Sensorless fault diagnosis of centrifugal pumps

Analysis of electrical signatures has been in use for some time to assess the condition of induction motors. In most applications, induction motors are used to drive dynamic loads, such as pumps, fans, and blowers, by means of belts, couplers and gear-boxes. Failure of either the electric motors or the driven loads is associated with operational disruptions. The large costs associated with the resulting idle equipment and personnel can often be avoided if the degradation is detected in its early stages, prior to reaching catastrophic failure conditions. Hence the need arises for cost- effective schemes to assess not only the condition of the motor but also of the driven load. This work presents an experimentally demonstrated sensorless approach for model- based detection of three different classes of faults that frequently occur in centrifugal pumps. A fault isolation scheme is also developed to distinguish between motor re- lated and pump related faults. The proposed approach is sensorless, in the sense that no mechanical sensors are required on either the pump or the motor driving the pump. Rather, fault detection and isolation is carried out using only the line voltages and phase currents of the electric motor driving the pump, as measured through standard potential transformers (PT's) and current transformers (CT's) found in industrial switchgear. The developed fault detection and isolation scheme is insensitive to electric power supply variations. Furthermore, it does not require a priori knowledge of a motor or pump model or any detailed motor or pump design parameters; a model of the system is adaptively estimated on-line. The developed algorithms have been tested on three types of staged pump faults using data collected from a centrifugal pump connected to a 3, 3 hp induction motor. Results from these experiments indicate that the proposed model-based detection scheme effectively detects all staged faults with fault detection times comparable to those obtained from vibration analysis. In addition to the staged fault experiments, extended healthy operation reveals no false alarms by the proposed detection algorithm. The proposed fault isolation method successfully classifies faults in the motor and the pump without any mis-classification