Keywords: Nonlinear Systems
Abstract: This talk proposes an overview on the use of a class of sign-indefinite quadratic Lyapunov functions comprising an extended quadratic form. We will first discuss stability/performance analysis results certified by these functions in linear saturated feedbacks in both local or global stability contexts.The direct design of linear plant-order saturated output-feedback controllers will be presented. Furthermore, we will then illustrate how the technique can be extended to deal with different nonlinearities and to handle other kinds of problems as for example the contraction problem or the way to deal with optimization algorithms.

Keywords: Computer Vision, Neural Networks, Machine Learning
Abstract: This study investigates the trade-off between segmentation accuracy and interpretability in chest X-ray lung segmentation using attention-augmented U-Net models. We introduce a novel Grad-CAM Coherence Score to quantify spatial alignment between segmentation masks and model attention maps. We compare three variants: a baseline U-Net without attention, a partially attentive model (UNet-Att1), and a fully attentive model (UNet-AttAll). All models incorporate Grad-CAM to assess interpretability. Trained on combined public datasets (Montgomery, Shenzhen, Darwin), the baseline model achieved the highest segmentation performance (Dice 0.9514, IoU 0.9070, coherence 0.9056). UNet-AttAll offered a strong balance between accuracy and interpretability (Dice 0.9246, coherence 0.8695), while UNet-Att1 underperformed across all metrics. While the classical U-Net achieves the highest Dice score (0.9514), our fully attentive variant (UNet-AttAll) demonstrates superior interpretability and a competitive Dice score (0.9246). Results suggest that deep attention integration improves model transparency without severely compromising performance, supporting the development of clinically trustworthy AI systems. These findings highlight the value of deep attention integration in medical image segmentation.

Keywords: Computer Vision, Machine Learning, Software Methods and Tools
Abstract: In this paper, we introduce a general-purpose framework for collecting large-scale datasets from computer games to support autonomous driving research. Unlike existing approaches that rely on direct memory access or API hooks, our method operates entirely at the user interface level, enabling non-invasive and highly adaptable data acquisition across different game environments. As a case study, we implement the pipeline on Grand Theft Auto V using the RAGE Multiplayer platform. Our system captures three key modalities - road imagery, vehicle speed, and controller inputs - and stores them efficiently using the HDF5 format. To enhance usability, we integrate robust screen capture with GPU acceleration, an SVM-based speedometer digit recognition system resilient to UI variability, and HSV-based route segmentation for improved navigation data. This framework enables the use of closed-source commercial video games for self-driving applications, which are designed for human players rather than automated systems and prohibit access to code or file modifications to mitigate cheating, providing a valuable tool for imitation learning and behavioral cloning experimentation in simulation.

Keywords: Computer Graphics, Computer Vision, Industrial Applications
Abstract: This paper presents a mobile application designed to acquire accurate anthropometric measurements by utilizing LiDAR technology and photogrammetry on iOS devices. The pipeline involves data acquisition through a mobile app, initial processing, server-side post-processing using a FastAPI backend, and final storage of measurements and metadata in a cloud database. The integration of LiDAR scanning and photogrammetry allows for the generation of detailed 3D models, which are further processed with the help of a SMPL model to extract key anthropometric measurements. This solution targets the fashion industry, focusing on custom clothing manufacturing and enhanced consumer experiences.

Keywords: Computer Vision, Human - Computer Interaction, Computer Graphics
Abstract: This paper presents a real-time, non-intrusive system for assessing cognitive fatigue and attention levels using webcam-based eye tracking. Leveraging the MediaPipe Face Mesh framework, the system analyzes the Eye Aspect Ratio (EAR) and gaze direction to detect blinking patterns and classify alertness states. An adaptive EAR threshold enhances accuracy across diverse facial structures, while real-time feedback and visual overlays provide intuitive user insights. Data is periodically logged to support retrospective analysis and future research. The solution is lightweight, hardware-agnostic, and applicable in various domains, including education, road safety, and cognitive ergonomics. Experimental validation using attention-focused tasks confirms the system’s reliability and usability in practical scenarios.

