Dynamical Flow Rate and Pressure Artificial Neural Network Estimators for a Centrifugal Fan Driven by an Induction Motor Drive

To provide significant energy saving in the air and water supply systems, centrifugal fans/pumps are driven by variable frequency induction motor drives. To reduce the cost of the required energy saving close loop control, the necessary expensive flow rate or pressure (head) sensors can be replaced by estimators based on the monitored variables of the drives and imbedded into the drives' software. Available estimation techniques are based on the steady state experimental or data sheet fans'/pumps' curves which is justified by quasi-steady modelling of the fans and pumps. The present paper identifies the problem of this approach during transients. It develops quasi-steady and dynamical estimators for the flow rate and pressure of a centrifugal fan with induction motor drive based on the neural networks trained based on the steady state and transient experimental data. It uses the referred frequency of the stator voltages, the measured rms stator current and the estimated input active power of the motor as inputs of the estimators avoiding the velocity estimation as in the available approaches. It demonstrates better dynamical performance of the dynamical estimators with one sample time delayed feedback as the additional input which is justified by dynamical modelling of the fans/pumps. The details of a specifically designed test rig based on the Nicotra-Gebhardt industrial centrifugal fan equipped with a three-phase induction motor and estimators' design procedure are explained.