ESTIMATION OF COMPONENTS RELIABILITY IN PETROCHEMICAL PLANTS USING A NEURAL-WEIBULL LIFETIME MODEL

Abstract

Reliability estimation is required for developing a proactive maintenance strategy for timely replacement of aging components in petrochemical plants. This paper presents a new generalized Weibull model to estimate the reliability of components in petrochemical plants. The model includes an artificial neural network that is trained using empirical values of cumulative failure rates. In a number of simulations, the neural-Weibull model is compared to generalized Weibull models recently reported in the literature to estimate the reliability and failure rates of three classes of valves in different petrochemical applications. The results show that in contrast to existing models, the proposed neural-Weibull model is capable of learning arbitrary shapes of empirical reliability plots and also significantly outperforms the existing reliability models. Furthermore, the average error of failure rate estimation using the neural-Weibull model is observed to be at least five times lower than the estimation error of the most recent model in the literature.