Keywords: Neural Networks, Computer Vision, Distributed Systems
Abstract: Human-bear interactions have become increasingly frequent in mountainous areas due to food scarcity in natural habitats. To prevent unpleasant encounters, this paper presents an autonomous multidisciplinary bear deterrent system (MBDS) developed for mountain highway areas. The system combines artificial intelligence detection techniques, remote observation, and remote access and positioning of cameras followed by distributed acoustic dispersion, an important part of this innovative system proposed due to its multiple integrated technologies. The problem of human-bear interactions in this tourist area requires a non-invasive approach that protects both people and wildlife. Using a central visual detection mechanism, the system identifies bear presence, and coordinately activates animal redirection without causing stress or aggressiveness.

Keywords: Computer Graphics, Computer Vision, Software Methods and Tools
Abstract: Transfer functions are essential in direct volume rendering, enabling the mapping of voxel properties to visual attributes for the purpose of revealing meaningful structures. Traditional one- and two-dimensional transfer functions often rely on density, gradient, or density–gradient histograms. While these methods allow for rapid voxel-to-visual mapping, they have limitations and can be difficult to adjust, often requiring domain expertise. We present an alternative approach based on transition histograms, which represent the relationship between voxel densities sampled along the local gradient direction. In this space, homogeneous materials cluster along the main diagonal, while material interfaces form compact, localized off-diagonal regions. This structure enables a more intuitive basis for specifying color and opacity mappings. To demonstrate the utility of this representation, we define transition-based transfer functions using simple rectangular regions placed in transition space. This significantly simplifies the isolation of important structures in volume data and supports a more interpretable and reproducible visualization process. Our work shows that transition-based transfer functions offer a promising direction for improving usability and precision in volume rendering workflows.

Keywords: Internet of Things, Machine Learning, Data - Driven Control
Abstract: The convergence of Artificial Intelligence (AI) and Low-Power Wide-Area Networks (LPWANs) has garnered substantial interest in recent years, particularly for their role in intelligent monitoring and data-centric decision-making. LPWAN technologies such as LoRa, NB-IoT, and Sigfox enable large-scale, low-power communication, making them well-suited for scenarios demanding extended connectivity and minimal power usage. The integration of AI into these networks has the potential to enhance data processing, anomaly detection, and predictive analytics, yet it also introduces challenges related to energy efficiency, computational limitations, security, and real-time performance. This paper offers a structured review of current research on AI-enabled solutions within LPWAN-based IoT frameworks. Our study assesses and categorizes recent research contributions, focusing on AI-enabled data acquisition, transmission protocols, edge and cloud computing frameworks, and machine learning methodologies for intelligent analytics in IoT applications.

Keywords: Smart Cities/Houses/Grids, Machine Learning, Internet of Things
Abstract: Efficient waste collection is a critical aspect of sustainable urban management. This paper presents an IoT-enabled system designed to optimize waste collection routes using real-world data and advanced algorithmic techniques. The proposed solution integrates GPS and RFID technologies to gather precise location data from smart waste bins at the time of emptying. The collected data is processed into distance matrices and optimized using three algorithms: Nearest Neighbor (NN), Google OR-Tools, and a Genetic Algorithm (GA) with various parameter configurations. Experiments were conducted on three real-world waste collection routes covering a total of 735 locations. Results demonstrate that the GA, particularly with a large population running a big number of generations, significantly outperforms other methods in reducing total route distance, fuel consumption, and CO2 emissions. The system achieved reductions of up to 52% in travel distance and over 14 kg in CO2. These findings highlight the practical and environmental benefits of integrating IoT infrastructure with optimization algorithms in municipal waste collection.

