On-line reliability estimation of individual components, using degradation signals

This paper provides a unique approach that allows ´determination of a component´s reliability as it degrades with time´ by monitoring its degradation measures. The concepts have been implemented using: finite-duration impulse response multi-layer perceptron neural networks for modeling degradation measures, and self-organizing maps for modeling degradation variation. The specific application considered is in-process monitoring of the condition of the drill-bit in a drilling process, using the torque and thrust signals. An approach to compute prediction limits for any feedforward neural network, critical for on-line performance reliability monitoring of systems using neural networks, is introduced by combining the network with a self-organizing map. Experimental results show that neural networks are effective in: modeling the degradation characteristics of the monitored drill-bits, and predicting conditional and unconditional performance reliabilities as they degrade with time or usage. In contrast to traditional approaches, this approach to on-line performance reliability monitoring opens new avenues for better understanding and monitoring systems that exhibit failures through degradation. Essentially, implementation of this ´performance reliability monitoring´ reduces overall operations costs by facilitating optimal component-replacement and maintenance strategies