End of Studies Projects 2024-2025

Permanent URI for this collectionhttp://dspace.hns-re2sd.dz:4000/handle/123456789/23

This collection contains the dissertations of students who completed their studies at the school in June 2025.

Browse

Start date: 2025End date: 2025

Search Results

Now showing 1 - 2 of 2
  • No Thumbnail Available
    Item
    Damage Detection in PV Support Structures Using Smart Materials and Finite Element Analysis.
    (2025-06-15) SAHRAOUI Amina
    Structural Health Monitoring (SHM) is essential for ensuring the reliability and longevity of photovoltaic (PV) systems, particularly in detecting damage in their support structures. This study presents a numerical approach based on piezoelectric smart materials (PZT) and Finite Element Analysis (FEA) to detect structural damage. Various shapes and types of PZT elements were tested by bonding them to aluminum samples, and the square-shaped PIC151 was selected for its superior performance. A full 3D ground-mounted PV support structure was modeled in SolidWorks, and the inclined support column, which was identified as a structurally sensitive component, was selected for detailed simulation in ANSYS with an integrated PZT sensor. Several damage scenarios were simulated, including cracks of varying depth, orientation, dimensions, and position; corrosion represented by gradual material degradation; and overloading modeled using compressive forces. The variations in electromechanical impedance (EMI) responses were analyzed to evaluate the sensor’s capability in damage detection. Results showed that PZT sensors effectively distinguished between damage types: cracks produced sharp, localized changes in EMI signals; corrosion caused smoother frequency shifts, reflecting its cumulative nature; and overloading led to noticeable impedance variations due to internal deformations, which could be either temporary or permanent. Furthermore, a comparative physical analysis was conducted to differentiate the impedance responses associated with cracks, corrosion, and overloading, highlighting the distinct mechanical and material behaviors underlying each damage mechanism. These findings confirm the potential of PZT-based SHM techniques for accurate and cost-effective monitoring of PV support structures. The results of this study pave the way for future experimental validation on real-world solar panel support structures to confirm the simulation outcomes. This work can also be extended to various mounting types. Additionally, it is recommended to integrate the EMI technique with real-time smart monitoring systems to enhance damage detection capabilities and improve structural health management.
  • No Thumbnail Available
    Item
    AI-Powered Platform for Automated Quiz and Question Generation.
    (2025-06-15) AISSAOUI Ayoub; BETTAHAR Akram; BOUREK Khalil; SAIGHI Ahmad Yasser
    This dissertation presents the design, implementation, and evaluation of an AI-powered platform developed for the automated generation of diverse assessment materials from user-provided PDF documents and raw text inputs. Addressing the significant time investment traditionally required for manual quiz creation in educational and training contexts, this work leverages advancements in Natural Language Processing (NLP) to streamline the process. The platform features a modular architecture implemented using Python and the Flask web framework, offering a user-friendly web interface for interaction. The core intelligence for generating Multiple-Choice Questions (MCQs) and Short Answer questions resides in a T5-base transformer model, specifically fine-tuned on the SQuAD v1.1 dataset adapted for question generation. This core model is supplemented by various NLP techniques and libraries (including NLTK, spaCy, and potentially Sense2Vec and Sentence-BERT based on implementation details) to facilitate the generation of Fill-in-the-Blanks questions, Matching tasks, and concise Summaries of the source material. The system allows users to specify the desired types and quantities of questions, providing flexibility in assessment creation. Evaluation of the fine-tuned T5 model demonstrated promising quantitative results, achieving average ROUGE-1 and ROUGE-L scores of 0.5247 and 0.4844, respectively, indicating a strong capability to capture semantic content and structure from the source text. While the average BLEU score was lower at 0.1988, this is often observed in generation tasks where content overlap (measured by ROUGE) is more critical than exact phrasal matching (measured by BLEU). Compared to existing commercial and academic solutions, the developed platform offers a distinct combination of input flexibility (PDF and text), a curated set of pedagogically relevant question types, and the use of an explicitly finetuned, adaptable T5 model. While acknowledging areas for future refinement, particularly concerning the robustness of PDF parsing across diverse formats and enhancing user interaction features, this project successfully demonstrates a viable and versatile approach to automating assessment generation. The platform holds significant potential to support educators, reduce administrative workload, and ultimately contribute to more dynamic and responsive learning environments.