swell, and interruption conditions, respectively. In the same manner, in the case of the IEEE 33-bus system, the THD of sag, swell and interruption are 0.58% 0.26% and 0.33%. The results confirm that the proposed FFNN-based DSTATCOM controller effectively mitigates a wide range of power quality disturbances while enhancing system stability and reliability.
Fast Energy harvesters that combine multiple conversion mechanisms, such as piezoelectric and electromagnetic, offer an improved solution for powering small, portable devices. These hybrid systems exploit the complementary strengths of each method, where piezoelectric materials excel at generating high voltages in response to mechanical stress. At the same time, EM harvesters are more effective at low frequencies, addressing the limitations of piezoelectric devices. However, despite overcoming these individual shortcomings, hybrid harvesters still face the challenge of generating sufficient power output under low-frequency, random excitations commonly found in real-world environments. This study addresses this limitation by integrating Fast Fourier Transform (FFT) analysis to identify dominant frequencies and optimise system performance. Additionally, signal rectification methods, including the use of a voltage doubler circuit, were explored to enhance the energy conversion efficiency. Experimental results show that under random excitation frequencies ranging from 1 Hz to 4 Hz, the hybrid harvester consistently generated stable peak voltages, with the voltage doubler achieving a significant improvement, producing 3.8 V compared to the 1.5 V generated by the full bridge circuit. This work demonstrates that hybrid energy harvesting systems can provide efficient power generation under unpredictable, low-frequency conditions when coupled with optimized signal processing techniques. The findings contribute to the advancement of energy harvesting technologies, offering a sustainable and maintenance-free power source for low-power electronics in dynamic and mobile environments.
Convolutional Neural Networks (CNNs) have significantly advanced object detection recently. However, achieving the right balance between model size and detection accuracy is a persistent challenge, particularly for face detection in resource-constrained environments. While large models provide high accuracy, their computational demands make them impractical for real-time or embedded applications. Conversely, lightweight models are more efficient but often sacrifice accuracy. Existing lightweight YOLO variants trade accuracy for efficiency, particularly in detecting small or occluded faces. To address this, we proposed GCB-YOLO, a lightweight face detection model that balances compactness and performance. The design incorporates the Ghost Module (GM) to reduce model size, the Convolutional Block Attention Module (CBAM) to enhance feature representation, and the SiLU activation function to improve detection accuracy. Our evaluation on the Wider Face dataset demonstrates the effectiveness of GCB-YOLO. On the medium subset, it achieved 79.6% mAP50 and an 81.04% F1 score, surpassing other state-of-the-art lightweight models. The model maintained strong performance on the challenging hard subset with 60.5% mAP50, and a 65.37% F1 score. For the efficiency metrics, the parameter count was reduced by 49.44% (from 6.015M to 3.041M), and CPU inference speed improved from 4.265 FPS to 5.130 FPS. These results highlight GCB-YOLO's suitability for real-time face detection on resource-constrained devices.
The optimization and unification of cable cross-sections in urban distribution electrical networks (DENs) is an important technoeconomic problem, particularly for 10 kV medium-voltage (MV) networks where construction, installation, and operational costs are significant. This study aims to develop an integrated criteria-based techno-economic optimization model for determining economically justified network parameters while considering cable cross-section unification. The proposed model jointly optimizes the number of outgoing feeders, feeder cable cross-sections, and the number of standardized cable cross-sections within a 10 kV urban distribution network supplied from a common power source (PS). The model is based on topological representations of urban distribution network sections and incorporates capital expenditures, installation costs, operating costs, and electric energy losses. The optimization is performed using criteria-based programming, which allows analytical expressions to be obtained without numerous variant-based calculations. The results show that cable cross-section unification is economically justified when it is optimized together with other network parameters. For urban areas with load density 6 >= 10 MVA/km2, the use of one or two standardized cable cross-sections is optimal, whereas for lower load densities 6<10 MVA/km2, two or three standardized cross-sections provide a more suitable solution. Comparative evaluation indicates that the proposed approach can reduce cable-type diversity, material consumption, and operational complexity compared with conventional non-unified design strategies, while maintaining thermal and voltage-drop constraints. Stability and sensitivity analyses confirm that the optimized unification parameter remains sufficiently robust under moderate variations in techno-economic input data. The proposed framework can therefore support preliminary planning, standardization-oriented design, and long-term modernization of 10 kV urban DENs.