Keywords: Internet of Things, Sensor Networks, Smart Cities/Houses/Grids
Abstract: This paper introduces an approach to sustainable IoT deployments through a system, which addresses critical challenges facing the future of environmental monitoring networks. As IoT devices proliferate in urban and remote environments, the twin challenges of energy sustainability and maintenance overhead present significant barriers to widespread adoption. Our solution leverages solar energy harvesting coupled with innovative power management strategies to create truly autonomous environmental sensing nodes with zero carbon footprint and minimal maintenance requirements. By eliminating batteries and optimizing energy consumption patterns, this approach offers a scalable model for future IoT deployments that can operate indefinitely under diverse environmental conditions. The system demonstrates how renewable energy integration at the device level can transform IoT implementations from maintenance-intensive, environmentally-problematic deployments to sustainable, set-and-forget solutions. This study presents ClimateCube, an environmental sensor that measures temperature, humidity, pressure, light intensity and air quality (including eCO2 & TVOC), transmitting wireless data, powered exclusively by solar cells and supercapacitors. The system demonstrated continuous operation for 24 hours using only the energy stored with optimized hibernation consumption without solar power, and achieved a transmission success rate of approximately 98 % while maintaining full autonomy. Experimental validation demonstrated that a 5.0 V supercapacitor charge sustained 30 consecutive transmissions at 30-minute intervals, equivalent to 15 hours of autonomous operation, validating the energy budget calculations and confirming the viability of supercapacitor-based energy storage for IoT.

Keywords: Sensor Networks, Internet of Things
Abstract: This paper explores the technical feasibility of deploying Agentic AI workflows within edge-fog computing architectures for real-time biomedical monitoring. The system is intended to gather heart rate data from a wireless heart rate device located at the edge of the network and process it with an immediate device located at the fog layer using AI methods and machine learning approaches. We have built an artificial neural network model based on the heart rate synthetic data classified into three categories. This paper aims to demonstrate the potential of Agentic AI and Agentic workflows in processing physiological data, specifically heart rate, at the fog layer of a distributed system. In critical situations, the system is capable of independently issuing alerts to caregivers, enabling timely intervention without requiring constant human supervision.

Keywords: Sensor Networks, Cloud Computing, Web services and applications
Abstract: One of the main challenges nowadays in medicine is to prevent health issues as early as possible. This can be done by using different digital tools/software which can monitor a person in real time. Cardiovascular disease is the leading cause of death globally. Identifying in due time the cardiovascular diseases prevent premature death, and the patient may receive appropriate treatment. This paper presents the modular HealthPocket system that has the potential to detect early signs of cardiovascular disease. The HealthPocket system includes a mobile application, a web-based application and an IoT module. To ensure the continuity of care, the system sends medical data of the patient to other medical units, using HL7 FHIR standard (Fast Health Interoperability Resources), ensuring the interoperability between medical units. For analyzing the data received by the IoT module was used Machine Learning techniques which analyses the data in real time, can personalize user experience, optimize the resources and can detect patterns. This type of monitoring the health status of the patient may help improve the wellbeing of the citizens because it can detect anomalies of cardiovascular behavior in time.

Keywords: Internet of Things, Smart Cities/Houses/Grids, Web services and applications
Abstract: Indoor mold growth remains a significant concern in modern buildings, with direct implications for human health and indoor air quality. Despite the availability of environmental monitoring devices, many existing solutions rely on static thresholds applied directly to raw sensor data (e.g., temperature and relative humidity), rather than integrating scientifically validated models for mold risk prediction. This paper introduces MoldSafe, an IoT-based system for real-time mold risk assessment, which transforms the theoretical Viitanen mold growth model (VTT) into a practical application. MoldSafe collects temperature and humidity data using ESP8266 nodes equipped with BME280 sensors, transmits this information via MQTT to a modular back-end infrastructure, and provides a user-friendly interface implemented as a PWA, enabling accessibility across all devices. This data is processed to estimate the risk of mold growth, providing users with early notifications to prevent it. The proposed solution is scalable and efficient due to the use of a modern architecture and serves as a practical foundation for validating the VTT model by facilitating continuous data collection in real-world conditions through the deployment of physical sensors. The system evaluation demonstrated reliable and consistent communication, resource-efficient and scalable back-end processing, and high performance data handling for 50 active sensors.

