Zimbabwe's national power generation sector heavily depends on two primary sources: hydropower and coal-fired thermal power stations. The country, with a total installed capacity of approximately 1,700 MW against a demand of 5,000 MW, faces persistent power shortages, leading to imports and frequent blackouts. These two sources present distinct operational characteristics, environmental implications, and resilience to climate and economic pressures. This paper presents a comparative analysis of hydropower and thermal-fired plants in Zimbabwe’s national grid, focusing on their current status, challenges, and future prospects. The study evaluates generation capacity, reliability, cost structure, environmental impact, and long-term sustainability under climate variability. The research will use data from the Zimbabwe Power Company, Zambezi River Authority, policy documents, and relevant government ministries to assess the performance of the power stations. The methodology involves techno-economic and environmental performance assessments. The key indicators to be examined include average annual generation output, operational efficiency, carbon emissions, fuel availability, vulnerability to climate change, and maintenance downtime. Expected results include a clearer understanding of the relative strengths and weaknesses of hydropower and thermal power in Zimbabwe's energy system, including their suitability for long-term sustainability and climate resilience. The study also anticipates identifying policy and investment pathways that support a more diversified, reliable, and low-carbon electricity mix. The water–energy–climate nexus approach will serve as the analytical framework to understand the research's interdependencies, trade-offs, and policy gaps. Finally, this paper aims to contribute to ongoing national and regional discussions on energy security and infrastructure modernization in line with sustainable development goals.

Humanoid robots equipped with emotion-sensitive artificial intelligence (AI) are set to redefine human–robot interactions by enabling robots to understand, adapt, and respond to emotional cues in real time. This research introduces a neuromorphic AI-driven electronic skin (e-skin) designed to mimic human somatosensory functions, enhancing the social cognition of humanoid robots. Traditional tactile sensors in robotics primarily focus on force and pressure detection, lacking the ability to interpret nuanced emotional signals. To bridge this gap, we propose a novel multimodal e-skin architecture integrating flexible high-resolution pressure sensors, temperature sensors, and electrostatic detectors. These sensors capture complex tactile feedback, which is processed using neuromorphic computing models, specifically spiking neural networks (SNNs), a bio-inspired AI model optimized for real-time sensory data interpretation. Unlike conventional deep learning models, SNNs offer low-power computation, rapid adaptation, and real-time emotion recognition, making them ideal for embedded robotic systems. The AI-driven e-skin enhances context-aware decision making, allowing humanoid robots to distinguish between stress, comfort, and affection through touch patterns and environmental stimuli. To ensure fast response times and energy efficiency, the system incorporates edge AI processing, reducing reliance on cloud-based computation while maintaining low-latency interaction. The proposed framework is validated through experimental tests, demonstrating enhanced emotion perception accuracy and adaptability in human–robot collaboration, healthcare assistance, and interactive robotics. By integrating neuromorphic AI with advanced tactile sensing, this study paves the way for the next generation of socially intelligent humanoid robots, fostering a seamless blend between artificial intelligence and human emotional intelligence.

The quality of computational meshes plays an important role in the accuracy and efficiency of simulations, particularly in finite element analysis frameworks. Traditional smoothing techniques for unstructured meshes, such as Laplacian and optimization-based methods, typically employ Gauss–Seidel-style iterative schemes, where interior nodes are updated sequentially over multiple passes. While effective at capturing local, high-frequency changes, these methods often suffer from high computational costs and slow convergence. Recent advances in machine learning, particularly neural networks, offer a promising alternative. Existing neural-network-based mesh smoothing approaches commonly rely on separate models for each valence configuration, which limits their generalizability across diverse mesh topologies.

This work proposes a valence-aware, feedforward neural network architecture that learns to surrogate traditional quality metric optimization processes. By explicitly encoding the local valence degree of mesh umbrellas, the network can operate across varying topological configurations within a single model. The method specifically targets improvement of the weighted inverse-Jacobian quality metric for triangular elements. A fully connected network is trained on synthetic unstructured meshes paired with optimal node placements, which are derived from conventional optimization routines. Numerical experiments demonstrate that the proposed approach provides a faster alternative to traditional smoothing and optimization methods while achieving high-quality meshes. The improvement in the quality/time ratio underscores the potential of neural networks to address complex, non-linear relationships in mesh optimization tasks.

