Contrôle
Permanent URI for this collectionhttps://dspace.univ-boumerdes.dz/handle/123456789/3081
Browse
161 results
Search Results
Item Hybrid fault estimation scheme of insulin delivery system(2025) Chehboune, Mohamed Amine; Bouarour, Mohamed Amine; Kouadri, AbdelmalekType 1 diabetes requires continuous insulin infusion, but sensor and pump failure can compro-mise closed-loop control. We designed a bank of observers for Continuous Glucose Monitor(CGM) sensor faults, insulin actuator faults, and meal disturbances independently. Observer estimates feed a two-stage classifier :fir st todete ctmea lsa nddisab lefau ltobserv eroutpu tsdur-ing disturbance, and second to classify sensor/actuator states and disable corrupted estimates due to fault coupling. Simulation results confir mtha tth epropose dapproac haccuratel ydetects and estimates both faults and disturbances, and works reliably across differen tscenarios.Item Real-time SCADA system for Industrial Inspection using YOLOv8 segmentation(University M’hamed Bougara : Istitute of Electrical and Electronic engineering (IGEE), 2025) Meneceur, Abdelatif; Kahloul, Noureddine; Loubar, HocineThis thesis presents the design and implementation of an intelligent, real-time SCADA-based inspection and rejection system for defective yogurt cups on a conveyor belt, integrating deep learning, programmable logic controllers (PLCs), and SCADA technologies. At the core of the system is a YOLOv8 segmentation model trained to detect four classes of cups: Good, Defect, Open, and Missed Label. The model achieved high performance during training, with mAP@0.5 reaching up to 96.1% for the "Good" class and 80.6% for the "Open" class, confirming its effectiveness in segmenting and classifying objects in industrial inspection tasks. The trained model was first deployed and executed on a PC(CPU), which performed all necessary tasks including image acquisition, inference, communication with a Siemens S7-1200 PLC, and real-time data transmission to a SCADA dashboard. The PC also functioned as an OPC UA server, enabling centralized monitoring and efficient control of the entire system. The model was also tested on a Raspberry Pi 4 Model B with 2GB RAM to explore the possibility of edge deployment. However, the system did not achieve the required processing performance for real-time cup inspection, as the inference speed was too slow for practical use on a fast-moving production line. Upon identifying a defective cup, the PLC triggers a rejection mechanism by sending a signal to an Arduino Uno, which drives a stepper motor through an optocoupler-isolated L298N driver. The complete SCADA system demonstrates a modular, scalable, and cost-efficient approach to automated visual quality inspection in manufacturing environments. It highlights the integration of embedded AI, industrial control systems, and open communication protocols within the context of Industry 4.0.Item Machine learning based approach for system identification ; a cement Bag filter case study(University M’hamed Bougara : Institute of Electrical and Electronic Engineering (IGEE), 2025) Benlaharche, Imene; Aziz, Abderazak; Kouadri, AbdelmalekTraditional system identificatio nmethod shav elon gbee nfundamenta ltool sfo rmodeling dynamic systems. Rooted in system theory and statistical estimation, these approaches have proven effectiv ei nman yapplications .However, as real-world system sgro wmore complex and data-driven, conventional techniques often struggle, particularly when system dynamics are complex, nonlinear, or time-varying. System identification methods —whether based on structure dmodels or more flexi-ble learning techniques — have proven effective for modeling a wide range of dynamic systems. However, as systems become more complex, nonlinear, or time-varying, these approaches may encounter limitations. As a contribution toward addressing these challenges, this thesis proposes a data-driven modeling approach called Polynomial Kolmogorov–Arnold Networks (PolyKAN). In-spired by the Kolmogorov–Arnold representation the orem, PolyKAN employs polynomial basis functions to approximate nonlinear input–output relationships directly from sensor data, without requiring prior knowledge of the system’s structure. The method is validated through two case studies: a Three Tank pilot plant and a Cement Bag Filter system. Results show that PolyKAN effectively models the nonlinear dynamics of both systems, achieving superior performance compared to conventional methods. Interms of Root Mean Squared Error (RMSE), it reaches values as low as 4.87 × 10-3 and 2.24 × 10-2, and R 2 scores up to 0.9851 and 0.9865, respectively. These results highlight PolyKAN’s potential for accurate and more interpretable system modeling in complex environments.Item A Comparative study of airspeed control for the boeing 737 (NG), using MPC, LQR, and PID(University M’hamed Bougara : Institute of Electrical and Electronic