Sustainable Energy and Artificial Intelligence

Abstract In this paper, we investigate factors affecting the performance of breast cancer histopathological image classification using the BreakHis dataset and MobileNetV3-Large as the backbone feature extractor.
First, we examine two sources of data leakage: the presence of multi-label patients and image-level dataset partitioning. We evaluate three experimental settings: (a) multi-label patients with image-level partitioning, (b) removal of the duplicate image copies associated with the multi-label patient while maintaining image-level partitioning, and (c) leakage-free patient-wise partitioning. Compared with setting (a), setting (b) resulted in a performance decline of 4-12 percentage points across evaluation metrics, while setting (c) led to substantially larger declines, with the largest decrease observed in specificity. These findings demonstrate that data leakage can substantially inflate performance estimates.
Second, we evaluate several transfer-learning strategies and observe that fine-tuning the middle and late layers of the ImageNet-pretrained network provides the best overall trade-off among the evaluated metrics, achieving an image-level accuracy of 0.60, an F1-score of 0.72, and a sensitivity of 0.80.
Finally, we repeat the leakage-free experiments over ten independent runs using different random seeds to assess the stability and statistical robustness of the results and compare the original GAP-based classification head with a Flatten-based alternative. The mean, standard deviation, 95% confidence interval, and statistical significance of the results were evaluated using performance metrics. Although the Flatten-based classifier achieved higher mean image accuracy, mean patient accuracy, mean sensitivity, and mean F1-score, the differences between the two classification heads were not statistically significant according to the paired Wilcoxon signed-rank test.

Abstract Power quality has become a serious problem as wind power has grown more integrated into the power grid. Power quality includes issues like flicker and harmonics. Wind characteristics such as wind speed, tower shadow and wind shear effects, and grid conditions such as short circuit capacity ratio and grid impedance angle, and maximum power point tracking type are just some of the factors that influence flicker emission and power oscillations of grid-connected wind turbines. Doubly fed induction generators (DFIG)s have several advantages over other wind power generators such as low equipment price and flexibility. However, DFIG is very sensitive to the lack of voltage at the generator terminal. Because the voltage drop induces large excitation voltage in the rotor windings, and high transient current through the rotor may occur which increase the dc interface voltage in the electronic power converter, resulting in damage to the electronic power converter. In this paper, performance of two maximum power point tracking techniques optimal torque control (OTC) and tip speed ratio (TSR) control in fluctuations emission is examined after the ideal MPPT technique and grid conditions are identified, and fluctuations caused by the tower shadow are then reduced by adding filter. The aerodynamic, mechanical, and electrical aspects of a grid-connected wind farm are simulated using MATLAB/SIMULINK. The simulation results are shown, which clearly illustrate that the proposed technique is effective for decreasing power fluctuations.

Abstract This study presents a data-driven Deep Reinforcement Learning (DRL) framework to minimize power losses and mitigate voltage fluctuations in distribution networks with high photovoltaic (PV) penetration, without curtailing active power generation from PV systems. The voltage regulation problem is formulated as a Markov Decision Process (MDP) with a multi-objective reward function, and two advanced DRL algorithms—Soft Actor-Critic (SAC) and Trust Region Policy Optimization (TRPO)—are trained and evaluated on the IEEE 37-bus and 123-bus distribution test systems. Using real-world solar radiation, temperature, and residential load data for two representative days, the DRL agents learn optimal coordinated Volt-Var control policies for multiple smart PV inverters through continuous interaction with a power flow environment simulated in OpenDSS. Results demonstrate that both SAC and TRPO effectively regulate voltage profiles and coordinate reactive power support while preserving full PV active power output, achieving substantial reductions in network energy losses compared to no-control and conventional Volt-Var droop methods, all while maintaining voltages within standard limits. SAC exhibits faster convergence owing to higher sample efficiency, whereas TRPO produces marginally lower total losses. This work introduces a model-free, multi-tasking DRL approach that enables scalable and adaptive coordinated control across distributed inverters, offering a practical solution for modern active distribution grids facing dynamic PV and load variations.

