Introduction: The Pelican Optimization Algorithm (POA) is a meta-heuristic optimization algorithm, distinguished by its excellent ability to balance global exploration and local exploitation. This unique advantage enables it to effectively tackle complex multi-objective and multi-constraint optimization problems in mechanical design, where traditional algorithms often struggle with precision and efficiency. Thus, exploring POA’s application value in this field is of great significance for advancing mechanical design optimization.

Methods: This study applies the POA to mechanical design parameter optimization, taking advantage of its bionic mechanism that simulates pelicans’ natural hunting behavior. To verify its performance, POA is employed to solve three mechanical design optimization tasks: minimizing the self-weight of tension/compression springs, reducing the volume of rolling bearings, and lowering the manufacturing cost of reducers, with parameter optimization conducted by leveraging POA’s balanced exploration–exploitation capability.

Results: Comparative experimental tests show that the mechanical design solutions obtained by POA outperform those generated by other conventional meta-heuristic algorithms. Specifically, in terms of core objective function values, including the self-weight of springs, volume of rolling bearings, and production cost of reducers, POA achieves more optimal results, demonstrating its superior optimization performance.

Conclusions: The application of POA effectively enhances the precision and efficiency of mechanical design parameter optimization. This study confirms the feasibility, superiority, and practical applicability of POA in solving engineering optimization problems, providing a reliable new optimization tool for related mechanical design scenarios.

The deployment of computer vision models on resource-constrained hardware, such as the ESP32 microcontroller, requires a critical balance between classification accuracy and memory footprint. This study investigates the performance of MobileNetV1 and MobileNetV3-Small architectures in scenarios characterized by limited data, aiming to identify the most efficient configuration for mechanical industrial tool classification. A dataset containing 106 images across five distinct classes was developed, utilizing an 80/20 train–test split with an additional 15% of training data reserved for validation. Both architectures were implemented using transfer learning with frozen ImageNet backbones and custom classification heads. MobileNetV1 was configured with a width multiplier of $\alpha = 0.25$ to aggressively reduce filters, while MobileNetV3-Small employed $\alpha = 1.0$ in "minimalistic" mode to exclude high-latency activation functions and attention modules. Both models were optimized using the Adam optimizer and categorical cross-entropy loss to ensure a controlled experimental comparison. Experimental results demonstrate that both architectures achieved equivalent performance, with a global accuracy of 91% and a weighted F1-score of 0.91. Confusion matrix analysis revealed that errors were primarily confined to visually similar classes, such as Allen keys and screwdrivers. However, a significant disparity emerged regarding model size: MobileNetV1 produced a 1.1 MB binary, whereas MobileNetV3-Small resulted in 2.1 MB, a nearly 90% increase in storage requirements without any gain in predictive performance. This research concludes that increased architectural complexity does not inherently translate to superior performance in small-data regimes. For memory-constrained devices like the ESP32, the scaled-down MobileNetV1 provides a superior cost–benefit ratio, maintaining high accuracy with a substantially smaller memory footprint. The findings highlight the necessity of prioritizing structural simplicity over architectural novelty when designing deep learning solutions for embedded AI applications.

The large-scale integration of renewable energy sources (RESs), electric vehicles (EVs), and battery energy storage systems (BESSs) has significantly reduced system inertia and intensified frequency–voltage coupling in modern interconnected power systems, thereby challenging the effectiveness of conventional secondary control strategies. To address these emerging issues, this paper proposes an intelligent, control-aware evolutionary multi-objective Neural Architecture Search (EMO–NAS) framework for coordinated Load Frequency Control (LFC) and Automatic Voltage Regulation (AVR) in renewable-dominated multi-area power systems. Unlike existing approaches that rely on fixed or heuristically selected controller structures, the proposed framework treats the controller architecture itself as an explicit decision variable and autonomously synthesizes task-specific control policies through multi-objective evolutionary optimization. The coordinated LFC–AVR problem is formulated by simultaneously minimizing frequency deviation, tie-line power oscillations, voltage deviation, rate of change of frequency (RoCoF), control effort, and robustness degradation, while satisfying practical operational constraints including generation rate limits, actuator bounds, and BESS state-of-charge restrictions. A structured NAS search space incorporating feedforward, recurrent, and temporal architectures is evaluated using closed-loop time-domain simulations under realistic disturbances, renewable intermittency, EV variability, and parameter uncertainty. Feasibility and stability are enforced through constraint-aware penalties and robust domain randomization. Comprehensive simulation studies on three- and four-area interconnected systems demonstrate that the proposed EMO–NAS controller achieves substantial performance improvements compared with optimally tuned fractional-order PID, robust sliding mode, and fixed-architecture neural controllers. Quantitatively, reductions of approximately 30–35% in frequency deviation, 25–35% in tie-line power oscillations, and up to 30% in RoCoF are achieved, while completely eliminating constraint violations. Robustness analysis under ±50% parameter uncertainty and Monte Carlo simulations further confirm superior stability, generalization, and scalability. These results establish architecture-level optimization as a powerful and systematic pathway for designing robust, coordinated secondary controllers in future low-inertia, renewable-dominated power systems.