Keywords: Neural Networks, Internet of Things, Control Applications
Abstract: Air quality has become a topic of growing interest in recent years, especially after the pandemic made people more aware of the environments they live and work in. While many studies focus on air pollution outdoors, the quality of indoor air is not explored as much, even though it affects people just as directly—particularly now, when working from home has become common. In this context, it's not enough just to monitor the air; what really matters is to act in time and try to prevent the quality from dropping too much. One way to do that is by predicting how air quality will evolve, so that systems like ventilation can be turned on earlier, before things get worse. This paper presents an analysis of several neural network models based on time series, trained to predict the indoor air quality index (AQI) 15 minutes ahead. The data used for training was collected over a period of one year and includes the concentrations of eight different gases measured in an enclosed space. Several Nonlinear Autoregressive with eXogenous inputs (NARX) models with different layer configurations were trained using MATLAB, and their prediction results were compared to find out which setup performs best when it comes to forecasting indoor air quality.

Keywords: Computer Graphics, Information Systems Applications, Intelligent Systems
Abstract: Pediatric dental anxiety is an issue which has an important effect on children's oral health and willingness to receive dental care. Traditional ways of reducing fear, such as the "tell-show-do" method, have proven effective, but they sometimes lack engagement and involvement for younger patients. With the increased accessibility of smartphones and augmented reality (AR), new digital approaches are emerging to address this issue in new ways. This paper introduces Tooth Quests, an Android-based mobile application created with the Unity game engine and AR Foundation tool. The application is designed to reduce dental anxiety in children by using gamified learning and immersive mini-games that simulate common dental experiences. It integrates storytelling, visual feedback, and educational content based on pediatric behavioral approaches. A user survey conducted among doctors, parents and children supports the application's relevance and utility, demonstrating that interactive technologies can improve children's perceptions of dental care. The study suggests that digital solutions such as Tooth Quests provide an appealing option for anxiety reduction and oral health education in pediatric dentistry.

Keywords: Information Systems Applications, Human - Computer Interaction, Computer Vision
Abstract: In today’s society, first aid care is often overlooked, despite its crucial role in day-to-day lives. Specialists state that the basic measures regarding first aid, the so-called pre-medical first aid measures, are important for every person to know. These measures should be taught in schools, and knowledge should be refreshed periodically, because information can change, be forgotten or updates might appear. Paramedics suggest that ideally, these refreshers should be done every year [1]. Unfortunately, in 2023 [2], according to the annual reports of the Red Cross, only 0.36% of the Romanian population took first aid training courses. In comparison, Northern European countries state that their citizens are 100% prepared to deliver first aid to persons in need [3]. Our paper presents a possible solution for overcoming this issue – Vital AidVenture, a bespoke system using emergent technologies for teaching first aid methods. By combining virtual reality and a custom-built manikin, Vital AidVenture offers an easy-to-use alternative that can improve delivery of first aid assistance.

Keywords: System Biology, Biomedical Engineering
Abstract: Recent advances in sensors for medical applications enhance data acquisition for medical systems. The authors propose a Digital Twin architecture for glaucoma diagnosis and monitoring that integrates Contact Lens Sensors for real-time monitoring of glaucoma progression. The proposed framework is discussed in regard to ethical aspects associated with the use of Digital Twin technologies in medicine. The use of emerging technologies in medicine is associated with significant ethical challenges. Starting from this premise, an analysis is conducted on data traceability, and several questions are raised regarding the responsible integration of predictive technologies into diagnostic and treatment decision-making processes. However, in order to combine the respect for fundamental ethical principles with the maximization of benefits, this paper proposes a guideline that integrates a set of deontological rules and utilitarianism rules.

Keywords: Control Applications, Control Systems Design, System Identification and Modeling
Abstract: This work is focused on the tuning process and implementation of several controllers for an original Ball-and-Plate (B&P) laboratory prototype process. The developed nonlinear mechatronic B&P system was designed and implemented as a laboratory platform for engineering education, being very easy to use for design and testing various types of control approaches. The low-cost two-degree-of-freedom (2DOF) Ball-and-Plate system uses as feedback sensor a resistive plate and as actuators two servo-motors. The control system is implemented on an Arduino board which allows the use of open source library functions which are easy to customize. The main control objective is to stabilize the ball position on the plates' flat surface and compensate any external disturbance effect. The tuning and the implementation of two controllers is also presented. Testing was performed initially in simulation using Matlab and next on the B&P prototype.

