Muhammad Hamza Zafar

• Moosavi, Syed Kumayl Raza; Zafar, Muhammad Hamza; Bukhari, Syed Muhammad Salman; Hadi, Shahzaib Farooq & Sanfilippo, Filippo (2026). Exploring the Synergies between Federated Learning, Collaborative Robotics, and Autonomous Systems: Shaping the path to Industry 5.0. Proceedings - European Council for Modelling and Simulation (ECMS). ISSN 2522-2414. 2026, p. 92–101. doi: 10.7148/2026-0092. Full text in Research Archive Show summary

  The advent of Industry 5.0 marks a transformative era in industrial development, where advanced Technologies converge to create more intelligent, efficient, and humancentric systems. This paper examines the synergistic integration of Federated Learning (FL), Collaborative Robotics (cobots), and Autonomous Systems (AS), exploring their combined potential in shaping the trajectory toward Industry 5.0. We present a comprehensive analysis of how these technologies collectively enhance the capabilities of industrial systems, with emphasis on humanmachine collaboration, resilience, adaptability, and sustainability. A case study centred on a disaster response scenario is introduced to illustrate the practical benefits of this technological convergence in real-world, high-stress environments. Our findings clarify that the convergence of these technologies not only propels the industrial sector toward the visionary goals of Industry 5.0 but also opens new avenues for addressing complex challenges across various industrial and societal domains.

• Murtaza, Aitzaz Ahmed; Zafar, Muhammad Hamza; Moosavi, Syed Kumayl Raza; Aftab, Muhammad Faisal & Sanfilippo, Filippo (2026). Tool wear prediction using multi-domain features and quantised attention embedded CNN. Measurement: Sensors. 47. doi: 10.1016/j.measen.2026.102013. Full text in Research Archive Show summary

  In precision manufacturing, accurate prediction of tool wear is pivotal for maintaining production quality and efficiency. This paper introduces a novel predictive model that employs a multi-branch convolutional neural network (multi-CNN) enhanced by an attention mechanism and Bayesian optimisation to forecast tool wear with high accuracy. The proposed approach is evaluated using the publicly available dataset, which comprises force and torque measurements acquired from machining processes. By incorporating time-domain (TD), frequency-domain (FD), and discrete wavelet transform (DWT) features, the model comprehensively analyses signals from machining processes, enabling robust predictions. The attention mechanism strategically emphasises significant features, augmenting the model’s predictive capability. Demonstrated results show superior performance with a mean absolute error (MAE) of 2.8, root mean squared error (RMSE) of 3.4, and an R2 score of 0.986, significantly outperforming conventional models. Additionally, post-training quantisation optimises the model for deployment on edge computing devices, maintaining effectiveness while being lightweight, ideal for real-time applications in resource-constrained environments. This study exemplifies how advanced machine learning techniques can improve predictive maintenance in manufacturing which not only enhances operational efficiency but also reduces downtime.

• Al-Tawalbeh, Nedaa; Zainuri, Muhammad Ammirrul Atiqi Mohd; Bataineh, Ahmad Sami; Olaimat, Melad M.; Akhtar, Naureen & Zafar, Muhammad Hamza [Show all 7 contributors for this article] (2026). Design and analysis of a charge-pump-assisted high-gain DC–DC converter for low-voltage hybrid PV–TEG energy harvesting applications. Energy Conversion and Management: X. 31. doi: 10.1016/j.ecmx.2026.102216. Full text in Research Archive Show summary

  Low-voltage renewable energy harvesting applications, particularly hybrid photovoltaic–thermoelectric generator (PV–TEG) systems, require high-gain DC–DC converters capable of efficiently boosting low and fluctuating input voltages while maintaining high conversion stability. However, conventional converter topologies often suffer from limited voltage gain, increased semiconductor stress, and reduced performance under the variable operating conditions characteristic of hybrid PV–TEG energy harvesting. To address these challenges, this paper proposes a novel Charge-Pump-Assisted Voltage Booster (CPAVB) converter for low-voltage hybrid PV–TEG energy harvesting applications. The proposed topology integrates inductive and capacitive energy storage elements in a charge-pump-assisted series-stacking configuration to achieve a high static voltage gain at moderate duty cycles while reducing component stress and improving conversion stability. Comprehensive simulation studies were performed in MATLAB/Simulink under diverse operating conditions, including uniform irradiance and temperature, partial shading with thermal non-uniformity, and low-voltage scenarios. The proposed converter demonstrated stable operation, rapid dynamic response, and accurate maximum power extraction. Furthermore, its applicability to a single-phase grid-connected hybrid PV–TEG system was investigated through simulation, where it maintained a well-regulated DC-link voltage, enabled unity power factor operation, achieved a simulated grid-current total harmonic distortion (THD) of 0.08%, and injected active power closely matching the theoretical hybrid maximum. To validate the proposed converter, a hardware prototype was developed and experimentally evaluated. The experimental results confirmed the predicted voltage gain, stable low-voltage operation, and acceptable semiconductor voltage stresses, demonstrating the practicality of the proposed converter. Overall, the combined simulation and converter-level experimental results demonstrate that the proposed CPAVB converter is a promising high-gain power conversion solution for low-voltage renewable energy harvesting applications and provides a practical interface for future hybrid PV–TEG systems and microgrid integration.

• Das, Opy; Zafar, Muhammad Hamza; Rudra, Souman & Sanfilippo, Filippo (2026). Adaptive battery SOC estimation: Temporal attention-enhanced BiLSTM and quantized model comparison for scalable edge solutions. Energy Reports. 16. doi: 10.1016/j.egyr.2026.109544. Full text in Research Archive Show summary

  Lithium-ion batteries are the dominant energy storage technology in modern energy systems, making the accurate state-of-charge (SOC) estimation essential for effective battery management. Reliable SOC estimation helps determine remaining battery capacity, improves operational safety, and supports real-time energy control. However, SOC estimation is challenging due to the nonlinear electrochemical behavior of lithium-ion batteries and performance degradation caused by aging and temperature variations. This study proposes a Bidirectional Long Short-Term Memory network with a Temporal Attention Mechanism (BiLSTM-TAM) to enhance SOC estimation accuracy across a wide temperature range from −20 °C to 40 °C. The model is evaluated under several driving conditions, including the Urban Dynamometer Driving Schedule (UDDS), the US06 aggressive driving cycle, a mixed driving cycle that combines multiple standard cycles, and the Hybrid Pulse Power Characterization (HPPC) test commonly used for battery modeling. Model performance is assessed using root-mean-square error (RMSE), mean absolute error (MAE), and the coefficient of determination (R ). Results demonstrate that the proposed BiLSTM-TAM model improves SOC estimation accuracy and robustness. To enable practical deployment, post-training quantization (PTQ) is applied to reduce model size and computational requirements while maintaining predictive performance, making it suitable for edge-based electric vehicle applications.

• Masood, Muhammad Zeeshan; Zafar, Muhammad Hamza; Moosavi, Syed Kumayl Raza; Demrozi, Florenc; Shaukat, Furqan & Sanfilippo, Filippo (2026). Towards robust neuromorphic/event vision-based fall detection: Cross-dataset generalisation benchmarking. Expert Systems With Applications. ISSN 0957-4174. 325. doi: 10.1016/j.eswa.2026.132675. Full text in Research Archive Show summary

  Fall detection is a critical safety concern in many real-world environments, including smart homes, healthcare and assisted living facilities, workplaces, and other indoor/outdoor environments where individuals may be exposed to situations that increase the risk of accidental falls. Vision-based systems have emerged as a promising solution due to their non-intrusive nature and strong recognition performance. However, traditional RGB-based approaches suffer from limitations, including sensitivity to lighting conditions, occlusions, and privacy concerns. In contrast, event cameras capture asynchronous brightness changes with high temporal resolution and low latency, making them more robust in challenging conditions. Despite this potential, progress in event-based fall detection is limited by the lack of large-scale datasets and standardized evaluation protocols. In this paper, we present a benchmarking study based on four widely used RGB fall detection datasets–CAUCA Fall, GMDCSA24, Fall Vision, and Multiple Cameras Fall Dataset–converted into event data using the Metavision Simulator. These datasets are used to train and evaluate nine different deep learning models (i.e., I3D, 3DCNN, ConvLSTM, MobileNetV3-LSTM, TCN, SwinTransformer, TimeSFormer, EfficientViT, and TemporalMotionNet). To evaluate cross-domain generalization, the trained models are tested on a real-world event dataset, the Pedestrian Detection, Action Recognition, and Fall Detection (PAF) dataset. Our study highlights the challenges of simulation-to-real transfer in event-based fall detection and provides a comparative evaluation across synthetic and real event datasets. The results show that while TemporalMotionNet achieves high performance on synthetic data (98.65% accuracy), performance drops significantly on real data, with a 17.25% gap (81.4% accuracy on PAF). The converted datasets, together with the evaluation protocol, are publicly released (https://github.com/AIBCRG/EventFallDetection) to foster further research in this domain.

• Abdullah, Muhammad; Shaukat, Furqan; Zafar, Muhammad Hamza; Abdulkadir, Said Jadid & Ismail, Ahmed Omer Ahmed (2026). Attention-ResUNet and EfficientSASM-UNet: UNet-Based Frameworks for Lung and Nodule Segmentation. IEEE Access. 14, p. 59948–59958. doi: 10.1109/ACCESS.2026.3683298. Full text in Research Archive
• Alwan, Abdullah A.; Faraj, Sura H.; Darweesh, Salih Y.; Ali, Ahmed J. & Zafar, Muhammad Hamza (2026). MPPT for PV Systems Under Partial Shading Using Cuckoo Search With Perturb and Observe Optimizations. Al-Bahir Journal for Engineering and Pure Sciences. ISSN 2312-5721. 8(2), p. 248–264. doi: 10.55810/2313-0083.1127. Full text in Research Archive
• Hamad, Qusay Shihab; Samma, Hussein; Suandi, Shahrel Azmin; Hussien, Abdelazim G.; Zafar, Muhammad Hamza & Hamad, Yasameen Shihab (2026). QLESCA: A reinforcement learning based metaheuristic for efficient CNN filter pruning. Biomedical Signal Processing and Control. ISSN 1746-8094. 120. doi: 10.1016/j.bspc.2026.110214. Full text in Research Archive
• Zafar, Muhammad Hamza; Moosavi, Syed Kumayl Raza & Sanfilippo, Filippo (2026). CurFed-CHARM: Curriculum sequential federated learning with a channel-hierarchical action recognition model for non-IID heterogeneous data in neuromorphic/event cameras. Computer Vision and Image Understanding. ISSN 1077-3142. 265. doi: 10.1016/j.cviu.2026.104696. Full text in Research Archive
• Eman, Hafza; Shaukat, Furqan; Zafar, Muhammad Hamza & Anwar, Syed Muhammad (2026). EMeRALDS: Electronic Medical Record Driven Automated Lung Nodule Detection and Classification in Thoracic CT Images. Journal of Imaging Informatics in Medicine (JIIM). ISSN 2948-2925. doi: 10.1007/s10278-026-01901-7. Full text in Research Archive
• Langås, Even Falkenberg; Aalerud, Atle; Nyberg, Svein Olav Glesaaen; Zafar, Muhammad Hamza; Hagen, Daniel & Sanfilippo, Filippo (2026). Predictive Human Trajectory Classification in Industrial Settings with Transfer Learning and Data Augmentation. IEEE Access. 14, p. 47071–47084. doi: 10.1109/ACCESS.2026.3677599. Full text in Research Archive Show summary

  The successful realisation of Industry 5.0 relies on seamless collaboration between humans and robots in shared workspaces. This paper investigates predictive human trajectory classification (HTC) in a factory setting to enhance safety, productivity, and adaptability in human-machine environments. Using a state-of-the-art light detection and ranging (LiDAR) system with reflective mirrors to extend the field of view, motion data are collected across a 26 m ×20 m indoor industrial floor. Human trajectories are classified to predict workstation transitions using a stacked long-short term memory (LSTM) network. To address the challenges of limited real-world data, transfer learning is applied by first training the model on synthetically generated trajectories. The paper compares a traditional stochastic method, specifically the Ornstein-Uhlenbeck (OU) process, with a modern learning based method, represented by a timeseries generative adversarial network (TimeGAN). Further on, the algorithm is fine-tuned on the real-world dataset. The comparison shows that in data-scarce scenarios, OU outperforms TimeGAN in most metrics. The LSTM model trained on OU-based data and real-world data shows an overall accuracy gain of 4.8 percentage points compared to training on real-world data only. It is able to classify the goal workstation three seconds in advance with an accuracy of 95.6%, demonstrating the potential for predictive models to improve workplace operations by reducing stall time, optimising workstation readiness, and reducing the need for traditional safety cages.

