Abstract
Practical early fault detection and diagnosis systems must exhibit high level of detection accuracy and while exhibiting acceptably low false alarm rates. Such designs must have applicability to a large class of machines, require installation of no additional sensors, and require minimal detailed information regarding the specific machine design. Electromechanical systems, such as electric motors driving dynamic loads like pumps and compressors, often develop incipient failures that result in downtime. There is a large number of such failure modes, with a large majority being of mechanical nature. The precise signatures of these failure modes depend on numerous machine-specific factors, including variations in the electric power supply and driven load. In this paper the development and experimental demonstration of a sensorless, detection and diagnosis system is presented for incipient machine faults. The developed fault detection and diagnosis system uses recent developments in dynamic recurrent neural networks in implementing an empirical model-based approach, and multi-resolution signal processing for extracting fault information from transient signals. The signals used by the system are only the multi-phase motor current and voltage sensors, whereas the transient mechanical speed is estimated from these measurements using a recently developed speed filter. The effectiveness of the fault diagnosis system is demonstrated by detecting stator, rotor and bearing failures at early stages of development and during different levels of deterioration. Experimental test results from small machines, 2.2 kW, and large machines, 373 kW and 597 kW, are presented demonstrating the effectiveness of the proposed approach. Furthermore, the ability of the diagnosis system to discriminate between false alarms and actual incipient failure conditions is demonstrated.
