Stochastic Modelling and Availability Prediction of Emergency Shutdown System Using Supervised Learning Algorithms Subject to 2‐Out‐of‐3 Redundancy

ABSTRACT

The prominent objective of present study is to predict the steady state availability of an Emergency Shutdown (ESD) System extensively used in industries such as oil, gas etc. under the concept of redundancy. The ESD is a complex system having five subsystems including sensor units in 2‐out‐of‐3: G redundancy. For this purpose, a novel stochastic model is developed using Markovian approach and governing equations of transition probabilities are derived. All the random variables are statistically independent and exhibit constant behaviour. The sensitivity of the proposed model is carried out by taking 50% variations in the failure and repair rates of all the subsystems and observing the corresponding behaviour of ESD system availability. Further, supervised learning techniques namely regression analysis (RA) and artificial neural networks (ANN) are employed to predict the availability of the ESD system. It is observed that ANN outperforms RA in terms of all three‐performance metrics, exhibiting a higher value and lower MSE and RMSE values based on the simulated data. The performance measures are shown numerically and graphically. The maintenance engineers may use these findings in planning maintenance strategies.