Keywords: Machine/Reinforcement Learning, Signal Processing, Automotive Control Systems
Abstract: This paper presents a real-time EEG signal processing system for detecting driver fatigue, implemented as part of a Brain-Computer Interface (BCI) application. The system leverages EEG signals acquired using an OpenBCI headset, pre-processed through offset correction, band-pass filtering, and smoothing techniques to remove noise and artifacts. Key features, including blink periods and statistical descriptors (mean, standard deviation, kurtosis, and skewness), are extracted from the filtered signals. Two supervised machine learning algorithms—Support Vector Machine (SVM) and Logistic Regression—are employed to classify the signals into normal and abnormal blinking patterns indicative of driver fatigue. The system achieves a classification accuracy exceeding 85%, demonstrating its potential for real-time driver fatigue monitoring in automotive applications. Future work will explore the integration of additional cognitive indicators and data from multiple subjects to enhance the system’s generalizability.

Keywords: Neural Networks, Intelligent Systems, Biomedical Engineering
Abstract: Magnetoencephalography (MEG) offers high-resolution measurements of brain activity using hundreds of sensors, making it a valuable tool for diagnosing neurological disorders. Recent advances in machine learning have enabled the analysis of large-scale MEG datasets, uncovering intricate spatio-temporal patterns. In this study, we propose the use of 3D Convolutional Neural Networks (CNNs) for Parkinson’s disease diagnosis, leveraging compact Fourier feature volumes derived from MEG signals. The 3D architecture progressively expands the receptive fields across both spatial and frequency dimensions, improving feature extraction and classification accuracy. We also investigate alternative approaches using 2D CNNs on spectral data volumes. Unlike their 3D counterparts, these models aggregate frequency channels in the initial convolutional layer and expand receptive fields only across spatial dimensions. Furthermore, we explore hybrid models that combine 2D and 3D CNNs, where 2D CNNs perform initial spatial analysis and 3D CNNs integrate the resulting feature over the whole frequency band. Experiments on the OMEGA dataset show that the proposed methods accurately detect Parkinson’s disease, with 3D CNNs outperforming 2D CNNs due to their enhanced capabilities.

Keywords: Data - Driven Control, Neural Networks, Predictive Control
Abstract: This paper presents an AI-based solution for a data-driven control framework in the field of wastewater treatment, highlighting surrogate modeling and its advantages. To approximate the ammonium (SNHe) concentration in the effluent, a long short-term memory (LSTM) neural network (NN) was trained, using selected reference simulations from Benchmark Simulation Model no. 2 (BSM2) as input variables. To identify the most relevant predictors, recursive feature elimination (RFE) was performed. The surrogate model was first trained on open-loop simulation data and then integrated into a closed-loop control system by generating oxygen transfer rate (KLa) inputs based on SNHe predictions. This data-driven KLa signal was injected into the BSM2 simulation in place of the standard controller. A feedback-based retraining process was used to refine the model’s performance under controlled conditions. Experimental results demonstrate high prediction accuracy, a coefficient of determination (R²) with an average value of 0.974, and a substantial reduction in effluent SNH levels, with SNHe values decreasing from an average of 7.62 mg/L to 0.19 mg/L after control. These findings support the integration of AI-based surrogate models into supervisory control strategies for wastewater treatment plants.

