Generation of Real World Maintenance Data of Data Center Uninterruptible Power Supply Systems and Failure Prediction

Abstract

Industry 4.0 positions real-world data as a transformative asset that drives the development of new knowledge, advanced solutions, and industrial growth. The exponential increase in data generation has accelerated the demand for data centers and consequently, uninterruptible power supply (UPS) systems, which are critical for ensuring continuous power delivery. Predicting failures in UPS systems using both structured and unstructured data requires a well-curated data preparation process to enable effective analysis and modeling of the data. This study introduces techniques for transforming these diverse data types to extract actionable insights using a GenAI model. The model leverages two feature sets: one sourced from the Customer Relationship Management (CRM) system and the other derived from service completion reports documented by field service representatives. A key innovation of the model is the use of instructional prompts and a rule set that includes keyword mappings and acronym references, which enables the accurate interpretation of domain-specific language. This study also captures global performance insights for UPS systems and integrates data visualization. These visualizations facilitate the identification of failure patterns, including symptomatic and asymptomatic service order categories, failure origin, failure types, and criticality, enabling proactive maintenance strategies and enhancing system reliability. Model validation demonstrated a weighted accuracy and precision exceeding 90 % and an F1 score of 0.91.