• Zafar, Muhammad Hamza; Moosavi, Syed Kumayl Raza & Sanfilippo, Filippo (2026). MATRIX-HAR: Lightweight temporal motion-guided feature network with adaptive temporal resolution based transfer learning for resource-constrained event camera-based human action recognition. Expert Systems With Applications. ISSN 0957-4174. 303. doi: 10.1016/j.eswa.2025.130628. Full text in Research Archive Show summary

  Event-based cameras have emerged as a promising alternative to traditional sensors for human action recognition due to their high temporal resolution, low latency, and robustness to lighting variations. However, existing methods struggle to effectively capture the complex spatio-temporal dynamics inherent in event streams, particularly under challenging environmental conditions. In this paper, we propose Lightweight Temporal Motion-guided Feature Network with Adaptive Temporal Resolution Based Transfer Learning for Resource-Constrained Event Camera-based Human Action Recognition (MATRIX-HAR), a novel deep learning architecture specifically designed for event-based human action recognition that incorporates three key innovations: an Adaptive Temporal Resolution (ATR) module that dynamically adjusts temporal processing granularity based on input characteristics, a Motion Guided Feature Module (MGFM) that leverages intrinsic motion information encoded in event streams, and a comprehensive hierarchical transfer learning framework that enables effective knowledge transfer from simple to challenging environmental settings. We systematically evaluate four distinct transfer learning strategies, ranging from early feature transfer to comprehensive multi-level transfer, demonstrating their effectiveness in adapting models trained on simple scenarios to complex real-world conditions. To ensure practical deployment on resource-constrained edge devices, we implement post-training quantization that maintains performance while achieving 4 ×  compression ratios. Extensive experiments on four benchmark datasets (HARDVS, PAF, THU-E-ACT-50, and THU-E-ACT-50 CHL) demonstrate that MATRIX-HAR achieves state-of-the-art performance, surpassing existing methods by significant margins: 0.71 % on HARDVS (97.82 %), 0.74 % on PAF (97.23 %), 3.03 % on THU-E-ACT-50 (95.73 %), and 7.3 % on THU-E-ACT-50 CHL (65.8 %). Ablation studies confirm the individual contributions of each proposed module, while transfer learning analysis reveals that comprehensive feature transfer (Strategy 4) provides the most robust adaptation with 22.3 % relative performance improvement over baseline approaches. The proposed framework demonstrates superior learning dynamics, achieving faster convergence and enhanced stability compared to training from scratch, making it particularly suitable for real-world deployment scenarios where environmental conditions are variable and computational resources are limited.

• Al-Tawalbeh, Nedaa; Zafar, Muhammad Hamza; Radzi, Mohd Amran Mohd & Zainuri, Muhammad Ammirrul Atiqi Mohd (2025). An innovative design approach: optimized super-lift Luo converter for solar energy extraction. Energy Systems, Springer Verlag. ISSN 1868-3967. doi: 10.1007/s12667-025-00770-3. Full text in Research Archive Show summary

  This paper proposes a novel design methodology for a super-lift Luo converter to efficiently perform in maximum power point tracking (MPPT) systems, considering the wide range of weather conditions that can affect the converter’s effectiveness and performance. This approach optimizes the design of the super-lift Luo converters’ inductance, and capacitance components and output resistance, all of which are essential to the converter’s efficiency and ripple performance. The innovative aspect of the proposed design is to engage the photovoltaic (PV) array’s maximum power point as an optimal power point, which can be recognized based on the weather and the PV array’s characteristics. Optimizing the inductance size selection based on the extracted energy amount allows the MPPT super-lift Luo converter’s ripple quantity and duty cycle to be regulated in compliance with the necessary specifications. The robust performance of the proposed design strategy has been validated experimentally as well as by simulation. The prototype of the well-designed super-lift Luo converter is implemented and tested under uniform solar radiation, different shading situations, and sudden changes in incident radiation. The performance aspects are compared with that of the boost converter designed based on a recent design method. The findings confirm the super-lift Luo converter outperformance in terms of effectiveness, stability, diminished ripples, and readiness for radiation fluctuation. The stability offered by the proposed solution eliminates the need for an input capacitor at the convertor input, allowing a rise in steady PV voltage to a greater level than for the boost at virtually the same cost.

• Houran, Mohamad Abou; Srivastava, Gautam; Mirza, Jawad; Ranjha, Ali; Javed, Muhammad Awais & Zafar, Muhammad Hamza (2025). Centralized Task Allocation for Multiple UAVs in Time-Constraint Industrial IoT Operations. IEEE Internet of Things Journal. ISSN 2327-4662. 12(18), p. 37529–37537. doi: 10.1109/JIOT.2025.3584277. Full text in Research Archive Show summary

  The industrial Internet of Things (IoT) allows real-time monitoring and operational efficiency by enabling automated processes in industrial environments. In this article, we propose a centralized task assignment framework for industrial IoT scenarios, focusing on light cargo delivery to specific locations via unmanned aerial vehicles (UAVs). The task assignment problem is formulated as a capacitated vehicle routing problem (CVRP) with the objective of minimizing the maximum route distance among all UAVs. To solve CVRP, we employ a learning-based approach using an attention model (AM), which utilizes a deep learning framework with an encoder–decoder architecture to generate optimized UAV routes while satisfying the capacity constraints of UAVs. The AM is trained using policy gradient reinforcement learning (RL) to ensure that the solutions are both efficient and scalable. Numerical results demonstrate the effectiveness of the AM-based framework in delivering solutions that minimize the maximum tour length for deliveries.

• Khan, Muhammad Kamran; Khan, Muhammad Ammar; Kauhaniemi, Kimmo; Zafar, Muhammad Hamza & Rashid, Saad (2025). A New Algorithm for Multi-Objective Hybrid Economic Emission Dispatch Using Data-Driven Forecasting of Wind and Solar Power. IET Generation, Transmission & Distribution. ISSN 1751-8687. 19(1). doi: 10.1049/gtd2.70197. Full text in Research Archive Show summary

  This study introduces a forecasting-driven framework for solving the hybrid dynamic economic emission dispatch (HDEED) problem with uncertain wind and solar generation. Unlike conventional approaches that rely on probabilistic models, this research uses data-driven forecasting techniques to predict the output powers of wind and PV plants and uses them in the multi-objective optimisation algorithm to minimise costs and emissions. An arithmetic optimiser with a sine cosine assisted driving training-based optimisation algorithm (AOASC-DTBO) is presented to solve various single and multi-objective HDEED problems based on fuzzy decision-making and the Pareto dominance concept. The proposed dispatching algorithm is validated using 10-unit, 40-unit and 7-unit IEEE 57 bus systems. The results revealed that the AOASC-DTBO algorithm achieved 8.73% and 4.95% lower fuel costs when compared with the PSO and DTBO algorithms, respectively. In addition, the emissions were 6.47% and 2.25% lower than the BMO and DTBO algorithms, respectively. The work highlights the importance of integrating renewable energy sources (RES) into power systems to achieve cost savings and reduced emissions, while also emphasising the need for efficient dispatch algorithms to ensure grid stability and reliability. The results demonstrate that integrating RES into the power system can yield substantial economic and environmental benefits, achieving cost savings of up to $6,908.034 and reducing emissions by 13,233.691 tonnes per day.

• Khan, Muhammad Kamran; Kauhaniemi, Kimmo; Laaksonen, Hannu & Zafar, Muhammad Hamza (2025). Quantized deep learning model based Volt-Var control for hosting capacity maximization: a practical case study. International Journal of Electrical Power & Energy Systems. ISSN 0142-0615. 174. doi: 10.1016/j.ijepes.2025.111524. Full text in Research Archive Show summary

  This study introduces a data-driven Volt-Var control strategy aimed at maximizing hosting capacity (HC) and mitigating voltage violations in distribution networks. Central to the proposed methodology is a quantized one-dimensional convolutional neural network (QCNN), developed to emulate optimal Volt-Var control decisions. Post-training quantization is applied, resulting in tenfold reduction in model size, making QCNN well-suited for deployment on resource-constrained edge devices. The training data is prepared using Modified Reptile Search Algorithm, which determines optimal Volt-Var control set points for HC maximization. The QCNN model trained offline processes measurements from the test network to predict the optimal reactive power reference points, which are supplied to the Decoupled Finite Control Set Model Predictive Controller (D-FCS-MPC). The D-FCS- MPC subsequently determines the optimal inverter switching states to directly regulate reactive powers to the predicted reference values. The methodology is validated using model of a real-world medium voltage (MV) network situated in Vaasa, Finland, known as the Sundom Smart Grid (SSG). Both simulation and OPAL-RT based real time results confirms the effectiveness of the proposed methodology in maximizing hosting capacity and ensuring grid stability.

• Zafar, Muhammad Hamza; Moosavi, Syed Kumayl Raza & Sanfilippo, Filippo (2025). Applications of Neuromorphic/Event Camera in Robotics With Human in Loop: A Systematic Review, Datasets, and Challenges. IEEE Transactions on Human-Machine Systems. ISSN 2168-2291. 56(1), p. 32–47. doi: 10.1109/thms.2025.3628064. Full text in Research Archive Show summary

  The evolution of industrial robotics has advanced from isolated, caged systems through basic human–robot interaction (HRI) to sophisticated human–robot collaboration (HRC). However, conventional vision systems based on red, green, blue (RGB) cameras remain a significant limiting factor in realizing the full potential of collaborative automation. This comprehensive review examines the transformative role of event cameras in advancing HRC capabilities and addressing current limitations in industrial settings. Event cameras, with their microsecond-level temporal resolution and robust performance under challenging lighting conditions, offer substantial advantages over traditional RGB cameras that are constrained by fixed frame rates and ambient lighting dependencies. We present a systematic framework for leveraging event cameras to enhance the human state understanding in collaborative robotics, encompassing real-time detection of poses, gestures, facial expressions, and emotional states. This framework addresses fundamental challenges in workplace safety and collaborative efficiency while enabling more sophisticated and responsive HRC systems. Our review synthesizes recent research developments in event camera applications specific to HRC, providing a detailed comparative analysis of their advantages over conventional vision systems. We identify emerging opportunities and potential research directions for advancing event-based vision in industrial robotics. In addition, we examine integration challenges and propose strategies for implementing event camera technology in existing industrial infrastructure. This work contributes valuable insights into the future trajectory of adaptive and intuitive HRC systems, offering a roadmap for researchers and practitioners in the field of industrial automation.

• Moosavi, Syed Kumayl Raza; Zafar, Muhammad Hamza & Sanfilippo, Filippo (2025). Snake robots: A state-of-the-art review on design, locomotion, control, and real-world applications. Mechatronics (Oxford). ISSN 0957-4158. 112. doi: 10.1016/j.mechatronics.2025.103418. Full text in Research Archive Show summary

  Snake robots have emerged as a transformative class of bio-inspired robotic systems, offering unparallelled adaptability in navigating complex unstructured terrains. Their limbless design, inspired by biological snakes, enables efficient movement across diverse environments, including rough terrains, confined spaces, and hazardous conditions where traditional wheeled or legged robots struggle. This review provides a comprehensive analysis of the design and development of snake robots, covering their locomotion strategies, mechanical configurations, sensor integration, control schemas, and actuation mechanisms. The review further explores the evolution of mechanical structures from rigid, soft, and hybrid designs, emphasising advancements in actuation and sensor technologies in enhancing adaptability and navigation. Applications of snake robots extend across various domains, including search and rescue (SAR), industrial inspection, and exploration of extreme environments. Despite significant progress, challenges such as optimising energy efficiency, improving environmental perception, and achieving real-time adaptability remain open research areas. This review serves as a foundational reference for researchers and engineers working towards advancing the next generation of snake robots, paving the way for their integration into real-world applications

• Zafar, Muhammad Hamza; Moosavi, Syed Kumayl Raza & Sanfilippo, Filippo (2025). Neuromorphic Event Camera-Based Object Recognition and Grasping Position Detection Using a Transfer Learning-Enhanced Multi-Task Model. IEEE Transactions on Automation Science and Engineering. ISSN 1545-5955. 22, p. 20169–20181. doi: 10.1109/tase.2025.3598695. Full text in Research Archive Show summary

  Object recognition and grasping position detection are critical tasks in robotic manipulation, particularly when operating in dynamic and unstructured environments. This paper presents the Channel Sharpening Attention-based Adaptive Inception Network (CSA-AInceptNet), a novel multi-task learning model designed for these tasks using event camera data. The proposed architecture integrates channel sharpening attention with adaptive inception networks to enhance feature extraction and improve robustness. The model’s performance is evaluated on two state-of-the-art event camera datasets, E-Grasp and Neuro-Grasp. On the E-Grasp dataset, CSA-AIncepNet achieves a remarkable accuracy of 99.47% and a mean Intersection over Union (IoU) of 0.9370, significantly surpassing existing methods. On the Neuro-Grasp dataset, leveraging transfer learning, the model attains 98.58% accuracy and a mean IoU of 0.4897, demonstrating strong generalization capabilities across datasets. Comparative analyses and ablation studies further validate the effectiveness of the proposed architecture, highlighting its superiority over conventional models like ConvNeXt, DarkNet, DenseNet, and VGG16. The results establish CSA-AIncepNet as a robust solution for event-based object recognition and grasping detection, paving the way for advancements in human-robot collaboration and dynamic robotic manipulation. Note to Practitioners—This work provides a practical solution for improving object recognition and grasping position detection in robotic systems, particularly in unpredictable and fast-changing real-world environments. By leveraging event camera data, the proposed approach enables robots to efficiently identify objects and determine optimal grasping positions, even under challenging conditions. The results highlight the model’s ability to outperform existing methods, making it highly suitable for applications such as human-robot collaboration and precise object handling. This advancement has significant implications for industries like manufacturing, logistics, and healthcare, where robots must interact with objects quickly and accurately. Practitioners can adopt this method to enhance robotic performance, reduce errors, and improve operational efficiency. Future work could focus on testing the model in more complex environments and adapting it for real-time deployment in dynamic settings.

