Abstract

The increasing complexity of multi-layer mobile network systems—comprising the physical, network, service, and application layers—has made fault detection, diagnosis, and self-optimization more challenging and critical than ever. As user expectations for seamless connectivity and high-quality service intensify, traditional rule-based network management strategies have proven insufficient in meeting the demands of real-time responsiveness and operational efficiency. This paper explores the latest advances in Artificial Intelligence (AI)-augmented fault detection and self-optimization techniques for multi-layer mobile systems, offering a comprehensive review and conceptual synthesis of emerging methods. The study presents a multidimensional approach that combines machine learning, deep learning, and reinforcement learning models to detect anomalies, predict failures, and trigger automated optimization protocols. Techniques such as autoencoders for unsupervised anomaly detection, convolutional neural networks (CNNs) for pattern recognition, and deep reinforcement learning for closed-loop self-optimization are critically analyzed. Special emphasis is placed on their application in heterogeneous network environments, including 4G, 5G, and the evolving 6G architectures. Furthermore, the paper introduces a conceptual framework that integrates AI-driven monitoring agents with centralized orchestration layers to ensure cross-layer data correlation, intelligent root cause analysis, and proactive resource allocation. Real-world case studies and simulation results from recent deployments demonstrate significant improvements in fault resolution times, network throughput, and energy efficiency. The framework is also aligned with zero-touch network management paradigms and network slicing strategies, essential for scalable deployment in modern telecom infrastructures. The findings reveal that AI-augmented systems outperform traditional mechanisms in speed, accuracy, and adaptability, ultimately reducing downtime and enhancing user experience. However, challenges such as data privacy, model interpretability, and the need for large-scale labeled datasets remain key constraints. This paper concludes by recommending future research directions focused on federated learning, explainable AI, and hybrid models that combine symbolic reasoning with data-driven approaches to enhance fault resilience and operational autonomy in next-generation networks.