Engineering (IGEE), 2025) Hadj Mohammed, Zahafi Abdelhad; Boumaref, Chaima; Loubar, HocineThe Boeing 737 Next Generation (NG) aircraft, revolutionized commercial aviation by offering significant improvement sover the previos 737 Classic series since it has been introduced in 1997. Its redesigned wing with a larger area, a wider wingspan, and a greater fuel capacity, significally enhanced its performance, partucularly in fuel efficiency, rangeang passenger comfort. In this work, model predictive control, linear quadratic regulator, and Proportianal-Integral-Derivative control strategies are used to control the airspeed of the Boeing 737 (NG) aircraft. By employing a detailed nonlinear 6-degree-of-freedom (6DOF) dynamic model, which is linearized for control design, the study simulates airspeed changes over periods of 15, 100, and 2000 seconds, evaluating performance in both speed-up and slow-down sce-narios. The research includes thorough aerodynamic modeling, taking into account lift, drag, and thrust effects, all transformed across various reference frames (body, stability, and wind) to create a solid basis for assessing the controllers. The simulation results show that MPC stands out for its adaptability and accuracy in pre-dictive optimization, while LQR excels in delivering optimal feedback with strong stability, and PID offers practical tuning flexibility, though it does face some challenges with over shoot management. A comparative analysis showcases the trade-offs in responsiveness, stability, computational demands, and setpoint accuracy, providing useful insights for choosing control methods suited to different flight phases.Item Ros-Based real-time multi-waypoint path Planning(2025) Seghier, Mohamed Islam; Touzout, WalidThis thesis introduces an innovative optimization technique for multi-waypoint path planning, enhancing path efficiency The Multi-Waypoint Slime Mould Algorithm (MWSMA) leverages a metaheuristic approach inspired by the slime mould algorithm, incorporating a priority-based encoding scheme and an objective function to optimize waypoint sequences. A heuristic priority function guides the algorithm, with parameters tuned using Particle Swarm Optimization (PSO) for robust performance. The MWSMA is implemented in a real-time agricultural task using ROS2, demonstrating its practical applicability. This approach ensures efficient, adaptabl epat hplanning, maintaining the original algorithmic structure while significantly improving solution quality.Item Kalman-based localization of a mobile robot(University M’hamed Bougara : Institute of Electrical and Electronic Engineering (IGEE), 2025) Benali, Nadia; Bengherabi, lina; Guernane, RedaThis thesis explores the development of a localization system for mobile robots oper-ating in indoor environments. The proposed system fuses data from wheel encoders and ultrasonic sensors using sensor fusion techniques based on Kalman filtering .Tw ofiltering algorithms are evaluated: the Extended Kalman Filter (EKF) and its Lie group-based variant, the Left-Invariant EKF (LIEKF). To determine the initial pose of the robot, a hybrid strategy is employed, combining a particle-based global search with a fine-graine dgri drefinemen t.T heenti resyst emis implemented and tested in both simulated and real-world conditions using a commercial differential-driv erobo tan d aMATLAB-base dprogram. Results demonstrate that both filter simprove localization compared too dometry alone, with LIEKF providing more stable estimates, less drift, and faster convergence, es-pecially in nonlinear motion and poor initial conditions. This work highlights how simple sensors and filterin gtechnique sca nb ecombine dfo reffecti velocalizati on instructured indoor environments.Item Intelligent fault detection in gas turbine safety systems(University M’hamed Bougara : Institute of Electrical and Electronic Engineering (IGEE), 2025) Bourenane, Mohamed Akram; Hamidouche, Hamoud Abdelbasset; Boushaki, RazikaModel Predictive Control (MPC) strategies, the Triconex SIS demonstrated improved responsiveness and fault tolerance, particularly in scenarios such as overspeed events and combustion instability. The results underscore the viability of MOMEDA and mKurt as robust tools for real-time fault diagnosis, offering a framework for enhancing safety-critical systems in industrial applications. This research advances the development of intelligent, data-driven control systems, providing actionable insights for optimizing gas turbine reliability and operational safety through the Triconex SIS platform.Item PLC based Bottle filling system using neural network control(University M’hamed Bougara : Istitute of Electrical and Electronic engineering (IGEE), 2025) Fredj, Mohamed; Haddouche, RezkiThis thesis presents an intelligent control strategy for an automated liquid bottle filling system, where the operation time of an electric pump is determined using a