• Zafar, Muhammad Hamza; Moosavi, Syed Kumayl Raza & Sanfilippo, Filippo (2025). Federated Learning-Enhanced Edge Deep Learning Model for EMG-Based Gesture Recognition in Real-Time Human-Robot Interaction. IEEE Sensors Journal. ISSN 1530-437X. 25(5), p. 9139–9151. doi: 10.1109/JSEN.2025.3529841. Full text in Research Archive Show summary

  Electromyography (EMG)-based gesture detection plays a crucial role in human-robot interaction (HRI), providing a seamless interface for controlling robotic systems through muscle activity. Despite its potential, EMG systems face significant challenges related to the security and privacy of sensitive biometric data, as well as the computational limitations of deploying deep learning (DL) models on edge devices. To address these issues, we propose a federated learning (FL)-based DL model for EMG gesture recognition, specifically designed for edge devices. Our model utilizes a custom dataset collected using a Mindrove eight-channel EMG armband, capturing eight distinct hand gestures—rest, move left, move right, move down, move up, open fingers, close fist, and twist hand—from ten subjects with seven repetitions each, ensuring diverse and robust data for training. During preprocessing, a bandpass filter (50–450 Hz) was applied to remove noise and enhance signal quality, followed by a short-term frequency transform (STFT) with a 200-ms sample time and 50% overlap to extract relevant features from the EMG signals. The dataset was segmented into training and testing sets with a 70/30 split for evaluation. We evaluate several FL techniques, including FedAvg, FedProx, and FedSGD, demonstrating that FedAvg achieves the highest accuracy of 96.92% without quantization with Scenario 9 (15 epochs, 20 rounds) with minimal communication overhead. Additionally, our model is quantized, resulting in an 89% reduction in size and a high accuracy of 95.99%, representing a minimal loss of 0.93%, making it ideal for edge deployment without compromising performance. A comparative analysis with other DL models, such as multiconvolutional residual networks (MCRNs), multiconvolutional neural networks (MCNNs), temporal convolutional networks (TCNs), and InceptionNet, shows that our approach outperforms these models in both accuracy and efficiency. Experimental results validate the high accuracy of our model in both training/testing and real-time disaster scenario simulations using the Spot robot. The proposed solution provides a secure, efficient, and highly accurate framework for EMG-based gesture recognition on edge devices, ideal for HRI and assistive technologies such as in search and rescue operations.

• Zafar, Muhammad Hamza; Poursina, Mohammad; Moosavi, Syed Kumayl Raza & Sanfilippo, Filippo (2025). Trajectory Optimisation of a Robotic Manipulator Using a Novel Evolutionary Intelligence Algorithm. In Scarpa, Marco; Cavalieri, Salvatore; Serrano, Salvatore & Vita, Fabrizio De (Ed.), Proceedings of the 39th ECMS International Conference on Modelling and Simulation (ECMS 2025), June 24th - June 27th, 2025, Catania, Italy. ECMS European Council for Modelling and Simulation. ISSN 9783937436869. p. 93–99. doi: 10.7148/2025-0093. Full text in Research Archive Show summary

  In this paper, we present a novel approach for Proportional–Integral–Derivative (PID) tuning of a robotic manipulator, modeled using the Lagrange method and validated through comprehensive modeling and simulation. To optimize the PID gains, we developed a hybrid algorithm that combines Greylag Goose Optimization (GGO) with the Sine Cosine Algorithm (SCA), leveraging the strengths of both optimization techniques. The proposed hybrid algorithm, termed GGOSCA, was tested against GGO and Particle Swarm Optimization (PSO) using three objective functions: Lyapunov Based Function (LBF), Integral of Absolute Error (IAE), and Integral of Time-weighted Absolute Error (ITAE). The results demonstrated that GGOSCA outperforms both GGO and PSO across all objective functions. Specifically, GGOSCA achieved the lowest costs of 0.1018, 0.2484, and 0.6601 for LBF, IAE, and ITAE, respectively, compared to 0.1022, 0.2580, and 0.6939 for GGO, and 0.1023, 0.2660, and 0.7273 for PSO. The superior performance of GGOSCA highlights its effectiveness in balancing exploration and exploitation, making it well-suited for complex control tasks such as PID tuning in robotic systems. This novel combination of GGO and SCA provides a robust and efficient solution for optimizing control parameters in dynamic environments, demonstrating its potential through rigorous modeling and simulation.

• Sanfilippo, Filippo; Zafar, Muhammad Hamza; Wiley, Timothy & Zambetta, Fabio (2025). From caged robots to high-fives in robotics: Exploring the paradigm shift from human–robot interaction to human–robot teaming in human–machine interfaces. Journal of manufacturing systems. ISSN 0278-6125. 78, p. 1–25. doi: 10.1016/j.jmsy.2024.10.015. Full text in Research Archive Show summary

  Multi-modal human–machine interfaces have recently undergone a remarkable transformation, progressing from simple human–robot interaction (HRI) to more advanced human–robot collaboration (HRC) and, ultimately, evolving into the concept of human–robot teaming (HRT). The aim of this work is to delineate a progressive path in this evolving transition. A structured, position-oriented review is proposed. Rather than aiming for an exhaustive survey, our objective is to propose a structured approach in a field that has seen diverse and sometimes divergent definitions of HRI/C/T in the literature. This conceptual review seeks to establish a unified and systematic framework for understanding these paradigms, offering clarity and coherence amidst their evolving complexities. We focus on integrating multiple sensory modalities — such as visual, aural, and tactile inputs — within human–machine interfaces. Central to our approach is a running use case of a warehouse workflow, which illustrates key aspects including modelling, control, communication, and technological integration. Additionally, we investigate recent advancements in machine learning and sensing technologies, emphasising robot perception, human intention recognition, and collaborative task engagement. Current challenges and future directions, including ethical considerations, user acceptance, and the need for explainable systems, are also addressed. By providing a structured pathway from HRI to HRT, this work aims to foster a deeper understanding and facilitate further advancements in human–machine interaction paradigms.

• Bukhari, Syed Muhammad Salman; Zafar, Muhammad Hamza; Moosavi, Syed Kumayl Raza & Sanfilippo, Filippo (2025). Emotion recognition with a Randomized CNN-multihead-attention hybrid model optimized by evolutionary intelligence algorithm. Array. 26. doi: 10.1016/j.array.2025.100401. Full text in Research Archive Show summary

  Emotion recognition systems are vital for various applications, yet existing models often face limitations in computational efficiency and accuracy, especially when handling complex emotional expressions in sequential data. To address these challenges, we propose an innovative emotion recognition framework that integrates a Randomised Convolutional Neural Network (RCNN) with a Multi-Head Attention model, further optimized by the Football Team Training Algorithm (FTTA) metaheuristic to enhance network parameters effectively. The RCNN, characterized by fixed random weights in its convolutional layers, efficiently extracts features from facial landmarks, enabling robust and diverse feature extraction while reducing computational load. This structure is complemented by a multi-head attention mechanism that processes temporal dynamics in emotion data, with both components optimized through FTTA to balance exploration and exploitation. Our hybrid model undergoes rigorous testing on a widely recognized emotion recognition dataset, outperforming conventional fully trainable models and alternative architectures. The results indicate a substantial improvement in classification accuracy, with an overall accuracy of 99%, and a significant reduction in computational demands, achieving a 65% faster training time on average compared to state-of-the-art models. These enhancements confirm the model’s efficiency and robustness across various emotional classifications. The synergy between the RCNN’s fixed-weight feature extraction and FTTA’s optimization capabilities demonstrates a powerful solution for emotion recognition systems. The combination of accuracy and efficiency renders our model suitable for real-world applications, particularly in fields like healthcare and mental health monitoring, where real-time emotion detection can have significant impacts.

• Bukhari, Syed Muhammad Salman; Dahmani, Nadia; Gyawali, Sujan; Zafar, Muhammad Hamza; Sanfilippo, Filippo & Raja, Kiran (2025). Optimizing fire detection in remote sensing imagery for edge devices: A quantization-enhanced hybrid deep learning model. Displays (Guildford). ISSN 0141-9382. 89. doi: 10.1016/j.displa.2025.103070. Full text in Research Archive Show summary

  Wildfires are increasing in frequency and severity, presenting critical challenges for timely detection and response, particularly in remote or resource-limited environments. This study introduces the Inception-ResNet Transformer with Quantization (IRTQ), a novel hybrid deep learning (DL) framework that integrates multiscale feature extraction with global attention and advanced quantization. The proposed model is specifically optimized for edge deployment on platforms such as unmanned aerial vehicles (UAVs), offering a unique combination of high accuracy, low latency, and compact memory footprint. The IRTQ model achieves 98.9% accuracy across diverse datasets and shows strong generalization through cross-dataset validation. Quantization significantly reduces the parameter count to 0.09M and memory usage to 0.13 MB, enabling real-time inference in 3 ms. Interpretability is further enhanced through Grad-CAM visualizations, supporting transparent decisionmaking. While achieving state-of-the-art performance, the model encounters challenges in visually ambiguous fire-like regions. To address these, future work will explore multi-modal inputs and extend the model towards multi-class classification. IRTQ represents a technically grounded, interpretable, and deployable solution for AI-driven wildfire detection and disaster response.

• Gonsholt, Alf Stian Sundo; Greca, Eivind Enea; Haddad, Mustapha; Zafar, Muhammad Hamza & Sanfilippo, Filippo (2025). Enhanced Human-Robot Teaming Through Attention Multi Convolutional Neural Network-Based Multi-Modal Sensor Fusion for Hand Gesture Recognition and Orientation Control. In Bui, Tung (Eds.), Proceedings of the 58th Hawaii International Conference on System Sciences (HICSS), Hilton Waikoloa Village, January 7-10, 2025. University of Hawai'i at Manoa. ISSN 9780998133188. p. 590–599. Full text in Research Archive Show summary

  Our study aims at enhancing Human-Robot Interaction, Collaboration, and Teaming (HRI/C/T) in industrial automation by developing a novel framework for real-time gesture control of a robotic hand. We use an Inertial Measurement Unit (IMU) sensor for precise orientation control of the end effector, and surface Electromyography (sEMG) sensors to detect muscle movements. The sEMG signals are processed by an Attention-based Multi Convolutional Neural Network (A-MCNN) for accurate gesture detection, enabling the robotic hand to mimic these gestures in real-time. Our method achieves notable results for gesture recognition, with the A-MCNN model attaining an accuracy of 97.89%, a precision of 97.49%, a recall of 97.71%, and an F1 score of 97.65%. This integration of IMU and sEMG technologies with advanced neural networks creates a responsive and intuitive control mechanism, improving safety, usability, and interaction of collaborative robots in shared workspaces. Our approach aims to transition towards Human-Robot Teaming (HRT), significantly advancing the seamless and safe integration of robots in industrial environments, enhancing productivity and collaboration.

