| description abstract | Condition-based maintenance (CBM) is an emerging technology, which seeks to develop sensors and processing systems aimed at monitoring the operation of complex machinery such as turbine engines, rotor craft drivetrains, and industrial equipment. The goal of CBM systems is to determine the state of the equipment (i.e., the mechanical health and status), and to predict the remaining useful life for the system being monitored. The success of such systems depends upon a number of factors, including: (1) the ability to design or use robust sensors for measuring relevant phenomena such as vibration, acoustic spectra, infrared emissions, oil debris, etc.; (2) real-time processing of the sensor data to extract useful information (such as features or data characteristics) in a noisy environment and to detect parametric changes that might be indicative of impending failure conditions; (3) fusion of multi-sensor data to obtain improved information beyond that available to a single sensor; (4) micro and macro level models, which predict the temporal evolution of failure phenomena; and finally, (5) the capability to perform automated approximate reasoning to interpret the results of the sensor measurements, processed data, and model predictions in the context of an operational environment. The latter capability is the focus of this paper. Although numerous techniques have emerged from the discipline of artificial intelligence for automated reasoning (e.g., rule-based expert systems, blackboard systems, case-based reasoning, neural networks, etc.), none of these techniques are able to satisfy all of the requirements for reasoning about condition-based maintenance. This paper provides an assessment of automated reasoning techniques for CBM and identifies a particular problem for CBM, namely, the ability to reason with negative information (viz., data which by their absence are indicative of mechanical status and health). A general architecture is introduced for CBM automated reasoning, which hierarchically combines implicit and explicit reasoning techniques. Initial experiments with fuzzy logic are also described. | |
