From Alerts to Insights: An Integrated AIOps Framework for Multi-Modal Fault Warning and Causal Root-Cause Recommendation

Abstract

The escalating complexity of modern IT infrastructures necessitates advanced operational frameworks capable of preemptive fault management. This paper presents the design and implementation of an AI-based operational system that integrates real-time fault warning with intelligent root-cause recommendation. By leveraging a multi-source telemetry data pipeline encompassing metrics, logs, and traces, our system employs a stacked ensemble of forecasting models—including Temporal Fusion Transformers (TFT) and Graph Neural Networks (GNN)—to detect service anomalies and resource degradation with high precision. Upon fault identification, a causality engine utilizing Bayesian inference and topological dependency analysis pinpoints probable root causes, presenting operators with contextualized evidence and remediation steps. In validation across a large-scale e-commerce platform, the system demonstrated a 92.3% fault detection rate—outperforming threshold-based monitoring by 34%—and reduced mean-time-to-resolution (MTTR) by over 40% through its diagnostic recommendations. The architecture further incorporates a continuous learning loop, where analyst feedback on recommendations refines both detection sensitivity and causal models. This research not only offers a scalable blueprint for AI-enhanced operations but also establishes a new benchmark in integrating prognostic alerting with interpretable diagnostic support, substantially elevating the efficacy and autonomy of modern IT management.