• Khan, Umer Amir; Asif, Mansoor; Zafar, Muhammad Hamza & Alhems, Luai (2025). Experimental validation of machine learning for contamination classification of polluted high voltage insulators using leakage current. Scientific Reports. 15. doi: 10.1038/s41598-025-97646-4. Full text in Research Archive Show summary

  This paper presents a comprehensive experimental validation of machine learning for contamination classification of polluted high voltage insulators using leakage current. A meticulous dataset of leakage current for porcelain insulators with varying pollution levels was developed under controlled laboratory conditions. Critical parameters of temperature and varying humidity were also included in the dataset to reflect the impact of environmental conditions and bring the dataset close to real world scenarios. The dataset generated was preprocessed and critical features were extracted from time, frequency, and time-frequency domains. Four distinct machine learning models, encompassing decision trees and neural networks, were trained and evaluated on this dataset. The Bayesian optimization technique was used to optimize the parameters of Machine Learning Models. The models demonstrated exceptional performance, with accuracies consistently exceeding 98 %. Notably, the decision tree-based models exhibited significantly faster training and optimization times compared to their neural network counterparts. This study underscores the effectiveness of machine learning in improving the reliability of insulator maintenance and monitoring systems, paving the way for more robust predictive maintenance strategies.

• Ibrahim, AL-Wesabi; Xu, Jiazhu; Farh, Hassan M. Hussein; Al-Shamma'a, Abdullrahman A.; Yang, Yuqing & Shi, Zhenglu [Show all 7 contributors for this article] (2025). An efficient linear-extrapolation catch-fish algorithm for maximizing the harvested power from thermoelectric generators sources. Applied Thermal Engineering. ISSN 1359-4311. 268. doi: 10.1016/j.applthermaleng.2025.125916. Full text in Research Archive Show summary

  This paper proposes a novel catch-fish optimization algorithm with linear extrapolation (LE-CFOA) to efficiently track the global maximum power (GMP) of thermoelectric generator (TEG) systems under irregular temperature conditions (ITCs). The integration of TEGs with batteries not only provides more stable power output but also allows for the development of self-powered health monitoring systems, potentially slowing battery degradation and improving overall system longevity. The proposed LE-CFOA improves upon the traditional CFOA by using the linear characteristics of TEGs within a specific search region around the GMP, enabling faster and more accurate GMP tracking without extra iterations. The performance of the proposed method was rigorously evaluated against several state-of-the-art algorithms, including CSA, DFO, GHO, GWO, IMFO, P&O, INC, PSO, and PSOGSA, through hardware-in-the-loop (HIL) experiments and MATLAB/Simulink simulations. The proposed method demonstrated superior performance, achieving a tracking efficiency of 99.92 %, exceeding CSA (99.54 %) and PSOGSA (99.85 %), with a significantly reduced tracking time of 0.003 s compared to CSA (0.5 s) and IMFO (0.33 s). Additionally, it achieved a power output of 133 W in static ITC scenarios, surpassing DFO (124 W) and GWO (105 W), while minimizing energy loss to 0.003 W, substantially lower than CSA (0.3 W) and GHO (26.1 W). These findings hold significant promise for modern applications such as wheelchairs and electrical vehicles, where thermoelectric systems can be employed to power auxiliary systems, including temperature regulation, energy recovery, and battery charging. The results confirm that LE-CFOA outperforms other algorithms in terms of tracking speed, efficiency, tracking time and energy loss under static, dynamic, and stochastic temperature variations. Finally, this paper offers valuable insights for TEG specialists and researchers working on TEG MPPT techniques.

• Al-Tawalbeh, Nedaa; Zafar, Muhammad Hamza; Abuishmais, Ibrahim Abed; Zainuri, Muhammad Ammirrul Atiqi Mohd & Radzi, Mohd Amran Mohd (2025). Maximizing energy harvesting in grid-connected hybrid photovoltaic and thermoelectric generator systems: An adaptive global power tracking technique. Environmental Progress & Sustainable Energy. ISSN 1944-7442. 44(4). doi: 10.1002/ep.14621. Full text in Research Archive Show summary

  This paper proposes an innovative strategy to optimize the integration of thermoelec-tric generator (TEG) and photovoltaic (PV) technologies into a hybrid system linked toa three-phase grid, aiming to enhance renewable energy harvesting and raise overallsystem efficiency. An adaptive overstepping tracking algorithm is presented to meetthe requirements of hybrid PV-TEG systems for grid-connected applications, combin-ing the advantages of conventional perturb and observe techniques with a modifiedtrapezoidal rule based technique. The validations were conducted under a wide rangeof weather fluctuations, including varying irradiance, temperature gradients, complexpartial shading levels, and dynamic sun irradiation. The results of the simulation showthat the proposed method performs better than some of the most widely used meta-heuristic approaches, exhibiting notable gains in energy yield of over 99.7% efficiencywith a response time of less than 10 ms, reducing steady-state error (SSE) to less than0.2%. Experimentally, the proposed technique has been put into practice and evaluatedin a variety of additional weather scenarios using a solar simulator and a microcontrol-ler (TMS320F28335) as the main part of the implemented setup. In all tests that werecarried out, the reliable and stable response with an efficiency of over 99% wasapproved. The findings demonstrate that the adaptive MPPT technique can signifi-cantly enhance the stability and efficiency of solar systems, paving the way for morerobust and high-performing renewable energy solutions.

• Sanfilippo, Filippo; Moosavi, Syed Kumayl Raza & Zafar, Muhammad Hamza (2025). Autonomy Uprising: How Civilians, First Responders, and Robots Are Redefining the Future of Search-and-Rescue (SAR). In NN, NN (Eds.), 2025 9th International Conference on Robotics and Automation Sciences: AI‑Powered Robotics Advancements (ICRAS 2025). IEEE (Institute of Electrical and Electronics Engineers). ISSN 9798350392487. p. 1–8. doi: 10.1109/ICRAS65818.2025.11108899. Full text in Research Archive Show summary

  This paper explores the dynamics of human-robot interaction (HRI) in emergency response scenarios, focusing on how autonomous robots engage with civilians and first responders in search-and-rescue (SAR) operations. We examine the role of trust-building and emotional engagement in crisis situations, where human lives depend on effective HRI. The potential for autonomous systems to support decision-making while maintaining human oversight is investigated. Using the Autonomy Levels for Unmanned Systems (ALFUS) framework, we assess SAR robots' autonomy in terms of mission complexity (MC), environmental complexity (EC), and external system independence (ESI). Additionally, the Autonomy and Technology Readiness Assessment (ATRA) method supports gradual capability enhancement, providing a roadmap to higher autonomy while ensuring safety and reliability. Based on this established methodology, we introduce the Emergency Response Interaction Framework (ERIF), a novel conceptual framework defining the roles of actors involved, to connect theoretical insights with real-world applications. This work highlights the significance of human-robot trust in crisis scenarios and proposes directions for future research on collective decision-making in emergency response HRI.

• Aftab, Muhammad Faisal; Bounoua, Wahiba; Zafar, Muhammad Hamza & Sanfilippo, Filippo (2025). Improved Stiction Detection via Hybrid Residual Embedded Inception Module Networks. IFAC-PapersOnLine. ISSN 2405-8971. 59(6), p. 145–150. doi: 10.1016/j.ifacol.2025.07.136. Full text in Research Archive
• Sanfilippo, Filippo; Salvietti, Gionata; Blažauskas, Tomas; Gabriele, Guido; Zafar, Maria & Hua, Minh Tuan [Show all 14 contributors for this article] (2025). Integrating VR, AR, and Haptics in Basic Surgical Skills Training: A Review and Perspective. IEEE Access. 13, p. 99203–99220. doi: 10.1109/ACCESS.2025.3574265. Full text in Research Archive Show summary

  Traditional surgical training faces challenges related to sustainability, cost, and ethical considerations. Limited access to resources, ethical concerns about cadaver-based learning, limited availability of surgical cases, and the high costs associated with hands-on training impede the effectiveness and accessibility of surgical education. Despite these challenges, the need for direct tactile experience remains critical, as it fosters the development of essential motor skills and tactile abilities vital for surgical practice. However, advancements in immersive technologies such as virtual reality (VR), augmented reality (AR), and haptics offer a promising solution to address these challenges. VR and AR technologies offer realistic, controlled environments where learners can visualise anatomical structures and practice surgical procedures without risks. The integration of haptic feedback in these immersive simulations further enriches the learning process by adding the sense of touch and physical interaction, aiding in mastering critical surgical skills. The incorporation of immersive technologies in surgical training presents an opportunity to address the constraints of conventional methods. These technologies offer the possibility of creating platforms that replicate real-life surgical scenarios, allowing learners to hone their skills and build confidence in a safe and controlled environment. Moreover, immersive learning offers an ethical alternative to cadaver-based training and extends access to training materials irrespective of geographical limitations. Such an approach allows for step-based training, where students may begin acquiring skills through immersive learning, then move to special surgical simulators, and finally to patients. This article deals with the impact of immersive technologies such as VR, AR, and haptics on reshaping conventional surgical training, by focusing specifically on basic surgical skills. Targeting educators, medical experts, and involved parties, the article aims to stimulate additional exploration and adoption of immersive technologies to elevate surgical competencies and patient care standards.

• Das, Opy; Dahlioui, Dounia; Zafar, Muhammad Hamza; Akhtar, Naureen; Moosavi, Syed Kumayl Raza & Sanfilippo, Filippo (2025). Ultra-short term PV power forecasting under diverse environmental conditions: A case study of Norway. Energy Conversion and Management: X. 27. doi: 10.1016/j.ecmx.2025.101072. Full text in Research Archive Show summary

  Accurate short-term solar forecasting is critical for power plant operations, grid balancing, real-time dispatching, automatic generation control, and energy trading. In Norway, where solar radiation is limited in winter and highly variable in summer, accurate predictions are essential. This study focuses on ultra-short-term forecasting of solar radiation and power output from a 37.8 kWp solar photovoltaic (PV) power plant at the University of Agder (UiA), Grimstad (58.335322°N, 8.577718°E), in southern Norway. We propose a novel forecasting model that integrates Spatial Attention, Temporal Attention, Self-Attention, CNN, and BiLSTM architectures to enhance prediction accuracy. Using a custom dataset collected from the UiA PV plant, the model’s effectiveness was validated through comprehensive ablation studies and comparative analysis with state-of-the-art methods. The proposed model achieved a low RMSE of 0.162 kW using seven days of data, demonstrating its superiority in predicting short-term PV power outputs and associated uncertainties, outperforming conventional forecasting techniques.

• Langås, Even Falkenberg; Zafar, Muhammad Hamza & Sanfilippo, Filippo (2025). Exploring the synergy of human-robot teaming, digital twins, and machine learning in Industry 5.0: a step towards sustainable manufacturing. Journal of Intelligent Manufacturing. ISSN 0956-5515. 37, p. 99–1022. doi: 10.1007/s10845-025-02580-x. Full text in Research Archive Show summary

  Sustainable manufacturing remains a central objective of Industry 5.0. By successfully implementing harmonic human-robot teams in intelligent industrial systems, the efficiency and well-being of human workers can be increased. Achieving this requires a gradual approach from caged robots to advanced, seamless collaboration between humans and robots. Initially, that means transitioning to human-robot interaction (HRI) where there is an exchange of commands between the human and the robot. Further advancements within safety considerations, including collision avoidance through advanced machine vision, enable the exchange of workspace that defines human-robot collaboration (HRC). The next stage is physical HRC (pHRC) which requires safe and controlled exchange of forces through impedance and admittance control. Finally, this paper describes human-robot teaming (HRT), which is defined by the exchange of solutions as teammates. This is enabled by combining cutting-edge technologies such as digital twin (DT), advanced vision sensors, machine learning (ML) algorithms and mixed reality (MR) human–machine interfaces for operators. A key contribution of this work is reviewing the integration of HRT with DT and ML, highlighting how these technologies enable seamless perception, prediction, and decision-making in human-centric industrial systems. By reviewing these technologies, the paper highlights current challenges, limitations and research gaps within the field of HRT and suggests potential future possibilities for HRT, such as advanced disassembly of used goods for a more sustainable manufacturing industry.