neural network-based identification approach. The objective is to improve filling precision and efficiency by predicting the optimal pump control based on the weight measurement. A Nneural network is trained on system data to model the dynamic behavior of the fillin gprocess, serving as a digital twin that captures the system’s dynamics. This model is then used to train a second neural network that functions as a controller, predicting the optimal pump runtime needed to achieve precise filling. The resulting control logic is implemented on a Siemens S7-1200 Programmable Logic Controller (PLC), ensuring reliable and consistent operation in an industrial environment. Experi-mental results illustrate the potential of data-driven system identification in enhancing automated filling processes, offering a foundation for further refinement and optimization.Item Dual PLC-based HMI control and monitoring of siemens VFDs using profibus and profinet networks(University M’hamed Bougara : Istitute of Electrical and Electronic engineering (IGEE), 2025) Siad, Rihab; Rebali, Lilya; Boushaki, RazikaThis project presents the design and implementation of an advanced industrial automation system for controlling Siemens MICROMASTER 420 Variable Frequency Drives (VFDs) using a programmable logic controller (PLC)-based architecture. The central control unit is a Siemens S7-1500 PLC programmed with TIA Portal V15.1, which manages communications with an S7-300 PLC via a PROFINET network, where the S7-1500 operates as the server and the S7-300 as the client. In parallel, the S7-300 PLC functions as a PROFIBUS master to control three VFDs, each of which governs an induction motor. The VFDs act as DP slaves, receiving commands and returning operational data such as faults, alarms, and load conditions. An integrated Human-Machine Interface (HMI) replaces the conventional Basic Operator Panel, enabling users to monitor and control system variables such as motor speed, direction, and operating status. The HMI also supports real-time diagnostics and control functionalities for improved system responsiveness and user accessibility. Data exchange between PLCs is achieved through the implementation of the PUT/GET communication blocks in STEP 7, while SFC 14 and SFC 15 functions facilitate PROFIBUS-based communication with the VFDs. The proposed system demonstrates a flexible and scalable solution for modern industrial motor control, combining high-performance communication protocols with intuitive user interaction.Item Holistic data driven fault estimation approach in complex systems(University M’hamed Bougara : Institute of Electrical and Electronic Engineering (IGEE), 2025) Bendjedia, Djoumana; Kouadri, AbdelmalekThis project focuses on addressing the issue of fault estimation in dynamical industrial systems through a three-tank experimental pilot plant. The system under differen toper-ating conditions experiences multiple types of faults, including actuator faults and sensor faults. In this study, we aim to develop and compare data-driven methods that can accurately estimate the sensor/actuator faults based on input-output measurements. Toaccomplish this, a complete dataset was generated under normal, faulty, and disturbed conditions. The input features of machine learning-based model consist of the manip-ulated and measured variables, while the outputs correspond to sensor/actuator fault signals. Multiple fault estimation techniques were applied and evaluated. Independent Component Analysis (ICA) was used firs ta s alinea rtechniqu efo rsourc eseparation, However it failed to distinguish between faults and often produced switching and flipping errors due to linear mixing assumptions. Consequently, machine learning approaches were adopted, including Support Vector Regression (SVR), Decision Trees (Random Forest), Gradient Boosting, Neural Networks (NN), and Kolmogorov-Arnold Networks (KAN). Performance was evaluated using Root Mean Squared Error (RMSE), Mean Absolute Error (MAE), and Coefficie nt ofDeterminati on(R 2 ) metrics to assess the balance be- tween prediction accuracy and model robustness.The results showed that Neural Net-works achieved the highest estimation accuracy overall, demonstrating strong capability in capturing complex nonlinear relationships. Random Forest also delivered robust per-formance, particularly for sensor faults, benefitin gfro mit sensembl elearnin gstructure. XGBoost and Support Vector Regression provided satisfactory results for actuator faults but showed some variability across fault types. Although Kolmogorov–Arnold Networks offe rstron gtheoretica lflexibilit y,the irperforman ce inth isstu dyw asle ssconsiste ntcom- pared to the other methods. Overall, the data-driven approach proved effectiv efo rfault estimation even under disturbance influence ,offeri n gareliab lealternati vewh enaccurate physical models and full disturbance decoupling are difficu lt toachieve.