Detection limits of MES-embedded carbon-intensity monitoring for energy anomalies: a calibrated simulation study in machining-style processes

Real-time monitoring of carbon intensity (CI) — emissions per unit produced — at the work-centre level is increasingly deployed to surface energy anomalies in manufacturing before they accumulate into significant emissions. Its detection performance, however, has not been characterised: what is the smallest anomaly it can resolve, how quickly, and which faults does it systematically miss? We address this gap with a calibrated, fully reproducible simulation framework for MES-embedded CI monitoring in machining-style processes, whose energy substrate and no-load spindle parameters are anchored to the open Brillinger et al. (2025) CNC dataset and whose every parameter carries a provenance tag. Across a sensitivity campaign of 4,356 scenarios we establish two results: a detection floor — the minimum resolvable anomaly, expressed in relative (fraction-of-baseline) units — and a quantitative characterisation of the adaptive-baseline inertia trade-off, whereby faults that develop slowly relative to the monitoring baseline are absorbed into it and systematically missed. We characterise this limit as a function of a dimensionless onset-to-window ratio at a representative operating point, and map which industrial fault archetypes the deployed rule-based architecture catches and which it misses. All code, configurations, and raw results are released openly; field validation on instrumented machine tools is the natural next step.