Characterisation of Imminent System Failure and Replacement Policies

Abstract

Discriminant analysis techniques can select, from a large number of feature variables available, a relatively small subset, which will in combination produce optimum prognostic power. This combination of discriminators will then characterise system failure. Induced Gaussian normality in discriminant scores will lead naturally to logistic failure profiles, which compare favourably with observed failure rates. Replacement policies for systems at risk can be developed based on the overall efficiency of allocation rules or perhaps better their sensitivity to locate failures. In the example considered, the choice of sensitivity (within the neighbourhood of optimum efficiency) appears to have little effect on the total cost of misclassification.

Citation

Jerwood, D., Keller, A.Z. and Georgiakodis, F.A. (1988), "Characterisation of Imminent System Failure and Replacement Policies", International Journal of Quality & Reliability Management, Vol. 5 No. 4, pp. 56-64. https://doi.org/10.1108/eb002913