Search for a command to run...
Abstract Chang-Prompt is an AI-driven decision-support assistant designed to enable rapid, structured, and reliable troubleshooting in critical industrial operating environments. Inspired by clinical decision-support systems used in intensive care units (ICUs), Chang-Prompt employs a disciplined, stepwise diagnostic framework that integrates deterministic rule-based logic with probabilistic machine-learning models to support timely and consistent operational decisions. Architecturally, Chang-Prompt consists of two major functional components. The first component focuses on failure-part identification, or bleeding-point localization, which aims to determine the most probable source of degradation within the machine. The second component equips operators with quick-fix decision-support tools, enabling faster and more consistent response actions once the failure location is identified. In practice, Chang-Prompt comprises multiple diagnostic case groups addressing different classes of abnormal operating conditions across turbomachinery systems. This paper focuses specifically on one representative and high-impact case group: chip detection events identified by chip detectors installed within turbomachinery lubrication systems. Chip detection represents a critical early-warning signal of potential mechanical degradation, yet it is often associated with high uncertainty and false alarms when assessed in isolation. For this case group, Chang-Prompt integrates chip sensor diagnostics, trending analysis of supporting process and condition-monitoring sensors, and multiple scoring models within a unified in-house developed platform, eliminating fragmentation across analytical tools and data sources. Deterministic checks—such as chip level shift and chip flatness evaluation—serve as early decision gates to rapidly assign high-confidence outcomes. Ambiguous conditions are subsequently evaluated using trending behavior and probabilistic scoring, with model outputs aggregated using quantile-based methods to preserve sensitivity to emerging abnormalities. Machine-specific diagnostic workflows are implemented for AGB, IGB, 4R, and 5R configurations to account for differences in sensor availability and dominant failure mechanisms. Field application of the chip-detection case demonstrates that Chang-Prompt improves early fault detection, reduces false alarms, and shortens troubleshooting time by enforcing an ICU-style diagnostic flow. The results confirm the effectiveness of structured, hybrid AI decision-support in enhancing operational reliability and stability in turbomachinery systems.