• Zafar, Muhammad Hamza; Bukhari, Syed Muhammad Salman; Houran, Mohamad Abou; Mansoor, Majad; Khan, Noman Mujeeb & Sanfilippo, Filippo (2024). DeepTimeNet: A novel architecture for precise surface temperature estimation of lithium-ion batteries across diverse ambient conditions. Case Studies in Thermal Engineering. 61. doi: 10.1016/j.csite.2024.105002. Full text in Research Archive
• Sanfilippo, Filippo & Zafar, Muhammad Hamza (2024). Human-Robot Teaming: A Universal Controller with Multi-Modal Feedback for Emergency Response Robots. In NN, . (Eds.), 9th International Conference on Robotics and Automation Engineering (ICRAE 2024). IEEE (Institute of Electrical and Electronics Engineers). ISSN 9798331518301. p. 181–187. doi: 10.1109/ICRAE64368.2024.10851531. Full text in Research Archive Show summary

  This work presents the development of a universal wireless controller that supports the Robot Operating System (ROS), with a focus on enhancing system reliability and safety in Human-Robot Teaming (HRT). The controller is designed to provide multi-modal auditory, visual, and tactile feedback mechanisms, ensuring robust and reliable communication between human operators and various robots during collaborative emergency scenarios. The primary objective is to create an intuitive and unified interface that not only facilitates efficient collaboration but also significantly enhances the safety of operations by minimising the risk of human errors in dynamic and unpredictable environments. By integrating ROS capabilities with diverse sensorial feedback modalities, the controller improves situational awareness (SA) and operational reliability, enabling human operators, such as first responders, to maintain control and makeinformed decisions under pressure. A use case involving the navigation of a quadruped robot through an obstacle-filled environment is presented to demonstrate the controller’s practical benefits in ensuring safety and reliability during complex tasks. The sequence of operations, including safe navigation, reliable victim proximity indication, and secure emergency stops, highlights the controller’s effectiveness in real-world scenarios, emphasising its role in promoting both system reliability and operational safety.

• Zafar, Muhammad Hamza; Langås, Even Falkenberg; Aftab, Muhammad Faisal & Sanfilippo, Filippo (2024). Enhanced Intrusion Detection in Robot Operating Systems via Grid Search Based Multi-Head Attention Stacked Convolutional Network. IEEE International Conference on Automation Science and Engineering. ISSN 2161-8070. p. 3880–3885. doi: 10.1109/CASE59546.2024.10711785. Full text in Research Archive
• Khan, Noman Mujeeb; Khan, Umer Amir; Asif, Mansoor & Zafar, Muhammad Hamza (2024). Analysis of deep learning models for estimation of MPP and extraction of maximum power from hybrid PV-TEG: A step towards cleaner energy production. Energy Reports. 11, p. 4759–4775. doi: 10.1016/j.egyr.2024.04.035. Full text in Research Archive
• Zafar, Muhammad Hamza; Moosavi, Syed Kumayl Raza & Sanfilippo, Filippo (2024). Enhancing unmanned ground vehicle performance in SAR operations: integrated gesture-control and deep learning framework for optimised victim detection. Frontiers in Robotics and AI. 11, p. 01–16. doi: 10.3389/frobt.2024.1356345. Full text in Research Archive
• Ahmad, Nisar; Yi, Xu; Tayyab, Muhammad; Zafar, Muhammad Hamza & Akhtar, Naureen (2024). Water resource management and flood mitigation: hybrid decomposition EMD-ANN model study under climate change. Sustainable Water Resources Management. ISSN 2363-5037. 10(2). doi: 10.1007/s40899-024-01048-9. Full text in Research Archive
• Mansoor, Majad; Houran, Mohamad Abou; Al-Tawalbeh, Nedaa; Zafar, Muhammad Hamza & Akhtar, Naureen (2024). Thermoelectric power generation system intelligent Runge Kutta control: A performance analysis using processor in loop testing. Energy Conversion and Management: X. 23. doi: 10.1016/j.ecmx.2024.100612. Full text in Research Archive
• Moosavi, Syed Kumayl Raza; Zafar, Muhammad Hamza & Sanfilippo, Filippo (2024). Collaborative robots (cobots) for disaster risk resilience: a framework for swarm of snake robots in delivering first aid in emergency situations. Frontiers in Robotics and AI. 11, p. 1–12. doi: 10.3389/frobt.2024.1362294. Full text in Research Archive
• Zafar, Muhammad Hamza; Khan, Noman Mujeeb; Houran, Mohamad Abou; Mansoor, Majad; Akhtar, Naureen & Sanfilippo, Filippo (2024). A novel hybrid deep learning model for accurate state of charge estimation of Li-Ion batteries for electric vehicles under high and low temperature. Energy. ISSN 0360-5442. 292, p. 1–22. doi: 10.1016/j.energy.2024.130584. Full text in Research Archive
• Bukhari, Syed Muhammad Salman; Zafar, Muhammad Hamza; Moosavi, Syed Kumayl Raza; Mansoor, Majad & Sanfilippo, Filippo (2024). An integrated stacked convolutional neural network and the levy flight-based grasshopper optimization algorithm for predicting heart disease. Healthcare Analytics. 7. doi: 10.1016/j.health.2024.100374. Full text in Research Archive Show summary

  Cardiovascular disease is the leading cause of death worldwide, including critical conditions such as blood vessel blockage, heart failure, and stroke. Accurate and early prediction of heart disease remains a significant challenge due to the complexity of symptoms and the variability of contributing factors. This study proposes a novel hybrid model integrating a Stacked Convolutional Neural Network (SCNN) with the Levy Flight-based Grasshopper Optimization Algorithm (LFGOA) to address this challenge. The SCNN provides robust feature extraction, while LFGOA enhances the model by optimizing hyperparameters, improving classification accuracy, and reducing overfitting. The proposed approach is evaluated using four publicly available heart disease datasets, each representing diverse clinical and demographic features. Compared to traditional classifiers, including Regression Trees, Support Vector Machine, Logistic Regression, K-Nearest Neighbors, and standard Neural Networks, the SCNN-LFGOA consistently outperforms these methods. The results highlight that the SCNN-LFGOA achieves an average accuracy of 99%, with significant improvements in specificity, sensitivity, and F1-Score, showcasing its adaptability and robustness across datasets. This study highlights the SCNN-LFGOA’s potential as a transformative tool for early and accurate heart disease prediction, contributing to improved patient outcomes and more efficient healthcare resource utilization. By combining deep learning with an advanced optimization technique, this research introduces a scalable and effective solution to a critical healthcare problem.

• Langås, Even Falkenberg; Zafar, Muhammad Hamza; Nyberg, Svein Olav Glesaaen & Sanfilippo, Filippo (2024). Human Trajectory Simulation in Industrial Settings Using the Ornstein-Uhlenbeck Process and Deep Learning Based Classification. In NN, . (Eds.), International Conference on Automation, Robotics and Applications (ICARA). Institute of Electrical and Electronics Engineers (IEEE). ISSN 9798350394252. p. 427–432. doi: 10.1109/ICARA60736.2024.10553211. Full text in Research Archive Show summary

  This paper presents a novel method of simulating human trajectories using the Ornstein-Uhlenbeck (OU) process in addition to deep learning (DL) based classification. The OU process is a stochastic process and is used in this paper to simulate the movement of a person on a typical factory floor. This work aims at developing systems that increase machines' awareness of people and make predictions about their behaviour to improve efficiency and safety in industrial settings. Sequences of simulated 2D coordinates of people moving on the factory floor are generated. Successively, these synthetic data are used to classify the path that the human is following, using a stacked long short-term memory (LSTM) network and a stacked bidirectional LSTM (BiLSTM) network. The results from this study suggest that, for such applications, it should be possible to predict future movements in 2D for human-robot collaboration (HRC) and teaming (HRT).

• Zafar, Muhammad Hamza; Langås, Even Falkenberg & Sanfilippo, Filippo (2024). Real-Time Gesture-Based Control of a Quadruped Robot Using a Stacked Convolutional Bi-Long Short-Term Memory (Bi-LSTM) Neural Network. In NN, . (Eds.), International Conference on Automation, Robotics and Applications (ICARA). Institute of Electrical and Electronics Engineers (IEEE). ISSN 9798350394252. p. 81–86. doi: 10.1109/ICARA60736.2024.10553163. Full text in Research Archive
• Zafar, Muhammad Hamza; Langås, Even Falkenberg; Nyberg, Svein Olav Glesaaen & Sanfilippo, Filippo (2024). Multimodal Fusion of EEG and EMG Signals Using Self-Attention Multi-Temporal Convolutional Neural Networks for Enhanced Hand Gesture Recognition in Rehabilitation, International Conference on Omni-layer Intelligent Systems (COINS). IEEE (Institute of Electrical and Electronics Engineers). ISSN 9798350349597. p. 245–249. doi: 10.1109/COINS61597.2024.10622144. Full text in Research Archive Show summary

  In this work, we introduce an innovative approach to hand gesture recognition aimed at rehabilitation applications, utilising the synergistic potential of multimodal data fusion from electroencephalogram (EEG) and electromyogram (EMG) sensors. Our approach exploits the strength of Self-Attention Multi-Temporal Convolutional Networks (SAMTCN), which adeptly combine the distinct and complementary insights provided by EEG and EMG signals. The core of our methodology is the strategic application of self-attention mechanisms with multi-temporal convolutional architectures. This design choice allows our model to capture and analyse temporal patterns in multimodal data with unprecedented precision, significantly enhancing its ability to generalise to new, unseen data. The effectiveness of our approach is evidenced by the model's exceptional performance, achieving an accuracy of over 97% in recognising diverse hand gestures. This high level of accuracy highlights the model's potential to revolutionise how interactions are facilitated in rehabilitation contexts.

• Langås, Even Falkenberg; Bukhari, Halima Zahra; Hagen, Daniel; Zafar, Muhammad Hamza & Sanfilippo, Filippo (2024). Inclusive Digital Twins with Edge Computing, Cloud Communication and Virtual Reality to Achieve Remote Human-Robot Interaction. In NN, . (Eds.), 12th International Conference on Control, Mechatronics and Automation (ICCMA). IEEE (Institute of Electrical and Electronics Engineers). ISSN 9798331517519. p. 323–329. doi: 10.1109/ICCMA63715.2024.10843896. Full text in Research Archive Show summary

  Digital twins, advanced robotics, edge computing, and immersive technologies have come together to create novel solutions that increase flexibility and operating efficiency. This work is motivated by the need to harness these advancements to develop a robust architecture that supports edge intelligence and real-time remote monitoring and control. The aim of this work is to define a comprehensive framework for digital twins of complex mechatronic systems. The framework enables seamless connection between the physical environment and a virtual representation accessible from remote locations. Key components of the framework include the Robot Operating System 2 (ROS 2) for robot control, Node-RED for data processing and edge communication, a Message Queuing Telemetry Transport (MQTT) broker for cloud-based communication, and a virtual reality (VR) application for immersive interaction. A case study is presented to demonstrate the framework’s capabilities. Through the VR application, users can interact with a digital twin of a mechatronic system consisting of a collaborative robot, a programmable logic controller (PLC), a conveyor belt and numerous sensors. The user interface lets the operator manipulate the robot and monitor sensor data in real-time. Latency is measured to validate the performance of the framework, resulting in a mean latency of around 116 ms.

• Murtaza, Aitzaz Ahmed; Saher, Amina; Zafar, Muhammad Hamza; Moosavi, Syed Kumayl Raza; Aftab, Muhammad Faisal & Sanfilippo, Filippo (2024). Paradigm shift for predictive maintenance and condition monitoring from Industry 4.0 to Industry 5.0: A systematic review, challenges and case study. Results in Engineering (RINENG). 24. doi: 10.1016/j.rineng.2024.102935. Full text in Research Archive
• Zafar, Muhammad Hamza; Moosavi, Syed Kumayl Raza & Sanfilippo, Filippo (2024). Hierarchical Recurrent-Inception Residual Transformer (HRIRT) for Multidimensional Hand Force Estimation Using Force Myography Sensor. IEEE Sensors Letters. 8(9). doi: 10.1109/LSENS.2024.3431433. Full text in Research Archive
• Khan, Muhammad Kamran; Houran, Mohamad Abou; Kauhaniemi, Kimmo; Zafar, Muhammad Hamza; Mansoor, Majad & Rashid, Saad (2024). Efficient state of charge estimation of lithium-ion batteries in electric vehicles using evolutionary intelligence-assisted GLA–CNN–Bi-LSTM deep learning model. Heliyon. 10(15). doi: 10.1016/j.heliyon.2024.e35183. Full text in Research Archive
• Bukhari, Syed Muhammad Salman; Zafar, Muhammad Hamza; Moosavi, Syed Kumayl Raza; Khan, Noman Mujeeb & Sanfilippo, Filippo (2024). FireNet: A Hybrid Deep Learning Approach for Enhanced Fire Detection in Remote Sensing Imagery. In NN, NN (Eds.), Intelligent Systems Conference (IntelliSys). Springer. ISSN 9783031663291. p. 1–15. doi: 10.1007/978-3-031-66329-1_1. Full text in Research Archive Show summary

  In this study, we address the critical task of early bushfire detection in remote areas through UAV imaging, a vital effort for protecting inhabitants, infrastructure, and ecosystems within the smart cities paradigm. We introduce FireNet, a deep learning framework that employs a hybrid Inception Residual Neural Network (Inception-ResNet) model, specifically designed for the rapid and precise classification of fire and non-fire regions in real-time image processing scenarios. FireNet synergises the strengths of both Inception and ResNet architectures, enhancing the model’s feature extraction capabilities while significantly reducing computational overhead. A distinctive aspect of FireNet is the adoption of the HardSwish activation function, demonstrating superior performance over the conventional Rectified Linear Unit (ReLU) in our fire detection cases. Through rigorous evaluation of a robust dataset, FireNet achieves an impressive accuracy of 96%, with a corresponding AUC of approximately 96%. These results not only affirm FireNet’s efficacy in accurately identifying fire and non-fire areas but also underscore its importance as a crucial tool for real-time fire monitoring and rescue operations, especially in the context of smart cities. With its outstanding efficiency and accuracy, FireNet represents a significant leap forward in the domain of fire detection technologies, highlighting the role of deep learning in enhancing urban resilience against fires.

