Air purifier performance in an occupied office: Field CADR and effectiveness from a reconstructed device-off condition

Air-purifier performance in an occupied building is hard to assess because the indoor concentration without the device cannot be observed under the same conditions. This study reconstructs that device-off condition from long-term field measurements in an occupied office. Two sequentially coupled physics-informed neural networks (PINNs), constrained by the CO 2 and PM 2.5 mass balances, estimate the time-varying air change rate, penetration factor and removal rate, and the removal rate is re-evaluated at each instant with the air-purifier input switched off. The difference between the two removal rates gives field CADR, and the modelled concentrations with the device on and off give steady-state and dynamic effectiveness. Field CADR is an equivalent removal rate specific to this room and device, not a chamber CADR. With the window and door closed, the median field CADR was 175 m 3 /h and the steady-state effectiveness was 0.79. With the window, the door or both open they fell to 50–66 m 3 /h and 0.09–0.31; the open-condition values are less precise, whereas the closed-open difference was reproduced in synthetic-data tests within the same model class. A conventional before/after comparison on the same record agreed with the reconstruction in the aggregate under the closed condition but not event by event, returned negative values under full opening, and yielded 59 estimates against 3613 from the proposed field CADR estimation. Its generality beyond one office remains to be established.