This research develops Smart Twin PM, a six-layer digital twin system for predictive maintenance in manufacturing. By combining real-time data analytics, physics-based validation, cybersecurity checks, and smart scheduling, it reduces unexpected failures by 15% and false alarms by 20%, enabling proactive, trustworthy, and efficient machine maintenance.

This research uses drone imagery and a hybrid AI model to classify rangeland cover as green vegetation, dead vegetation, or bare soil. Combining two neural network approaches achieved 96% accuracy while requiring only simple, low-cost sensors. The method enables fast, large-scale monitoring to combat invasive shrubs and support sustainable land management.

Traditional neural networks are powerful but difficult to interpret and vulnerable to small input changes. This research develops wavelet-based neural networks with provable stability guarantees, extending the scattering transform to texture modeling. The approach reduces feature complexity while improving interpretability, enabling more reliable and mathematically explainable AI systems.

This thesis developed a real-time system for detecting, classifying, and locating sound events using only audio data. A network of 16 microphones and deep learning techniques achieved 96% classification accuracy and average localization error of 1.4 meters, demonstrating that sound-based analysis can effectively replace vision in monitoring applications.

This project uses hive sound recordings and machine learning to detect early signs of bee swarming. By identifying acoustic differences between swarming and stable colonies, the system predicts swarming with 93% accuracy. This enables beekeepers to intervene early, prevent colony loss, and even create new healthy colonies.

This study modeled wild edible mushroom yields in Mediterranean forests using Planet satellite imagery, LiDAR, climate data, and field measurements. Results show that seasonal NDVI differences, precipitation, and forest structure are key predictors. Integrating high-resolution intra-annual remote sensing significantly improves yield prediction and ecological understanding.

This study evaluated a PointNet++ deep learning model for binary classification of Pinus sylvestris and Quercus pyrenaica using only LiDAR 3D point clouds. A balanced dataset of 160 trees achieved 91% accuracy, showing that geometric features alone can effectively discriminate species, highlighting the potential of lightweight AI models for forest inventories.

This study mapped land use changes in the Grombalia Region of Tunisia using Sentinel-2 imagery and machine learning. Three classifiers—Random Forest, SVM, and CNN—were compared. Random Forest achieved the highest accuracy. Results highlight agricultural changes over time and demonstrate the effectiveness of remote sensing for environmental monitoring.

This study developed a real-time IoT-based system to optimize fishway performance in fragmented rivers. Using sensors, PIT-tag tracking, and machine-learning models, it links climate triggers with hydraulic controls. Adaptive sluice-gate regulation improved fish passage efficiency by 166% without reducing hydropower output, offering scalable, sustainable river management.

This research applies machine learning to genetic data to distinguish harmless DNA variations from cancer-causing mutations. By treating DNA like a crime scene, the model learns to identify which genetic changes truly drive breast cancer risk, supporting more accurate diagnosis and informed clinical decision-making.