Abstract -The fast development of 6G communication networks greatly increases the complexity of Software-Defined Networking (SDN) and Network Function Virtualization (NFV)-enabled cloud infrastructures. While these technologies enhance network flexibility, scalability, and efficiency, they also introduce dynamic failures, outages, and degradation of performance caused by virtualized network functions and heterogeneous traffic load. Traditional fault management solutions either utilize threshold-based monitoring or employ black-box deep learning models which lack transparency, fail to provide timely fault localization and insufficient self-healing. In order to address these issues, this work introduces an Explainable Hierarchical Deep Self-Healing Network (EHDSHN), which represents a new deep learning approach for autonomous detection and healing of faults within 6G cloud networks. The introduced framework incorporates a temporal traffic representation model based on Bi-directional Long Short-Term Memory (Bi-LSTM), topology-aware fault localization with the help of Graph Attention Networks (GAT), and SHapley Additive exPlanations (SHAP) to provide clear decision-making transparency. Optimization of corrective actions for fault recovery is done with the help of Reinforcement Learning. Evaluations are performed in the context of simulated SDN/NFV cloud networks with realistic fault datasets. Through experimental assessment, the proposed Explainable Hierarchical Deep Self-Healing Network (EHDSHN) can detect faults with an accuracy rate of 98.18%, precision of 97.86%, recall of 97.74%, and F1-Score of 97.80%, while at the same time lowering the network recovery delay to an average of 110 ms compared to other models including Bi-LSTM, Graph Attention Network, and DRL based self-healing approaches.