Weak-Signal Interpretation for Industrial Fault Detection and Predictive Maintenance

Industrial systems rarely fail as a single, fully legible event. More often, they drift toward failure through weak, noisy, partial, and sometimes converging signals such as vibration irregularity, slight thermal rise, intermittent acoustic anomaly, pressure fluctuation, current drift, or modest performance degradation. This paper argues that industrial fault detection and predictive maintenance are strong candidates for a stratified weak-signal architecture in which bounded local findings are represented provisionally, compared against local expectations, escalated through short-horizon salience when deviation becomes informative, and only later promoted into stronger intervention through persistence, convergence, and governance. The proposal is architectural and early-warning oriented. It does not claim exact failure forecasting, plant-wide optimization, safety certification, or replacement of engineers and maintenance teams. Its narrower contribution is to show that industrial systems need a disciplined place for weak precursor signals to matter before they harden into alarms, maintenance action, or ignored noise.