Accurate induction motor eccentricity detection from phase currents is attractive for industrial monitoring because it can be implemented non-invasively using existing electrical measurements. However, developing robust neural classifiers is often limited by the scarcity of labeled fault data across operating conditions. This work proposes a compact convolutional neural network (CNN) for eccentricity-level classification, trained on a broad set of simulated current signals enhanced to better reflect measurement imperfections, and designed as a foundation for future transfer to real-machine recordings.

Three-phase stator currents were generated with an eccentricity simulation model and enriched using a residual-injection scheme motivated by frequency-domain inspection and correlation analysis, which indicate notable non-ideal components affecting signal consistency. The dataset spans five eccentricity levels (0.0–0.4, step 0.1), steady loads from 0 to 10 (step 2), and 1 Hz sinusoidal load profiles within 0–4, 4–6, and 6–10, for steady speeds of 1500, 1350, and 1200 rpm. From each case, 50 windows of 900 samples were extracted and reshaped into a 30×30×3 representation (three channels for phase currents). The CNN includes three convolutional feature-extraction blocks (convolution, batch normalization, ReLU, max pooling) and a classifier head with adaptive average pooling, dropout, and a fully connected layer.

The proposed network achieved approximately 96% test accuracy, with comparable validation accuracy and ~97% training accuracy across the considered conditions.

A lightweight CNN can accurately classify eccentricity levels using current-only inputs when trained on condition-diverse, residual-augmented simulation data. In future work, the trained model will serve as a pretraining baseline for transfer learning to laboratory measurements, enabling practical eccentricity detection on real motors.

Electric motor diagnostic systems dedicated to rolling bearing fault detection impose specific signal processing requirements to ensure high effectiveness in technical condition assessment. In the case of mechanical faults, the most critical requirements include relatively long measurement durations, stable operating conditions, and high measurement accuracy. However, such conditions are difficult to achieve in real industrial drive systems, where transient states and variable operating conditions are common. Consequently, there is an increasing demand for fast and reliable fault detection methods based on the shortest possible data records.

This study investigates the influence of the data acquisition system and operating conditions on the quality of fault-related information in the rolling bearing diagnostics of induction motors. The primary objective is to determine the minimum measurement duration that enables reliable fault detection while preserving essential diagnostic features.

Experimental investigations were conducted for various types and severities of rolling bearing damage under both steady-state and transient operating conditions. To enable fully automated classification, a shallow neural network was employed. Specifically, a classifier based on Kohonen Self-Organizing Maps (SOMs) was used, with input features extracted from measurement signals using envelope spectrum analysis and selected statistical indicators.

The results demonstrate that the quality of fault-related information strongly depends on both the measurement duration and the operating conditions of the motor. Reducing the data vector length significantly degrades diagnostic performance, particularly under dynamic operating conditions. The obtained results highlight the importance of appropriate measurement duration selection in the design of fast and practically applicable diagnostic systems for induction motor bearing fault detection.

This work evaluates how varying the methanol energy share affects efficiency, combustion behavior, and pollutant formation in a methanol–diesel-fueled Reactivity-Controlled Compression Ignition (RCCI) engine. A comprehensive experimental campaign was performed under both constant and dynamically varying operating conditions, covering engine speeds from 1400 to 2000 rpm and load levels ranging between 25% and full load. The methanol substitution rate (MSR) was systematically adjusted from 0% to 40% to assess its influence on combustion and emission performance. The findings demonstrate that a higher methanol contribution significantly improves brake thermal efficiency, with gains reaching approximately 14% at elevated load conditions. However, excessive methanol addition slightly increases cycle-to-cycle variations, indicating a modest reduction in combustion stability. Changes in MSR were also observed to strongly influence gaseous and particulate emissions. Compared with conventional diesel operation, increasing methanol fraction led to notable reductions in nitrogen oxides and particulate matter, while carbon monoxide levels exhibited sensitivity to combustion phasing and mixture reactivity. Analysis of the measured data revealed that NOx emissions declined by nearly 30–50% at higher methanol fractions, and particulate emissions remained considerably lower than those produced by diesel-only combustion. In addition, vibration-based virtual sensing combined with time–frequency domain analysis showed that methanol enrichment modifies heat release characteristics and pressure rise rates, which subsequently alters engine structural vibration responses and emission trends. Exergy assessment further indicated that a methanol substitution ratio of approximately 30% minimizes irreversibility losses, signifying superior energy utilization efficiency and the most thermodynamically favorable combustion condition within the tested range.

