AI-AUGMENTED ROOT-CAUSE ANALYSIS WORKFLOWS FOR LARGE-SCALE FIXED BROADBAND FIBER ACCESS NETWORK OPERATIONS
TL;DR
An evidence-centred model for artificial-intelligence-augmented root-cause analysis in large-scale fixed broadband Fiber/GPON operations is developed and concludes that reliable AI for GPON assurance should operate as a governed diagnostic co-worker rather than an autonomous cause generator.
Abstract
Fixed access broadband networks based on fibre-to-the-home and Gigabit-capable Passive Optical Network technologies have become critical infrastructure for residential connectivity, enterprise access, public services, and cloud-dependent economic activity. Their operational simplicity at the physical layer masks a complex service chain linking optical line terminals, passive splitters, feeder and distribution fibres, optical network units, aggregation switches, broadband network gateways, authentication platforms, policy systems, and customer-premises equipment. A single optical degradation, provisioning error, capacity constraint, or software change can consequently generate alarms and symptoms across several systems. This review synthesises research and standards published from 2020 to 2025 and develops an evidence-centred model for artificial-intelligence-augmented root-cause analysis in large-scale fixed broadband Fiber/GPON operations. It compares rules, optical telemetry analytics, time-series learning, topology and graph reasoning, multimodal fusion, digital twins, and language-model-assisted diagnostics. The proposed workflow frames the incident, resolves subscriber and network identities, builds a service-aware PON dependency graph, detects abnormal change, ranks testable hypotheses, verifies them with targeted evidence, controls remediation, and converts verified incidents into reusable operational knowledge. Particular attention is given to optical power drift, loss-of-signal events, branch faults, ONT registration failures, OLT port congestion, dynamic bandwidth allocation anomalies, VLAN and service-profile errors, and broadband session failures. The review concludes that reliable AI for GPON assurance should operate as a governed diagnostic co-worker rather than an autonomous cause generator. Production value depends on evidence provenance, topology freshness, calibration, unseen-fault generalisation, safe remediation, and measurable reductions in diagnosis time and unnecessary field intervention.