Development and experimental demonstration of a model-based fault detection system for electromechanical systems

Electromechanical systems, such as electric motors driving dynamic loads like pumps and compressors, often develop incipient faults that result in down-time. There is a large number of such fault classes, and their precise signatures depend on numerous parameters including variations in the motor power supply and driven load. Practical fault detection and diagnosis systems must exhibit high level of detection accuracy and acceptably low false alarm rates. They must have broad applicability, require installation of minimal extra sensors, and not require the use of detailed machine information for operation. In this paper the development and experimental demonstration of a model-based detection system for incipient electric machine faults is presented. The developed fault detection system uses recent developments in dynamic recurrent neural networks and multi-resolution signal processing. The sensors utilized axe only those measuring the motor current and voltage. The effectiveness of the developed system is demonstrated by detecting stator, rotor and bearing faults at the early stages of development. Furthermore, the ability of the system to discriminate between false alarm caused by poor power quality, variations in the driven load level, and actual incipient faults is demonstrated.