A comprehensive approach to diagnosing rolling bearing damage in permanent magnet synchronous motors (PMSMs) is an important research topic due to the growing use of PMSM drives in industrial systems. The proposed methodology is based on the analysis of diagnostic signals, including mechanical and electrical quantities, obtained under varying operating conditions. Experimental studies were conducted for different motor load levels, as well as for different settings of the current controller parameters in a field-oriented control system. This made it possible to assess the impact of operating conditions and control structure on the effectiveness of bearing damage detection. Particular attention was paid to the process of signal acquisition, preprocessing, and extraction of features characteristic of rolling bearing damage. The selected methods of time and frequency domain signal analysis were used to identify damage-sensitive symptoms associated with bearing component defects. The results obtained show that the effectiveness and sensitivity of individual diagnostic symptoms strongly depend on both the type of signal measured and the operating conditions of the drive system. A comparative evaluation of the extracted features is presented, emphasizing their usefulness for reliable damage detection and accurate condition assessment. The presented approach shows potential for the practical application of signals other than mechanical vibrations.

This paper presents the design and implementation of an energy monitoring system based on the SDM630 multifunction meter, developed for applications involving three-phase electrical machines. The main objective of the proposed system is to provide a practical solution for monitoring and analyzing the operating behavior of electrical machines in both industrial and laboratory settings. The system performs real-time measurement and continuous logging of key electrical parameters, including phase and line voltages, currents, active and reactive power, energy consumption, power factor, and frequency. Data communication between the SDM630 meter and the unit of monitoring is implemented using the Modbus RTU protocol, selected for its suitability and reliability for industrial environments. Measured data are stored locally and processed offline to enable further analysis. Statistical processing is applied to assess operating conditions, identify energy consumption trends, and evaluate the stability of machine performance under different loading conditions. Parameters such as mean values, fluctuations, and temporal trends are considered to support energy evaluation and basic diagnostic analysis. The system architecture was designed with flexibility in mind, allowing straightforward integration with supervisory systems and potential future extensions, including remote monitoring and data visualization platforms. Experimental tests were conducted on three-phase electrical machines operating under various load conditions levels. The experimental results demonstrate that the proposed system provides consistent and accurate measurements, confirming its suitability for continuous monitoring tasks. Owing to its straightforward design, low implementation cost, and reliance on commercially available components, the developed system offers an effective and accessible solution for energy management, performance assessment, and power quality monitoring in three-phase electrical machine applications.

This work presents an experimental comparison of speed sensor fault detection and classification methods in a vector-controlled permanent magnet synchronous motor (PMSM) drive. Three approaches are investigated: a model-based Sliding Mode Observer fault detector and compensator, a multilayer perceptron (MLP), and a convolutional neural network (CNN) fault classifiers. The study is entirely based on experimental results obtained on the dSPACE DS1103 Controller Board, ensuring a realistic and reproducible validation environment. The experiments cover a range of operating conditions (variable speed and load), allowing evaluation of each method’s detection accuracy, reliability, and robustness. Furthermore, the operation of the classifiers is based on a different type of speed estimator—the Model Reference Adaptive System (MRAS)—which enables not only fault classification but also fault compensation for each analyzed system.

The MLP and CNN approaches utilize data-driven techniques to classify faults, while the Sliding Mode Observer provides a model-based reference, enabling direct comparison between signal-based, shallow learning, and deep learning approaches. The findings reveal distinct performance differences, with each method showing particular strengths and limitations under the tested conditions. This comparison highlights the trade-offs between computational complexity, accuracy, and practical applicability, offering guidance for selecting appropriate diagnostic strategies for industrial PMSM drives and other applications.

Manual handling and packing of thin metal plates remains a labor-intensive operation in the automotive manufacturing sector, frequently requiring multiple operators and resulting in limited productivity, reduced process repeatability, and ergonomic constraints. This paper presents a preliminary virtual and experimental proof of concept for the automation of such a packing task using a delta robot, motivated by a representative industrial scenario involving automotive heat exchanger plates. The proposed study adopts an intentionally simplified problem formulation to support early-stage feasibility assessment and system design. In the considered scenario, the approximate initial position of each plate is assumed to be known at the moment it is released onto a tray, allowing the study to focus on the robotic packing stage rather than on part detection. A virtual robotic packing cell is developed using MATLAB-based simulation tools, with particular emphasis on workspace definition, packing layout design, and end-effector selection. The target placement positions inside the packing tray are fully defined in the robot coordinate system, enabling the analysis of reachability, packing density, and achievable cycle times without introducing additional complexity related to sensing or perception. A set of performance-oriented indicators is defined and evaluated in simulation, including workspace utilization, throughput, and packing efficiency. Based on the virtual results, a simplified experimental demonstrator is implemented to provide initial validation of the proposed concept, using representative plate geometries and a programmed packing sequence. This experimental stage is not intended as a full industrial validation, but rather as a functional verification of the feasibility of delta robot-based packing under controlled conditions. By deliberately limiting the system scope, the study establishes a baseline for robotic packing performance and provides a structured foundation for subsequent research, which will address more realistic industrial conditions such as imprecise plate positioning, surface contamination, vision-based perception, robot–sensor calibration, and scenarios involving multiple plates simultaneously.