Abstract
Continuous verification of Personal Protective Equipment (PPE) compliance in CNC machining environments is a measurement problem: compliance state must be inferred from noisy, intermittent RFID readings, time-stamped with finite sampling resolution and network latency. Manual inspection offers no quantifiable measurement uncertainty and cannot serve as auditable evidence for quantitative risk assessment. This paper reports a distributed multi-agent measurement architecture, comprising an RFID-based Smart Gate and a dual Digital Twin feedback layer for a CNC Z32 lathe, together with a full measurement-uncertainty budget for the two indicators characterising the detection chain: the Positive Detection Ratio (PDR) and the Authorisation Latency (AL). Following the Guide to the Expression of Uncertainty in Measurement (GUM), five uncertainty sources are identified and classified as Type A or Type B (sampling quantisation, debounce-window variability, network time-synchronisation offset, RFID detection stochasticity, HTTP/TCP transmission jitter), allocated to each measurand's budget, combined with a stated coverage factor, and complemented by a rule-of-three upper confidence bound for the observed zero False Authorisation Rate (FAR). Experimental validation gives FAR = 0, PDR in the range 50.7-82.7 % across PPE categories, authorisation latency below 1 s at 10 Hz, and unit sensitivity/specificity in detecting spindle-guard tampering, with zero packet loss in the machine-feedback layer. Expressing these indicators with a stated coverage interval, rather than as bare point estimates, turns them into metrologically traceable quantities usable in a quantitative risk-assessment matrix under EN ISO 12100:2010, and illustrates a general method for attaching a defensible uncertainty budget to distributed, RFID-based access-control instrumentation.