Extended Reality (XR) is an umbrella term that encompasses Virtual Reality (VR), Augmented Reality (AR), and Mixed Reality (MR), each offering unique ways to blend the digital and physical worlds. VR immerses users entirely in computer-generated environments, isolating them from the real world. AR overlays digital content onto the physical environment, while MR allows for more dynamic interaction between real and virtual elements. XR pushes these boundaries further by engaging multiple senses, enabling users not only to see but also to feel virtual objects — perceiving their weight, texture, and even temperature — thereby enhancing realism and immersion.

Artificial Intelligence (AI) plays a pivotal role in advancing XR technologies. It enables semantic enrichment of virtual scenes, allowing systems to understand and contextualize environments. AI can reconstruct and interpret 3D spaces, correct and enhance data during 3D digitization, and facilitate the creation of virtual worlds based on real-world inputs. Moreover, AI supports seamless interaction between physical and virtual realms, making the user experience more intuitive and responsive. Despite these advancements, challenges remain — including hardware limitations, latency, and the complexity of real-time data processing. In this context, we will explore both the current limitations and the vast opportunities that AI brings to the evolution of Extended Reality.

In the field of assistive technology (AT), individuals with visual impairments (VI) and speech impairments (SI) often encounter significant barriers to effective daily communication and social participation. This work presents the design and evaluation of an AI-powered integrated communication system aimed at bridging these gaps. The proposed solution combines multimodal interfaces that include voice commands, with real-time text and speech conversion to enhance user interaction. Adopting a user-centered design methodology, the system was iteratively refined through usability testing conducted in both controlled environments and real-world contexts. The results demonstrated notable improvements, with average task completion times decreasing from 45 to 28 seconds and communication success rates increasing from 76% to 91%. Additionally, user feedback emphasized clearer interactions, improved adaptability across contexts, and reduced frustration during use. Despite remaining challenges related to device compatibility, latency, and cost, the system demonstrated practical feasibility and significant value in supporting daily communication for users with VI and SI. This research provides a strong foundation for future developments, including multilingual capabilities, broader device support, and advanced AI features (e.g., predictive text and emotion recognition) to increase user autonomy and inclusion. Furthermore, the integration of customizable settings allows users to tailor the system according to their specific needs and preferences, enhancing accessibility and personal comfort. Continuous updates and machine learning algorithms ensure the system adapts dynamically to user behavior, improving efficiency over time. The scalability of the system suggests potential application beyond individual use, including educational and workplace environments, thereby promoting inclusivity on a larger scale.

In order to maintain public health, food security, and livestock productivity, animal health monitoring is essential. Traditional illness detection techniques frequently depend on laboratory testing and manual observation, which are expensive, time-consuming, and prone to causing delays in early intervention. Humans can easily communicate and share a problem; however, when it comes to animals, they cannot communicate or share their discomfort. Our objective in this paper is to detect and predict various health risks in different animals with high accuracy, thereby promoting a healthy lifestyle and minimizing the risk of death. Additionally, our model facilitates early intervention, contributing to reduced mortality rates. We used machine learning algorithms like random forest, SVM, LoGR, DT, Naive Bayes, and KNN, which were trained and evaluated for baseline performance. Apart from this, we also used the DL models ANN and CNN to capture complex nonlinear patterns and high-dimensional feature interactions. The performance metrics were evaluated to determine which model performed well. The proposed ML and DL models achieved high classification performance, with the random forest and CNN models outperforming the others. We measured the accuracy of the CNN model to be 94.8%, with a precision of 93.2%, recall of 95.1%, and F1 score of 94.1%. This means that the CNN has one of the strongest predictive capabilities for identifying early health risk across multiple animal species. We also found that the AUC-ROC score was 0.96, which indicates that our model perfectly classifies healthy and diseased cases for early diagnosis and intervention in animal healthcare.

In this study, several well-established convolutional neural network (CNN) architectures were employed to enable the automatic and reliable interpretation of Pile Integrity Test (PIT) data, with a particular emphasis on integrating soil characteristics as auxiliary input features. While previous studies have primarily focused on analyzing signal data alone, the significant impact of subsurface soil conditions on wave propagation and signal interpretation has been widely recognized by geotechnical experts. However, this critical aspect has often been overlooked in data-driven approaches. To address this limitation, we collected PIT data from 278 foundation piles constructed across sites with diverse geotechnical profiles. Each data sample was augmented with relevant soil-related features, including soil classification, stiffness parameters, and localized stratigraphic information. Among the tested CNN models, the highest classification accuracy achieved was 95%. Importantly, all data used in the study were obtained from real-world PIT measurements, as opposed to synthetic reflectograms commonly used in earlier research, thereby enhancing the practical relevance and generalizability of the results. The inclusion of soil characteristics was found to substantially improve model performance, increasing accuracy from 90% (signal-only input) to 95% when soil features were incorporated. This study represents the first comprehensive effort to explicitly include soil influence in PIT data analysis using deep learning and offers a novel contribution to AI-powered decision-support systems in structural and geotechnical engineering. The proposed soil-aware approach opens up new opportunities for more accurate and context-sensitive defect detection in pile foundations.