Abstract Surface defects on crankshaft journals, such as micro-scratches and pitting, can significantly degrade engine performance and initiate catastrophic failure. This study presents a real-time, vision-based inspection system for detecting surface defects on stationary crankshaft journals under severe data scarcity constraints. A high-resolution dataset of real journal surfaces was acquired using a controlled optical setup to minimize specular reflections. Pixel-level defect localization was performed using a U-Net semantic segmentation model with a MobileNet backbone, trained with Categorical Focal Loss and advanced data augmentation to address class imbalance and limited samples. The MobileNet-based model achieved a classification F1-score of 88.9%, demonstrating effective learning without overfitting. For segmentation, the system attained a mean IoU of 71.5% without augmentation, improving to 81.7% with augmentation. The optimized model was deployed on an NVIDIA Jetson Nano at a bearing assembly station, validating its suitability for real-time, in-line industrial inspection. The proposed framework provides an efficient and scalable solution for automated quality control, advancing zero-defect manufacturing of critical engine components.

Abstract The transition to clean energy demands advanced thermal energy storage (TES) solutions, especially for high-temperature applications like concentrated solar power (CSP) and industrial processes. Molten salt nanofluids—formed by dispersing nanoparticles in molten salts—offer a promising pathway to enhance thermal properties while maintaining high thermal stability and cost-effectiveness. This review summarizes the recent progress in the development, properties, preparation, and potential applications of these materials. Key focus areas include enhancements in specific heat capacity and thermal conductivity, which are critical for efficient heat storage and transfer. Notably, experimental studies report up to 100% increases in specific heat—defying classical predictions—possibly due to interfacial nanolayers, ionic rearrangement, or secondary nanostructures. Thermal conductivity improvements vary depending on nanoparticle type, morphology, and dispersion quality. The review also covers common base salts (nitrates, carbonates, chlorides) and a wide range of nanoparticle additives. Preparation methods such as ultrasonication and in-situ synthesis are discussed, along with challenges related to nanoparticle agglomeration, sedimentation, and long-term stability. Viscosity, corrosion behavior, and thermal cycling stability are also examined, as they critically affect system efficiency, pumping power, and material compatibility. Molten salt nanofluids hold strong potential for CSP, geothermal energy, enhanced oil recovery, and next-generation nuclear systems. However, commercialization is hindered by uncertainties in scalability, lifecycle impacts, and regulatory readiness. The review highlights the need for standardized methodologies, cross-disciplinary collaboration, and integrated performance-sustainability assessments to advance these materials toward practical deployment.

Abstract The frequency stability is a crucial aspect of an islanded microgrid, especially considering the presence of RES with low inertia. These RES, such as wind turbines and photovoltaic systems, pose a potential threat to the frequency stability of the microgrid. To address this challenge, the concept of VIC has been introduced in islanded microgrids. This paper investigates the application of VIC not only to the ESS but also to the WT and PVS. The proposed method aims to enhance the frequency stability of the microgrid. The results of this study compare the performance of the proposed method, which includes VIC for PVS, WT, and ESS, with other scenarios. These scenarios include The VIC for PVS and ESS, VIC for ESS only, and a method without VIC. The simulation results, obtained using MATLAB software, demonstrate that the proposed method significantly improves the frequency stability of the microgrid under load disturbances and disturbances originating from RES. Moreover, the proposed method exhibits robustness in the face of uncertainties associated with microgrid parameters.

Abstract Creating a formal common language, beyond the ambiguities of natural languages, between different stakeholders, from analysts to test engineers, is one of the key goals of software modeling. Although notations are standard in software modeling languages such as UML, junior engineers’ interpretation of models varies. Model interpretation in the presence of experienced people increases the learning rate for junior engineers. One of the potentials of large language models is the ability to interpret images and models. This research aims to use large language models as an assistant to interpret UML models to increase junior engineers’ learning rate and understanding of the software models. We conducted an evaluation study to examine how helpful an LLM can be to help interpret the software models. Although large language models are still not very accurate in interpreting UML models, the experiment’s results showed that students’ learning rates increased by LLMs as model interpretation assistants. In other words, the large language model worked well as a teaching assistant. The detailed results of this exploratory study are reported in this paper.

Abstract Electric motors are vital in energy systems for converting energy efficiently. Their reliability ensures consistent performance, making them indispensable in renewable energy applications and industrial processes. Bearings are crucial for motor operation, yet they are vulnerable to faults that can impair performance and reduce lifespan. Conventional fault detection methods require high sampling rate vibration data and expensive sensors. This paper presents a cost-effective solution by introducing a hybrid model combining Multiscale Convolutional Neural Networks (MSCNN) with Long Short-Term Memory (LSTM) networks for bearing fault diagnosis using low sampling rate data. We developed dedicated IoT-based hardware and a web server for real-time monitoring. Our approach leverages MSCNN for spatial feature extraction and LSTM for temporal pattern recognition, achieving high diagnostic accuracy with lower resolution data. Experimental results show that our MSCNN-LSTM model provides diagnostic accuracy comparable to or surpassing that of high sampling rate methods, offering a robust and economical solution for bearing fault detection.

