Academy of Technology

The integration of Internet of Things (IoT) and Artificial Intelligence (AI) technologies in underground coal mining has received growing attention due to its potential to improve operational efficiency and reduce energy consumption. IoT-based sensing systems enable real-time monitoring of environmental and operational conditions, while AI techniques support predictive analysis and decision-making. Existing studies indicate that such integration can contribute to improved ventilation control, predictive maintenance, and system-level energy management. This review synthesizes recent developments in IoT and AI applications in underground mining and examines their role in supporting energy-efficient operations. The analysis also highlights current limitations related to scalability, long-term validation, and data reliability. Overall, the literature suggests that IoT–AI integration offers promising opportunities for improving energy performance and sustainability in underground coal mining.

Parity-time (PT) symmetry in time-delay oscillators such as lasers and optoelectronic oscillators provides a potential route to enhanced spectral purity, including reduced phase noise and improved sidemode suppression. Existing theoretical descriptions are typically based on coupled-mode formulations derived under slowly varying envelope and near-degeneracy assumptions, which restrict their validity to weak coupling, small gain/loss contrast, and small detuning. In this work, a non-perturbative formulation of PT symmetric time-delay oscillators is developed based on a delay-difference equation and a scattering matrix representation of the coupling network. The approach treats propagation delay explicitly and does not rely on modal truncation, remaining valid for arbitrary coupling strength, gain/loss imbalance, and resonance detuning. The exact eigenvalue structure of the system is obtained in closed form, yielding a complete characterization of the unbroken and broken PT symmetric regimes as well as the associated exceptional points. A dimensionless order parameter is introduced that governs the symmetry transition over the full parameter space. It is further shown that conventional coupled-mode theory is recovered as an asymptotic limit of the exact formulation for small parameters. The results provide a unified and physically transparent framework for analysing PT symmetric delay systems beyond the weak-coupling limit, with direct implications for the design and optimisation of low-noise oscillators and photonic systems.

For accurate detection of thyroid cancer stages, including I, II, III, IVA, and IVB, proper treatment planning is necessary in a modern, sustainable medical system. Nowadays, researchers suggest the use of machine learning models for detecting complex patterns from laboratory test data. Hence, this work presents a novel, secure, and reliable fusion model that efficiently predicts stages of thyroid cancer. To design a strong ensemble model, it integrates BERT and XGBoost, which exploits both structural and categorical data to identify patterns from clinical data. Along with this, to improve security and transparency, this model was then integrated with a Blockchain platform, where patient records were securely deployed in an immutable platform. Finally, Performance analysis shows that the proposed framework outperforms other competitive classifiers in detecting the stages accurately. Thus, this work demonstrates an effective and secure framework that has the power to identify deep contextual meaning for predicting thyroid cancer stages and utilizes blockchain technology for enhancing privacy and transparency for securing diagnostic information.

Early detection of breast cancer is vital for effective therapy and improved survival rates. To mitigate this, we propose a federated learningintegrated vision transformer-based model, combined with Explainable AI, for privacy-preserving and accurate breast cancer diagnosis from histopathological images. The system uses FedAvgLoRA aggregation and fine-tuning to enable communication-efficient federated learning, thereby accelerating convergence and reducing resource usage. This framework uses Vision Transformer to capture fine-grained spatial and contextual information from histopathological images. To improve model performance, Optuna is used for hyperparameter optimization purposes. It successfully captures finegrained tissue structures and achieves higher accuracy and recall than baseline CNN models. Finally, the entire framework is integrated with Grad-CAM for rendering the model explanations. Performance evaluation demonstrates that it outperforms in terms of standard performance metrics than competing schemes, which is essential for early breast cancer detection in federated medical facilities.

Image segmentation plays a essential role in both the prevention of diagnostic errors as well as the early and precise detection of skin cancer. These identification techniques are becoming progressively more effective with the help of the latest advancements in computer vision. To increase segmentation accuracy, this study implements a superior U-Net architecture with dense blocks. Instead of using standard convolutional layers, the network utilizes densely connected layers at every step of the network it more effectively preserves and employs feature information. Due to this updated layout, based on the encoderdecoder architecture of U-Net, the model can now learn highly complex patterns in dermatological images. It also goes beyond that by patient variability and different lighting condition problems. The achievement of the proposed model has been evaluated with the help of the ISIC 2018 dataset, and the desired results led to a Dice coefficient of 91.38%, a IoU of 84.78%, and an accuracy of 97.72%. These findings are indicative of the model as a precise and reliable instrument for skin cancer segmentation.