• Bukhari, Syed Muhammad Salman; Zafar, Muhammad Hamza; Moosavi, Syed Kumayl Raza; Khan, Noman Mujeeb & Sanfilippo, Filippo (2024). Enhancing Alzheimer’s Disease Detection and Classification Through Federated Learning-Optimized Deep Convolutional Neural Networks on MRI Data. In NN, NN (Eds.), Intelligent Systems Conference (IntelliSys). Springer. ISSN 9783031663291. p. 693–712. doi: 10.1007/978-3-031-66428-1_44. Full text in Research Archive Show summary

  Early and accurate detection of Alzheimer’s Disease (AD) is vital for effective management and intervention. However, the subtle nuances of AD progression and the limited availability of comprehensive datasets pose significant diagnostic challenges. Our study introduces an innovative Deep Convolutional Neural Network (DCNN) model, optimised using Federated Learning (FL) for AD detection and staging from MRI images. This approach respects data privacy by enabling decentralised, multi-institutional model training, and demonstrates exceptional adaptability and precision. On the Open Access Series of Imaging Studies (OASIS) database, our FL-DCNN model achieved a remarkable 97% accuracy rate, suggesting significant potential for enhancing AD diagnostics and patient care.

• Moosavi, Syed Kumayl Raza; Sanfilippo, Filippo & Zafar, Muhammad Hamza (2024). Collaborative robots (cobots) for emergency situations: a snake robot as a team member for delivering first aid in emergency situations. In Bui, Tung X. (Eds.), Proceedings of the 57th Hawaii International Conference on System Sciences. The Hawaii International Conference on System Sciences (HICSS). ISSN 9780998133171. p. 443–452. Full text in Research Archive
• Sanfilippo, Filippo; Økter, Martin Bjaadal; Dale, Jørgen; Hua, Minh Tuan; Zafar, Muhammad Hamza & Ottestad, Morten (2024). Open-source design of low-cost sensorised elastic actuators for collaborative prosthetics and orthotics. HardwareX. 19. doi: 10.1016/j.ohx.2024.e00564. Full text in Research Archive
• ElRobrini, Feria; Bukhari, Syed Muhammad Salman; Zafar, Muhammad Hamza; Al-Tawalbeh, Nedaa; Akhtar, Naureen & Sanfilippo, Filippo (2024). Federated learning and non-federated learning based power forecasting of photovoltaic/wind power energy systems: A systematic review. Energy and AI. 18, p. 1–21. doi: 10.1016/j.egyai.2024.100438. Full text in Research Archive Show summary

  Renewable energy sources, particularly photovoltaic and wind power, are essential in meeting global energy demands while minimising environmental impact. Accurate photovoltaic (PV) and wind power (WP) forecasting is crucial for effective grid management and sustainable energy integration. However, traditional forecasting methods encounter challenges such as data privacy, centralised processing, and data sharing, particularly with dispersed data sources. This review paper thoroughly examines the necessity of forecasting models, methodologies, and data integrity, with a keen eye on the evolving landscape of Federated Learning (FL) in PV and WP forecasting. Commencing with an introduction highlighting the significance of forecasting models in optimising renewable energy resource utilisation, the paper delves into various forecasting techniques and emphasises the critical need for data integrity and security. A comprehensive overview of non-Federated Learning-based PV and WP forecasting is presented based on high-quality journals, followed by in-depth discussions on specific non-Federated Learning approaches for each power source. The paper subsequently introduces FL and its variants, including Horizontal, Vertical, Transfer, Cross-Device, and Cross-Silo FL, highlighting the crucial role of encryption mechanisms and addressing associated challenges. Furthermore, drawing on extensive investigations of numerous pertinent articles, the paper outlines the innovative horizon of FL-based PV and wind power forecasting, offering insights into FL-based methodologies and concluding with observations drawn from this frontier. This review synthesises critical knowledge about PV and WP forecasting, leveraging the emerging paradigm of FL. Ultimately, this work contributes to the advancement of renewable energy integration and the optimisation of power grid management sustainably and securely.

• Moosavi, Syed Kumayl Raza; Zafar, Muhammad Hamza; Saadat, Ahsan; Abaid, Zainab; Ni, Wei & Jamalipour, Abbas [Show all 7 contributors for this article] (2024). Transductive Transfer Learning-Assisted Hybrid Deep Learning Model for Accurate State of Charge Estimation of Li-Ion Batteries in Electric Vehicles. IEEE Transactions on Intelligent Transportation Systems (T-ITS). ISSN 1524-9050. 25(10), p. 14445–14459. doi: 10.1109/TITS.2024.3403518. Full text in Research Archive
• Bukhari, Syed Muhammad Salman; Zafar, Muhammad Hamza; Houran, Mohamad Abou; Moosavi, Syed Kumayl Raza; Mansoor, Majad & Muaaz, Muhammad [Show all 7 contributors for this article] (2024). Secure and privacy-preserving intrusion detection in wireless sensor networks: Federated learning with SCNN-Bi-LSTM for enhanced reliability. Ad hoc networks. ISSN 1570-8705. 155. doi: 10.1016/j.adhoc.2024.103407. Full text in Research Archive
• Ibrahim, AL-Wesabi; Farh, Hassan M. Hussein; Fang, Zhijian; Al-Shamma'a, Abdullrahman A.; Xu, Jiazhu & Alaql, Fahad [Show all 8 contributors for this article] (2024). A comprehensive comparison of advanced metaheuristic photovoltaic maximum power tracking algorithms during dynamic and static environmental conditions. Heliyon. 10(18). doi: 10.1016/j.heliyon.2024.e37458. Full text in Research Archive
• Das, Opy; Zafar, Muhammad Hamza; Sanfilippo, Filippo; Rudra, Souman & Kolhe, Mohan Lal (2024). Advancements in digital twin technology and machine learning for energy systems: A comprehensive review of applications in smart grids, renewable energy, and electric vehicle optimisation. Energy Conversion and Management: X. 24. doi: 10.1016/j.ecmx.2024.100715. Full text in Research Archive
• Khan, Muhammad Kamran; Zafar, Muhammad Hamza; Riaz, Talha; Mansoor, Majad & Akhtar, Naureen (2024). Enhancing efficient solar energy harvesting: A process-in-loop investigation of MPPT control with a novel stochastic algorithm. Energy Conversion and Management: X. 21, p. 1–17. doi: 10.1016/j.ecmx.2023.100509. Full text in Research Archive Show summary

  PV systems currently generate 4% of the world's energy needs, and their share is growing quickly. The maximum power point tracking (MPPT) is a complex non-convex optimization problem because the electrical characteristics of the PV model are nonlinear. Changes in temperature, partial shading (PS), and irradiance levels can all affect the amount of power that can be extracted from the solar system. Therefore, in this work, a novel energy valley optimizer (EVO) based MPPT algorithm is suggested to extract maximum power from solar. The classical perturb and observe (P&O), whale optimizer algorithm (WOA), cuckoo search algorithm (CSA), and particle swarm optimization (PSO) algorithms are all compared to EVO. Five case studies, including a field atmospheric data study, partial shading, variable temperature, and irradiance, are used to conduct in-depth analytical and statistical analysis. Furthermore, the successful verification of the MPPT control algorithm on the real microcontroller (Arduino MKRZERO board) through the PIL test is a critical milestone in this research. Quantitative, comparative, statistical and experimental results indicate that the proposed EVO-based MPPT achieves superior performance through 30% quicker tracking time and 80% faster settling time, which result in 4–8% higher power efficiency. The results indicate that the suggested MPPT controller successfully addresses the shortcomings of the current MPPT methods.

• Bukhari, Syed Muhammad Salman; Zafar, Muhammad Hamza; Houran, Mohamad Abou; Qadir, Zakria; Moosavi, Syed Kumayl Raza & Sanfilippo, Filippo (2024). Enhancing cybersecurity in Edge IIoT networks: An asynchronous federated learning approach with a deep hybrid detection model. Internet of Things: Engineering Cyber Physical Human Systems. 27. doi: 10.1016/j.iot.2024.101252. Full text in Research Archive
• Zafar, Muhammad Hamza; Langås, Even Falkenberg & Sanfilippo, Filippo (2024). Exploring the synergies between collaborative robotics, digital twins, augmentation, and industry 5.0 for smart manufacturing: A state-of-the-art review. Robotics and Computer-Integrated Manufacturing. ISSN 0736-5845. 89. doi: 10.1016/j.rcim.2024.102769. Full text in Research Archive
• Lodhi, Ehtisham; Dahmani, Nadia; Bukhari, Syed Muhammad Salman; Gyawali, Sujan; Thapa, Sanjog & Qiu, Lin [Show all 8 contributors for this article] (2024). Enhancing microgrid forecasting accuracy with SAQ-MTCLSTM: A self-adjusting quantized multi-task ConvLSTM for optimized solar power and load demand predictions. Energy Conversion and Management: X. 24. doi: 10.1016/j.ecmx.2024.100767. Full text in Research Archive
• Al-Tawalbeh, Nedaa; Zafar, Muhammad Hamza; Radzi, Mohd Amran Mohd; Zainuri, Muhammad Ammirrul Atiqi Mohd & Al-Wesabi, Ibrahim (2024). Novel initialization strategy: Optimizing conventional algorithms for global maximum power point tracking. Results in Engineering (RINENG). 22. doi: 10.1016/j.rineng.2024.102067. Full text in Research Archive
• Moosavi, Syed Kumayl Raza; Zafar, Muhammad Hamza; Sanfilippo, Filippo; Akhter, Malik Naveed & Hadi, Shahzaib Farooq (2023). Early Mental Stress Detection Using Q-Learning Embedded Starling Murmuration Optimiser-Based Deep Learning Model. IEEE Access. 11, p. 116860–116878. doi: 10.1109/ACCESS.2023.3326129. Full text in Research Archive Show summary

  Stress affects individual of all ages as a regular part of life, but excessive and chronic stress can lead to physical and mental health problems, decreased productivity, and reduced quality of life. By identifying stress at an early stage, individuals can take steps to manage it effectively and improve their overall well-being. Feature selection is a critical aspect of early stress detection because it helps identify the most relevant and informative features that can differentiate between stressed and non-stressed individuals. This paper firstly proposes a variance based feature selection technique that uses q-learning embedded Starling Murmuration Optimiser (QLESMO) to choose relevant features from a publicly available dataset in which stresses experienced by nurses working during the Covid’19 Pandemic is recorded using bio-signals and user surveys. Furthermore, a comparative study with other metaheuristic based feature selection techniques have been demonstrated. Next, to evaluate the efficacy of the proposed algorithm, 10 benchmark test functions have been used. The reduced feature subset is then classified through a 1D convolutional neural network (CNN) model (QLESMO-CNN) and is seen to perform well in terms of the evaluation metrics in comparison to other competitive algorithms. Finally, the proposed technique is compared with the State-of-the-Art methodologies present in literature. The experiments provide a strong basis to determine features that are most relevant for early mental stress classification using a hybrid model combining CNN, Reinforcement Learning and metaheuristic algorithms.