This study investigated the effect of varying proportions of rapeseed, avocado, and walnut oils on the oxidative stability of ternary lipid blends. A centroid simplex mixture design was applied, with oil fractions constrained to 100%. Oxidative quality was assessed using peroxide value (PV), acid value (AV), anisidine value (AnV), and the total oxidation index (Totox). Additionally, natural pigments, carotenoids, and chlorophylls were analyzed as response variables.

The findings showed that increasing walnut oil content above 7% markedly raised Totox, indicating enhanced susceptibility to oxidative degradation. In contrast, blends containing at least 72% rapeseed oil and 21–22% avocado oil, with a maximum of 6% walnut oil, demonstrated the lowest Totox values, reflecting the highest oxidative stability. This suggests a synergistic interaction between rapeseed and avocado oils, reducing both primary (PV) and secondary (AnV) oxidation.

The inclusion of carotenoids and chlorophylls as output variables highlighted their role as natural antioxidants, though their stability was strongly influenced by oil composition. Overall, the study confirms that mixture design methodology provides an effective framework for optimizing lipid systems and identifying interactions between components.

These results emphasize the potential of carefully balanced oil blends to achieve improved oxidative resistance and pigment stability. Such formulations may contribute to the development of functional fats and oils with superior nutritional quality and extended shelf life, supporting their application in the food industry.

This study examines the utilization of deep learning and transfer learning models for classifying photos of Indian cuisine. Indian cuisine, characterized by its extensive diversity and intricate presentation, poses considerable hurdles in food recognition owing to changes in ingredients, texture, and visual aesthetics. To tackle these challenges, we utilized a bespoke Convolutional Neural Network (CNN) and harnessed cutting-edge transfer learning models such as DenseNet121, InceptionV3, MobileNet, VGG16, and Xception. The research employed a varied dataset comprising 13 food categories and executed preprocessing techniques like HSV conversion, noise reduction, and edge identification to improve image quality. Metrics for performance evaluation, including accuracy, precision, recall, and F1-score, were employed to assess model efficacy. The CNN model demonstrated mediocre performance, revealing overfitting concerns due to a substantial disparity between training and validation accuracy. In contrast, transfer learning models, particularly DenseNet121, InceptionV3, and Xception, exhibited enhanced generalization ability, each attaining above 92% accuracy across all criteria. MobileNet and VGG16 produced reliable outcomes with marginally reduced performance. The results highlight the efficacy of transfer learning in food image classification and indicate that fine-tuned, pre-trained models markedly improve classification accuracy. This research advances the creation of intelligent food recognition systems applicable in dietary monitoring, nutrition tracking, and health management.

Escalating cloud adoption has multiplied organisations’ digital footprints—and their exposure to credential abuse, misconfiguration, and ransomware. This study examines how artificial intelligence (AI) analytics embedded in next‑generation Security Information and Event Management (SIEM) and Extended Detection and Response (XDR) platforms can fortify cloud defences while relieving Security Operations Centre (SOC) fatigue.

An extensive literature review was combined with an empirical evaluation in a production‑like enterprise cloud environment that fused a modern, data‑lake‑based SIEM, a vendor‑agnostic XDR layer, and a generative‑AI assistant orchestrating automated playbooks. Three realistic attack chains—phishing‑led account takeover, multi‑stage ransomware, and shadow‑IT data exfiltration—were replayed. Key metrics captured were mean time‑to‑detect/‑respond (MTTD/MTTR), incident‑correlation precision, and false‑positive rate.

AI‑assisted correlation collapsed hundreds of raw alerts into single contextual incidents, cutting analyst triage time by 96 %. Behaviour‑profiling models in the XDR layer reduced false positives by 89 %, while automated, AI‑guided playbooks contained live ransomware in under five minutes—an 18‑fold improvement over manual response. Overall, AI integration shortened MTTD and MTTR from hours to minutes across all scenarios.

The findings demonstrate that AI‑enabled SIEM/XDR can transform cloud security from reactive monitoring to proactive, autonomous defence, simultaneously boosting protection and SOC efficiency. Future work will explore reinforcement‑learning agents for dynamic policy tuning and assess interoperability among heterogeneous XDR components in complex multi‑cloud environments.