Generative artificial intelligence for root cause analysis optimization: a case study in 5G equipment manufacturing

Abstract

The complexity involved in manufacturing last-mile telecommunications equipment, such as the OptiXstar family, requires extremely agile fault diagnosis in order to maintain global competitiveness. This article presents the results of implementing the Cyber Nexus platform, which uses Large Language Models (LLMs) to automate Root Cause Analysis (RCA) at the Manaus Industrial Pole. The objective was to raise the process’s Industry 4.0 maturity level by shifting from reactive visibility to predictive transparency. The methodology consisted of an experimental case study with validation in a relevant environment (TRL 6), involving quality engineers and real production data. The results demonstrated a drastic reduction in the RCA cycle, from an average of 15 days to less than 48 hours, with an anomaly detection accuracy rate above 80%. The solution enabled real-time correlation between radio frequency calibration variables and failure logs, establishing the foundation for predictive maintenance and green manufacturing through the reduction of rework.