Keywords: Neural Networks, Machine Learning, Control Applications
Abstract: This paper presents a comparative study of data-preprocessing techniques for developing a gated recurrent unit (GRU)–based soft sensor to estimate dissolved oxygen (DO) levels in wastewater treatment plants (WWTPs). Simulation data from the Benchmark Simulation Model No. 2 (BSM2) were used to evaluate three feature-selection methods—SHAP, Boruta, and recursive feature elimination (RFE)—to identify the most informative input variables. Concurrently, three normalization schemes (MinMaxScaler, StandardScaler, and RobustScaler) were assessed for their impact on model accuracy. Each GRU network was trained and tested on datasets preprocessed according to every combination of feature-selection and normalization method. Model robustness was further evaluated under controlled input-noise conditions. Under nominal conditions, the best performance was achieved by pairing StandardScaler normalization with RFE-selected features, yielding the lowest prediction error. Under noisy inputs, models trained with RobustScaler demonstrated superior resilience to perturbations. Finally, the top-performing GRU models were integrated into a MATLAB Simulink environment for practical validation. The results underscore the critical influence of preprocessing choices on soft-sensor accuracy and robustness and confirm the feasibility of GRU networks for reliable DO estimation in wastewater treatment applications.

Keywords: Biomedical Engineering
Abstract: This study analyzes the potential of spiral and wave drawings, interpreted via deep learning, as non-invasive indicators of Parkinson’s disease. Early motor symptoms, such as hand tremors, often manifest in handwriting, making graphical input tasks a promising screening modality. A pre-trained ResNet50 model was adapted to classify grayscale biomedical sketches using a custom preprocessing pipeline that preserves spatial fidelity. The system was trained on a balanced dataset and enhanced with domain-specific data augmentation to improve generalization. Internal evaluation achieved classification accuracies of 97% for spirals and 93% for waves. The method emphasizes computational efficiency and stability under low-data conditions, supporting its suitability for real-world clinical use. A Python-based graphical interface was developed to enable clinicians or patients to submit hand-drawn inputs and receive real-time diagnostic feedback. These results demonstrate the feasibility of combining deep learning with simple graphical biomarkers for early PD detection and provide a foundation for future clinical validation. This study contributes a portable, interpretable module to support early neurological assessment.

Keywords: Biomedical Engineering, Neural Networks, Intelligent Systems
Abstract: Deep learning models for ECG signal classification have emerged as powerful tools for achieving accurate medical diagnoses. However, the supervised training of such models typically requires large, well-annotated, and balanced datasets, which are often challenging to obtain. These limitations can be mitigated through semi-supervised learning approaches that leverage partially labelled datasets. Despite their potential, semi-supervised methods face the risk of incorrectly interpreting unlabelled examples during the training process, or a lack of diversity for the labelled data, which can compromise model performance. To address these issues, we propose a semi-supervised method that combines self-labelling and consistency regularisation for a more refined selection of unlabelled data at each training iteration, thereby reducing the likelihood of ineffective training. The approach does not require a pretrained teacher model and introduces a refined consistency regularisation mechanism that integrates both the model’s predictions and the corresponding gradCAM (Gradient-weighted Class Activation Mapping) visualisations. The method is evaluated on patient-independent models using data merged from diverse ECG datasets. Experimental results demonstrate that the combined use of self-labelling and consistency regularisation enhances classification accuracy by leveraging noise injection from consistency regularisation and labelling reliability provided by proxy-based classification.

Keywords: Optimization and Optimal Control, Data - Driven Control, System Biology
Abstract: This paper presents a control strategy to track an optimal biomass concentration setpoint for microalgal growth in a simulated closed photobioreactor under natural sunlight. We investigate a feedback linearizing control approach with a lumostatic setpoint, which adjusts the dilution rate to maintain, at each time step, the biomass concentration that yields maximum productivity. To achieve this, we employ established radiative and kinetic models for Chlorella vulgaris from the literature. The radiative model describes light attenuation in the culture, while the kinetic model provides the growth rate for a given light intensity. The control strategy was tuned and tested using several idealized light profiles. We then evaluated the biomass productivity over an entire year of solar data and compared the results with those from a simple open-loop operation and a more complex model predictive control (MPC) strategy. The proposed lumostatic operation, under ideal conditions and with proper tuning, achieved production levels approaching those of the more complex strategies.

