A new approach to the challenge of machinery prognostics

Current generation mechanical diagnostic equipment is designed to identify individual events or trends in the output of sensors mounted on a mechanical component, subsystem, or system. Such equipment can provide a useful indication that a failure condition may be developing, but it cannot provide reliable prediction of the remaining safe or operational life. Typically, these diagnostic systems simply compare the output of individual sensors against a priori thresholds to establish a measure of the system's health. Two problems result from this approach: (1) There is no advantage taken of possible synergy among the sensors, i.e., the determination of health is one dimensional; and (2) the diagnosis provides only a statement regarding the current equipment health, but does not provide a prediction of the time remaining to failure. This often leads to an operational environment in which diagnostic equipment outputs are either ignored because of frequent false alarms or frequent (and costly) time-based preventive maintenance is performed to avoid hazardous failures. This paper describes a new approach to the development of a more robuts diagnosis and prognostic capability. It is based on the fusion of sensor-based and model-based information. Sensor-based information is the essence of current diagnostic systems. Model-based information combines dynamic models of individual components with micromechanical models of relevant failure mechanisms, such as fracture and crack propagation. These micromechanical models must account for initial flaw size ditribution and other microstructural parameters describing initial component condition. A specific application of this approach is addressed, the diagnosis of mechanical failure in meshing gears. Four specific issues are considered: (a) how to couple a validated numerical simulation of gear transmission error (due to tooth spacing irregularity, or material inhomogeneity) with physically and empirically based descriptions of fatigue crack grawth to predict a failure precursor signature at the component level; (b) how to predict the manifestation of this signature at the subsystem or system level where sensors are located; (c) how to fuse this model-based information with the corresponding sensor-based information to predict remaining safe or operational life of a gear; and (d) issues associated with extending this methodology to bearings and other rotating machinery components