Abstract This paper presents an insulin dosing policy for individuals with Type 1 Diabetes (T1D), utilizing Linear Matrix Inequality (LMI) techniques in combination with the TakagiSugeno (TS) fuzzy approximator. The primary goal is to regulate blood glucose levels by employing robust control strategies for the nonlinear dynamics of the glucose-insulin system, which are described using the Bergman Minimal Model. The proposed approach systematically incorporates insulin pump constraints, such as maximum insulin delivery rates, to ensure practical applicability in real-world scenarios. Simulation results demonstrate that the proposed controller maintains blood glucose levels within a safe range for over 84% of the time, with average glucose levels reduced to as low as 95mg/dL under the least restrictive input constraints. Furthermore, the controller effectively mitigates meal-induced disturbances while minimizing hypoglycemia risks, demonstrating its robustness under varying parameter uncertainties. This research highlights the potential of the proposed method for use in closed-loop insulin delivery systems, offering a promising solution for personalized and adaptive diabetes management.

Abstract In recent years, due to the energy crisis, the attention of researchers and energy industry experts has been drawn to the use of renewable energy sources. Among renewable energy sources, hydroelectricity is of great importance due to its major advantages, such as its widespread distribution on the earth's surface. Also Studies have shown that hydrokinetic energy can be a suitable alternative to fossil fuels. Savonius turbines are one type of turbine that, when combined with hydrokinetic systems, will produce clean and accessible energy. The Savonius vertical wind turbine must be started non-automatically. Additionally in savonius wind turbine slow speed reduces productivity and increases costs. When this turbine To be used in a flowing water Its characteristics will change and need to be optimized. Accordingly, numerous studies have been conducted on the optimization of these turbines. Accordingly, this article will review and investigate published articles in In this field. It also describes how each of the effective parameters affects the performance of these systems.

Abstract This paper presents an approach for automated unsupervised video summarization, that means, nothing more than video is needed to train the model. The goal is to extract a sequence of frames from an input video and assign each frame a score between 0 and 1. By doing so, we can select a subset of the most informative and diverse shots to make a summarized video. We build upon the foundation of SUM-GAN, particularly SUM-GAN-SLA, which utilize Generative Adversarial Networks to compare and distinguish between the original video and its regenerated counterpart. A key contribution of our work lies in the novel biLSTM-based self-attention network that we introduce to handle the crucial scoring layer of our model. We adjusted several aspects of the model, particularly in the loss functions and learning steps, to enhance the training process and achieve superior performance compared to state-of-the-art unsupervised and even supervised methods. To ensure a fair comparison, we evaluate our proposed model using two widely used datasets: SumMe and TVSum. The experimental results highlight the effectiveness of our proposed approach in automated unsupervised video summarization, achieving a 1.2% improvement over the best-performing methods' average F-score on SumMe and TVSum datasets. Additionally, our method ranks second among state-of-the-art unsupervised methods on each dataset. Notably, the top-performing methods exhibited inconsistent results across datasets, underscoring the broader applicability of our approach to diverse types of videos. Furthermore, our method demonstrates competitive performance compared to supervised approaches, with the best supervised method surpassing our results by only 0.75%.

Abstract Frequency stability is the fundamental problem of keeping the PS's frequency within a reasonable range. The PS's stability would be threatened by frequency variation brought on by load disruptions and uncertainty pertaining to PS characteristics. The PS's LFC system is crucial for controlling and maintaining frequency stability. This work uses the adaptive Fuzzy-FOPID (FFOPID) controller in the LFC structure to address uncertainties and disturbances while improving frequency stability in a two-area PS (TAPS). Additionally, RL has been used to enhance the adaptive FFOPID controller's performance. The adaptive FFOPID controller's performance against the uncertainties and disruptions of the TAPS has been enhanced by RL, which has also increased the control system's environmental adaptability. In a number of scenarios, the suggested control (FFOPID-RL) method has been compared to FFOPID, PDSMC, and SMC control methods. The findings indicate that the control method is better at lowering frequency deviations, settling times, and damping frequency oscillations associated with the TAPS