• Zafar, Muhammad Hamza; Mansoor, Majad; Houran, Mohamad Abou; Khan, Noman Mujeeb; Khan, Kamran & Moosavi, Syed Kumayl Raza [Show all 7 contributors for this article] (2023). Hybrid deep learning model for efficient state of charge estimation of Li-ion batteries in electric vehicles. Energy. ISSN 0360-5442. 282. doi: 10.1016/j.energy.2023.128317. Full text in Research Archive Show summary

  State of charge (SoC) estimation is critical for the safe and efficient operation of electric vehicles (EVs). This work proposes a hybrid multi-layer deep neural network (HMDNN)-based approach for SoC estimation in EVs. This HMDNN uses Mountain Gazelle Optimizer (MGO) as a training algorithm for the deep neural network. Our method leverages the intrinsic relationship between the SoC and the voltage/current measurements of the EV battery to accurately estimate the SoC in real time. We evaluate our approach on a large dataset of real-world EV charging data and demonstrate its effectiveness in comparison to traditional SoC estimation methods. Four diverse Li-ion battery datasets of electric vehicles are employed which are the dynamic stress test (DST), Beijing dynamic stress test (BJDST), federal urban driving schedule (FUDS), and highway driving schedule (US06) with different temperatures of . The comparison is made with Mayfly Optimization Algorithm based DNN, Particle Swarm Optimization based DNN and Back-Propagation based DNN. The evaluation indices used are normalized mean square error (NMSE), root mean square error (RMSE), mean absolute error (MAE), and relative error (RE). The proposed algorithm achieves 0.1% NMSE and 0.3% RMSE on average on all datasets, which validates the effective performance of the proposed model. The results show that the proposed neural network-based approach can achieve higher accuracy and faster convergence than existing methods. This can enable more efficient EV operation and improved battery life.

• Zafar, Muhammad Hamza; Bukhari, Syed Muhammad Salman; Houran, Mohamad Abou; Moosavi, Syed Kumayl Raza; Mansoor, Majad & Al-Tawalbeh, Nedaa [Show all 7 contributors for this article] (2023). Step towards secure and reliable smart grids in Industry 5.0: A federated learning assisted hybrid deep learning model for electricity theft detection using smart meters. Energy Reports. 10, p. 3001–3019. doi: 10.1016/j.egyr.2023.09.100. Full text in Research Archive Show summary

  The integration of Smart Grid technology and conceptual Industry 5.0 has paved the way for advanced energy management systems that enhance efficiency and revolutionized the parallel integration of power sources in a sustainable manner. However, this digitization has opened a new stream of the threat and opportunities of electricity theft posing a significant challenge to the security and reliability of Smart Grid networks. In this paper, we propose a secure and reliable theft detection technique using deep federated learning (FL) mechanism. The technique leverages the collaborative power of FL to train a Convolutional Gated Recurrent Unit (ConvGRU) model on distributed data sources without compromising data privacy. The training deep learning model backbone consists of a ConvGRU model that combines convolutional and gated recurrent units to capture spatial and temporal patterns in electricity consumption data. An improvised preprocessing mechanism and hyperparameter tuning is done to facilitate FL mechanism. The halving randomized search algorithm is used for hyperparameters tuning of the ConvGRU model. The impact of hyperparameters involved in the ConvGRU model such as number of layers, filters, kernel size, activation function, pooling, GRU layers, hidden state dimension, learning rate, and the dropout rate is elaborated. The proposed technique achieves promising results, with high accuracy, precision, recall, and F1 score, demonstrating its efficacy in detecting electricity theft in Smart Grid networks. Comparative analysis with existing techniques reveal the superior performance of the deep FL-based ConvGRU model. The findings highlight the potential of this approach in enhancing the security and efficiency of Smart Grid systems while preserving data privacy.

• Mohyuddin, Hassan; Moosavi, Syed Kumayl Raza; Zafar, Muhammad Hamza & Sanfilippo, Filippo (2023). A comprehensive framework for hand gesture recognition using hybrid-metaheuristic algorithms and deep learning models. Array. 19. doi: 10.1016/j.array.2023.100317. Full text in Research Archive Show summary

  This paper presents a novel methodology that utilizes gesture recognition data, which are collected with a Leap Motion Controller (LMC), in tandem with the Spotted Hyena-based Chimp Optimization Algorithm (SSC) for feature selection and training of deep neural networks (DNNs). An expansive tabular database was created using the LMC for eight distinct gestures and the SSC algorithm was used for discerning and selecting salient features. This refined feature subset is then utilized in the subsequent training of a DNN model. A comprehensive comparative analysis is conducted to evaluate the performance of the SSC algorithm in comparison with established optimization techniques, such as Particle Swarm Optimization(PSO), Grey Wolf Optimizer(GWO), and Sine Cosine Algorithm(SCA), specifically in the context of feature selection. The empirical findings decisively establish the efficacy of the SSC algorithm, consistently achieving a high accuracy rate of 98% in the domain of gesture recognition tasks. The feature selection approach proposed emphasizes its intrinsic capacity to enhance not only the accuracy of gesture recognition systems and its wider suitability across diverse domains that require sophisticated feature extraction techniques.

• Bukhari, Syed Muhammad Salman; Moosavi, Syed Kumayl Raza; Zafar, Muhammad Hamza; Mansoor, Majad; Mohyuddin, Hassan & Ullah, Syed Sajid [Show all 8 contributors for this article] (2023). Federated transfer learning with orchard-optimized Conv-SGRU: A novel approach to secure and accurate photovoltaic power forecasting. Renewable Energy Focus. ISSN 1755-0084. 48. doi: 10.1016/j.ref.2023.100520. Full text in Research Archive Show summary

  Accurate photovoltaic (PV) power forecasting is pivotal for optimizing the integration of RES into the grid and guaranteeing proficient energy management. Concurrently, the sensitive nature of data obtained from individual PV systems underscores paramount concerns regarding data privacy and security. In this manuscript, we introduce an innovative approach for PV power forecasting that addresses these concerns, deploying federated learning (FL) combined with TL. This is orchestrated via a hybrid deep learning model, denominated as Federated transfer learning (TL) Convolutional Neural Network with Stacked Gated Recurrent Unit (FL-TL-Conv-SGRU). To optimize the performance of the Conv-SGRU model, we employ the OA for hyperparameter tuning, a novel bio-inspired technique inspired by orchard gardening practices. This algorithm presents a distinctive interplay between exploration and exploitation in the hyperparameter space, potentially elevating the model’s performance. Our exposition covers eight disparate datasets from PV systems, which are judiciously split into two cohorts, safeguarding data privacy. Through the prism of FL, we ensure data security by orchestratively training the Conv-SGRU model over distributed datasets. This strategy allows tapping into the shared wisdom across the datasets, all the while ascertaining individual data remains localized, boosting model generalization and predictive prowess. Additionally, TL is invoked to benefit from pre-trained feature representations, facilitating effective knowledge transmission across diverse PV setups with unique characteristics and locales. The put-forth FL-TL-Conv-SGRU design amalgamates the essence of FL, TL, convolutional neural networks, and stacked gated recurrent units. This ensemble aids in deciphering spatial–temporal intricacies intrinsic to PV power generation. Through empirical analyses, we evince that our FL-TL-Conv-SGRU model transcends conventional forecasting paradigms, emphasizing its adeptness in delivering meticulous forecasts over a range of PV installations. Our results accentuate the bifurcated importance of the federated TL framework: a capability for collaborative training with an unwavering commitment to data privacy, and a proficiency in exploiting decentralized data. This strategy is particularly salient given the shifting regulatory milieu centered on data safeguarding and confidentiality. As we transition towards a world more reliant on renewable energy, our proposed stratagem promises to be a cornerstone for efficient, sustainable energy management, heralding a future replete with green energy.

• Khan, Umer Amir; Khan, Noman Mujeeb & Zafar, Muhammad Hamza (2023). Resource efficient PV power forecasting: Transductive transfer learning based hybrid deep learning model for smart grid in Industry 5.0. Energy Conversion and Management: X. 20. doi: 10.1016/j.ecmx.2023.100486. Full text in Research Archive Show summary

  This paper presents an innovative approach for enhancing power output forecasting of Photovoltaic (PV) power plants in dynamic environmental conditions using a Hybrid Deep Learning Model (DLM). The hybrid DLM employs a synergy of Convolutional Neural Network (CNN), Long Short-Term Memory (LSTM) network, and Bidirectional LSTM (Bi-LSTM), effectively capturing spatial and temporal dependencies within weather data crucial for accurate predictions. To optimize the DLM’s performance efficiently, a unique Kepler Optimization Algorithm (KOA) is introduced for hyperparameter tuning, drawing inspiration from Kepler’s laws of planetary motion. By leveraging KOA, the DLM attains optimal hyperparameter configurations, elevating power output prediction precision. Additionally, this study integrates Transductive Transfer Learning (TTL) with the deep learning models to enhance resource efficiency. By leveraging knowledge gained from previously learned tasks, TTL enables the DLM to improve its forecasting capabilities while minimizing resource utilization. Datasets encompassing environmental parameters and PV plant-generated power across diverse sites are employed for DLM training and testing. Three hybrid models, amalgamating KOA, CNN, LSTM, and Bi-LSTM techniques, are introduced and evaluated. Comparative assessment of these models across distinct PV sites yields insightful observations. Performance evaluation, focused on short-term PV power forecasting, underscores the hybrid DLM’s superiority over individual CNN and LSTM models. This hybrid approach achieves remarkable accuracy and resilience in predicting power output under varying weather conditions, showcasing its potential for efficient PV power plant management.

• Moosavi, Syed Kumayl Raza; Zafar, Muhammad Hamza; Mirjalili, Seyedali & Sanfilippo, Filippo (2023). Improved Barnacles Movement Optimizer (IBMO) Algorithm for Engineering Design Problems. In NN, NN (Eds.), Artificial Intelligence and Soft Computing. Springer. ISSN 9783031425059. p. 427–438. doi: 10.1007/978-3-031-42505-9_36. Full text in Research Archive Show summary

  A better understanding of natural behavior modeling in mathematical systems has enabled a new class of stochastic optimization algorithms that can estimate optimal solutions using reasonable computational resources for problems where exact algorithms show poor performance. The position up-dating mechanism in various optimization algorithms utilizes similar chaotic random behavior which impedes the performance of the search for a globally optimum solution in monotonic nonlinear search space. In this work, an approach is proposed that tackles these issues on an already established algorithm; Improved Barnacle Mating Optimization (IBMO) Algorithm, inspired by the movement and mating of Gooseneck Barnacles. The algorithm introduces the mimicry of the movement and mating behavior in nature to model an optimization process. Several benchmark functions are employed to gauge the performance of the proposed optimization technique. Results are compared with several meta-heuristics and conventional optimization algorithms. It is observed that the IBMO algorithm performs generally better and provides a huge potential for solving real-world problems.

• Zafar, Muhammad Hamza; Moosavi, Syed Kumayl Raza & Sanfilippo, Filippo (2023). Inverse Kinematic Modelling of a 3-DOF Robotic Manipulator using Hybrid Deep Learning Models. Procedia CIRP. 120, p. 213–218. doi: 10.1016/j.procir.2023.08.038. Full text in Research Archive Show summary

  As the degrees of freedom (DOF) for a manipulator rise, so does the complexity of inverse kinematic modeling. This research provides an inverse kinematic model mapped with the aid of a Multilayer Deep Neural Network (DNN) trained using a unique meta-heuristic approach, namely the Gannet Optimization Algorithm (GOA), to decrease the computational weight and time lag for desired output transformation. The suggested design can automatically pick up on the kinematic characteristics of the manipulator. The sole observational basis for repeated learning is the link between input and output. Using the Robot Operating System (ROS), related simulations on a 3-DOF manipulator are performed. The simulation-generated dataset is split 65:35 for the purpose of training and testing the suggested model. Cost, time for the training data, mean relative error, normal mean square error, and mean absolute error for the test data are the metrics utilized for model validation. The efficacy and superiority of the suggested method are demonstrated by a comparison of the GOA-DNN model with the particle swarm optimization (PSO)-DNN and Grey Wolf Optimization (GWO)-DNN meta-heuristic DNN models.

