Project Leader:	Professor Dr-Ing Horst Rulfs
Research Assistant:	Dipl-Ing A Michaelsen
Duration:	01.10.1992 - 28.02.1996

 

In this project a method was developed and tested to enable automatic monitoring and fault recognition in centrifugal pumps. Preventive monitoring and differentiated early detection are measures which improve the cost effective and safe operation of technical systems. On the basis of advanced signal analysis processes and pattern recognition it was possible to achieve a case-oriented condition diagnosis of a reference rotary pump.

In the framework of extensive experimental studies using the reference centrifugal pump different error and failure scenarios for rotary pumps were documented on the basis of measurements. Based on questioning of experts from various pump operators extensive knowledge on the typical faults and damages in rotary pumps were gathered. In addition, a catalog of operation points for each fault scenario was developed, to highlight the feasibility of diagnosing the faults within the full operating range.

The necessary interconnection between the technical measurements and the diagnostic decision-making through classification arises from an efficient processing of the signal pattern through signal analysis. Within the framework of the signal analysis, processing in the time and frequency domains as well as the cepstral region were used. Besides statistical methods with selective frequency pre-processing this work also used specific methods of signal analysis such as the Hüll curve modulation.

For tackling complex applications a new classifying diagnostic network has been proposed. This enables to diagnostically monitor centrifugal pumps within a very wide operation range. On the basis of measurements of the body vibrations and fluid noise it is possible to early detect and safely diagnose pump flaws and failures.

In the industrial praxis, however, such diagnostic monitoring of technical systems will first be conceivable when the investment and operation costs of the diagnostic system are low enough to reflect the perceived low value of the monitoring in relation with the process itself.