Keywords: Control Applications, System Identification and Modeling, Stochastic Systems
Abstract: This paper aims to present the reactor poisoning effect, especially the dynamics behind 135Xe concentration variations. The experiment consists of constructing the mathematical model for 135Xe and precursor 135I and 135Te isotopes concentrations. The model was discretized with the Forward-Euler discretization method with ∆t=0.2s and rearranged into a recursive form to be suitable for implementation a LabVIEW-based nuclear powerplant simulator. By using NI’s LabVIEW instead other conventional simulation software the user is allowed to modify reactor parameters and analyze the interactions of other powerplant components with the reactor 135Xe level in real-time. Moreover, such built NPP simulator allows the operator to track the evolution of 135Xe concentration to test the nuclear powerplant simulator’s PID controller’s performances both for neutron-flux and reactor power stability. The obtained 135Xe model has been used to demonstrate the sensibility of fission nuclear reactors to power- and neutron-flux fluctuations.

Keywords: Aerospace Systems, System Identification and Modeling, Industrial Applications
Abstract: This paper presents a comprehensive framework for modeling and simulation of a 155mm spinning glide-guided projectile, a novel ammunition concept capable of range extension and trajectory correction by means of steering canards. The projectile's equations of motion are derived from classical flight dynamics, assuming the system is governed by rigid body mechanics. The range extension capability of the projectile is then assessed using an elementary control law. A model selection algorithm using multivariate orthogonal functions is introduced to obtain nonlinear analytic expressions of the aerodynamic coefficients, in a suitable form for nonlinear control purposes. Finally, these nonlinear models are studied through numerical simulations by introducing a new validation criterion.

Keywords: Adaptive and Robust Control, Control Applications, Industrial Applications
Abstract: This study proposes a novel framework to model, control, and optimize the turbine-generator ensemble using fractional-order calculus and artificial intelligence. The transfer function of the turbine with synchronous generator system is derived and approximated as a fractional-order transfer function to better represent its complex dynamics. Neural networks in MATLAB/ Simulink are then used to learn the response of the fractional-order transfer function under varying operating conditions, enabling data-driven identification of system behavior. Leveraging this model, a fractional-order proportional-integral-derivative controller is designed and implemented. To further enhance the system performances, an adaptive FOPID controller is developed, utilizing real-time AI-driven adjustments to tune fractional orders and gains in response to transient disturbances and load fluctuations. Simulation results demonstrate that the adaptive FOPID framework significantly improves system performance, achieving superior reference tracking, faster settling times, and enhanced robustness compared to the static FOPID controller. This paper highlights the potential of combining fractional-order modeling with AI-based adaptive control to advance the stability and efficiency of mini hydropower systems, offering a scalable solution for renewable energy integration in dynamic environments.

Keywords: Machine/Reinforcement Learning, Control Applications, Nonlinear Systems
Abstract: In this article, two methodologies are implemented for learning control laws in a Reinforcement Learning setup for a Two Rotor Aerodynamic System, a cross coupled and a highly nonlinear system, using Genetic Programming and NeuroEvolution of Augmented Topologies. The results of the two methods are presented and compared, highlighting their learning and step response performances. The initial results demonstrate significant potential of the methods, considering the dynamic complexity of the plant involved, however, future improvements are needed to refine performance. This work represents a notable starting point for the development of new Reinforcement Learning Control algorithms which, unlike traditional Reinforcement Learning aimed at optimizing machine learning models, enable the explicit discovery of control laws directly on the physical system.

Keywords: Control Systems Design
Abstract: This paper proposes a robust control scheme for Permanent Magnet Synchronous Motors (PMSMs) that integrates integral backstepping with a novel double-integral synergetic controller. By incorporating a double integral term into the synergetic control design, the proposed method achieves rapid disturbance rejection and ensures robust performance under parameter variations and external disturbances. The strategy also provides smooth dynamic response and enhanced stability, making it suitable for high-performance PMSM applications such as electric vehicle drives. Simulation and experimental results validate the proposed approach, demonstrating improved torque regulation, reduced steady-state error, and enhanced stability without chattering when compared to two existing approaches in the literature: Interconnection and Damping Assignment Passivity-Based Control (IDA-PBC), and the Integral Backstepping Double-Integral Sliding Mode Controller (IBC-DISMC).