• Houran, Mohamad Abou; Bukhari, Syed M. Salman; Zafar, Muhammad Hamza; Mansoor, Majad & Chen, Wenjie (2023). COA-CNN-LSTM: Coati optimization algorithm-based hybrid deep learning model for PV/wind power forecasting in smart grid applications. Applied Energy. ISSN 0306-2619. 349. doi: 10.1016/j.apenergy.2023.121638. Full text in Research Archive Show summary

  Power prediction is now a crucial part of contemporary energy management systems, which is important for the organization and administration of renewable resources. Solar and wind powers are highly dependent upon environmental factors, such as wind speed, temperature, and humidity, making the forecasting problem extremely difficult. The suggested composite model incorporates Long Short-Term Memory (LSTM) and Swarm Intelligence (SI) optimization algorithms to produce a framework that can precisely estimate offshore wind output in the short term, addressing the discrepancies and limits of conventional estimation methods. The Coati optimization algorithm enhances the hyper parameters CNN-LSTM. Optimum hyper parameters improvise learning rate and performance. The day-ahead and hour-ahead short-term predictions RMSE can be decreased by 0.5% and 5.8%, respectively. Compared to GWO-CNN-LSTM, LSTM, CNN, and PSO-CNN-LSTM models, the proposed technique achieves an nMAE of 4.6%, RE 27% and nRMSE of 6.2%. COA-CNN-LSTM outperforms existing techniques in terms of the Granger causality test and Nash-Sutcliffe metric analysis for time series forecasting performance, scores are 0.0992 and 0.98, respectively. Experimental results show precise and definitive wind power-making predictions for the management of renewable energy conversion networks. The presented model contributes positively to the body of knowledge and development of clean energy.

• Muqeet, Abdul; Israr, Asif; Zafar, Muhammad Hamza; Mansoor, Majad & Akhtar, Naureen (2023). A novel optimization algorithm based PID controller design for real-time optimization of cutting depth and surface roughness in finish hard turning processes. Results in Engineering (RINENG). 18. doi: 10.1016/j.rineng.2023.101142. Full text in Research Archive Show summary

  This paper proposes a novel method to improve surface finish in turning processes by effectively controlling the cutting depth. A metaheuristic algorithm based PID control system, in combination with a piezoelectric vibration sensor for feedback, is introduced to regulate the position of the servo motor that controls the cutting tool. The PID controller is optimized using Q-learning based Sand Cat Optimization algorithm to achieve the best performance in terms of cutting depth accuracy and surface finish quality. The piezoelectric sensor provides realtime feedback information about the cutting process and allows for precise adjustments to the cutting depth. The results demonstrate the proposed system’s ability to handle variations in cutting conditions and tool’s wear and tear. Compared to highly optimized standard PID control, improved robustness and stability has been achieved in experimental results by proposed framework. Experimental results demonstrate improved robustness and stability compared to standard PID control. Several materials with high hardness of 20–65 HRC including Phenolic Bakelite, Copper, Thermoplastic, and Stainless Steel (AISI-420, AA6061-T6, and AISI-316) are tested. Minimum value of Vibration amplitude achieved is 0.598 μm for cutting depth of 0.25 mm. The sustained minimum amplitude of vibration and Ra value of the surface finish is found comparable to standard models with less than 10% error.

• Zafar, Muhammad Hamza; Langås, Even Falkenberg & Sanfilippo, Filippo (2023). Empowering human-robot interaction using sEMG sensor: Hybrid deep learning model for accurate hand gesture recognition. Results in Engineering (RINENG). 20. doi: 10.1016/j.rineng.2023.101639. Full text in Research Archive Show summary

  In this paper, a novel approach using a Henry Gas Solubility-based Stacked Convolutional Neural Network (HGS-SCNN) for hand gesture recognition using surface electromyography (sEMG) sensors is proposed. The stacked architecture of the CNN model helps to capture both low-level and high-level features, enabling effective representation learning. To begin, we generated a dataset comprising 600 samples of hand gestures. Next, we applied the Discrete Wavelet Transform (DWT) technique to extract features from the filtered sEMG signal. This step allowed us to capture both spatial and frequency information, enhancing the discriminative power of the extracted features. Extensive experiments are conducted to evaluate the performance of the proposed HGS-SCNN model. In addition, the obtained results are compared with state-of-the-art techniques, namely AOA-SCNN, GWO-SCNN, and WOA-SCNN. The comparative analysis demonstrates that the HGS-SCNN outperforms these existing methods, achieving an impressive accuracy of 99.3%. The experimental results validate the effectiveness of our proposed approach in accurately detecting hand gestures. The combination of DWT-based feature extraction and the HGS-SCNN model offers robust and reliable hand gesture recognition, thereby opening new possibilities for intuitive human-machine interaction and applications requiring gesture-based control.

• Hua, Tuan; Langås, Even Falkenberg; Zafar, Muhammad Hamza & Sanfilippo, Filippo (2023). From rigid to hybrid/soft robots: Exploration of ethical and philosophical aspects in shifting from caged robots to human-robot teaming. In Yu, Wen (Eds.), 2023 IEEE Symposium Series on Computational Intelligence (SSCI). IEEE (Institute of Electrical and Electronics Engineers). ISSN 9781665430654. p. 1794–1799. doi: 10.1109/SSCI52147.2023.10372032. Full text in Research Archive Show summary

  This paper delves into the ethical, philosophical, and practical dimensions associated with the transition from caged robots to human-robot teaming (HRT). By exploring the evolving dynamics between humans and robots, this paper examines the ethical challenges, philosophical implications, and practical considerations that arise as collaboration and integration between humans and robots deepen. It emphasises the need for responsible design, implementation, and ethical frameworks to guide the development and deployment of human-robot teams. Particular focus is put into the ethical ramifications of choosing between rigid and soft actuators. The study underscores the significance of employing admittance and impedance control techniques to regulate interaction forces and compliance between humans and robots. By analysing the ethical implications of utilising soft actuators, the paper emphasises the potential advantages, such as enhanced safety and reduced risk of harm during close human-robot collaboration.

• Langås, Even Falkenberg; Zafar, Muhammad Hamza & Sanfilippo, Filippo (2023). Harnessing digital twins for human-robot teaming in industry 5.0: Exploring the ethical and philosophical implications. In Yu, Wen (Eds.), 2023 IEEE Symposium Series on Computational Intelligence (SSCI). IEEE (Institute of Electrical and Electronics Engineers). ISSN 9781665430654. p. 1788–1793. doi: 10.1109/SSCI52147.2023.10372069. Full text in Research Archive Show summary

  In the era of Industry 5.0, the convergence of humans and robots in collaborative work environments has brought forth the concept of digital twins (DTs) of humans and robots. These virtual replicas, mirroring their physical counterparts, have become integral to the design and operation of complex systems. This paper aims to explore the ethical and philosophical implications associated with the design and use of DTs of humans and robots in human-robot collaboration (HRC), and even further in human-robot teaming (HRT). By examining the potential benefits, challenges, and risks, this research seeks to shed light on the responsible development and application of DTs in the context of Industry 5.0.

• Murtaza, Aitzaz Ahmed; Amina, Saher; Mohyuddin, Hassan; Moosavi, Syed Kumayl Raza; Zafar, Muhammad Hamza & Sanfilippo, Filippo (2023). Enhancing Cardiovascular Disease Prediction via Hybrid Deep Learning Architectures: A Step Towards Smart Healthcare. In NN, NN (Eds.), 2nd International Conference on Emerging Trends in Electrical, Control, and Telecommunication Engineering (ETECTE). IEEE (Institute of Electrical and Electronics Engineers). ISSN 9798350305654. p. 1–6. doi: 10.1109/ETECTE59617.2023.10396716. Full text in Research Archive Show summary

  Cardiovascular disease presents a serious and increasing global health challenge, making a substantial contribution to morbidity and mortality rates on a global scale. This research study presents a novel methodology for predicting Cardiovascular Diseases by employing a, recently developed, metaheuristic optimisation algorithm within a neural network framework. The Coati Optimisation Algorithm (COA) is employed in an artificial neural network (ANN) to enhance the predictive accuracy of outcomes related to Cardiovascular Diseases. The enhanced performance of the COA can be ascribed to its adept utilisation of both exploration and exploitation phenomena. This research employs publicly available datasets pertaining to heart and stroke disorders, integrating two datasets focused on heart disease and one dataset focused on stroke disease. A comparison analysis is undertaken between the proposed COA-ANN and existing approaches, namely Particle Swarm Optimizer based ANN (PSO-ANN), Grey Wolf Optimizer based ANN (GWO-ANN), and backpropagation based ANN (BP-ANN). The findings of the study indicate that the COA-ANN model exhibits the highest level of predictive accuracy. The COAANN outperformed the other three networks, namely GWOANN, PSO-ANN, and BP-ANN, with an average accuracy of 98.43%. As a result, the utilisation of the COA-ANN leads to an improvement in predictive accuracy for these datasets, with an increase of up to 2.64%. Additional assessment metrics, such as F1-Score, Precision, and Recall, provide more insight into the balanced performance of the COA-ANN architecture when applied to imbalanced class datasets. These results prove that the integration of nature-inspired algorithms with cardiovascular diseases (CVDs) is a promising direction for future research.

• Zafar, Muhammad Hamza; Khan, Noman Mujeeb; Mansoor, Majad & Sanfilippo, Filippo (2023). Optimal Tuning of PID Controller for Boost Converter using Meta-Heuristic Algorithm for Renewable Energy Applications. In NN, NN (Eds.), International Conference on Mechanical, Automotive and Mechatronics Engineering (ICMAME 2023). ICMAME. ISSN 9786250015261. p. 1–6. doi: 10.53375/icmame.2023.32. Full text in Research Archive Show summary

  The Dynamic Levy Flight Chimp optimisation (DLFC) method is used in this study to optimise the Proportional- Integral-Derivative (PID) Controller for the Boost converter. As a possible application, the tuned PID controller is utilised to adjust voltages in the use of renewable power sources. The maximum power point tracking control approach based on machine learning (ML) is used to anticipate the reference voltages for the solar system based on the irradiance and the ambient temperature. The tuned PID controller uses this reference signal to regulate the maximum power point (MPP) voltages. To finetune the PID controller, comparisons are done with grey wolf optimiser (GWO), Harris hawk optimisation algorithms (HHO), and particle swarm optimisation (PSO) algorithms. The tuned PID controller has fewer oscillations and requires little tracking time to adapt to changing load and environment conditions. Additionally, statistical analysis, such as Mean Absolute Error (MAE) and Root Mean Square Error (RMSE) between the reference voltage and the output voltage, is presented. Since the DLFC tuned PID controller performs better than HHO, GWO, and PSO in terms of RMSE and MAE, it may be a promising way for optimising PID controller tuning for boost converters in photovoltaic (PV) system applications.

• Zafar, Muhammad Hamza; Younus, Hassaan Bin; Moosavi, Syed Kumayl Raza; Mansoor, Majad & Sanfilippo, Filippo (2023). Online PID Tuning of a 3-DoF Robotic Arm Using a Metaheuristic Optimisation Algorithm: A Comparative Analysis. In NA, NA (Eds.), Communications in Computer and Information Science. Springer. ISSN 9783031489815. p. 25–37. doi: 10.1007/978-3-031-48981-5_3. Full text in Research Archive
• Zafar, Muhammad Hamza; Sanfilippo, Filippo & Blažauskas, Tomas (2023). Harmony unleashed: Exploring the ethical and philosophical aspects of machine learning in human-robot collaboration for industry 5.0. In Yu, Wen (Eds.), 2023 IEEE Symposium Series on Computational Intelligence (SSCI). IEEE (Institute of Electrical and Electronics Engineers). ISSN 9781665430654. p. 1775–1780. doi: 10.1109/SSCI52147.2023.10371798. Full text in Research Archive Show summary

  As Industry 5.0 emerges by blending advanced technologies with human-centered approaches, the integration of machine learning (ML) in human-robot collaboration (HRC) becomes increasingly prominent. This paper explores the philosophy and ethics underlying the application of machine learning in Industry 5.0, specifically focusing on HRC. It examines the ethical considerations, philosophical implications, and potential challenges that arise in this evolving paradigm. The paper emphasises the need for a thoughtful and ethical approach to ensure the beneficial and responsible use of ML in Industry 5.0.

View all works in NVA

• Sanfilippo, Filippo & Zafar, Muhammad Hamza (2024). Hands-on Collaboration: Building Robot Muscle Memory. IEEE Transmitter. Full text in Research Archive
• Moosavi, Syed Kumayl Raza; Zafar, Muhammad Hamza; Das, Opy & Sanfilippo, Filippo (2024). Human-Robot Teaming for Intelligent Transportation Systems. Full text in Research Archive
• Sanfilippo, Filippo & Zafar, Muhammad Hamza (2024). Representing the Agder region at the AI, Data, Robotics Forum: live demos, posters and presentations. Full text in Research Archive

View all works in NVA