A diagnostic system for dry pumps is proposed. It predicts future pump motor current from time-series in-situ measurements. The prediction system has been constructed using a data acquisition system with an online system identification software algorithm. A field test on a low pressure chemical vapour deposition (LPCVD) system for the silicon nitride process predicted large values of motor current, some of which correlated well with actual motor currents as the pump became clogged. The combined use of the predicted motor current and the stability criteria shows promise in predicting the actual service life of a dry pump.