Accelerator complexes contain tens of thousands of interdependent components, and aging infrastructure amplifies the risk of equipment faults and costly, unscheduled shutdowns. At the Los Alamos Neutron Science Center (LANSCE), we are developing a data-driven framework that flags developing problems early enough to address them during scheduled maintenance, thereby improving reliability and increasing beam availability for users. Our approach analyzes all available signals within a subsystem to learn the facility’s “normal” operating envelope and to detect subtle deviations that precede failures. Unlike the current warning scheme, it captures hidden correlations among parameters and generates interpretable indicators of abnormal behavior. Predictions are validated against historical control-room log records. We report progress on three fronts: (i) extending anomaly prediction from a single beamline to all major LANSCE subsystems; (ii) expanding data archiving capacity by an order of magnitude to support broader coverage and longer look-back windows; and (iii) developing operator-facing algorithms that both warn of emerging anomalies and localize likely problem elements along the beamline. Together, these advances are designed to shift maintenance from emergency response to planned intervention, reducing downtime and enhancing overall facility performance.