Quantitative risk assessment of submersible pump components using Interval number-based Multinomial Logistic Regression(MLR) model

Abstract

This paper aims at developing a methodology to explore the association among the risk factors affecting failure, concerning Potential Failure Mode (PFMs) using Multinomial Logistic Regression (MLR). MLR model helps in assigning different weights to the risk factors in the form of MLR equations, where the response variable ‘Failure’ is treated as a categorical variable with three defined levels (Critical ‘1′; Medium ‘2′ & Less Critical ‘3′). This model appointed three experienced decision-makers to articulate their opinions using linguistic variables, expressed as interval numbers about PFMs concerning nine risk factors, unlike the traditional approach, and provides flexibility to accommodate as many risk factors as possible. The proposed approach is demonstrated with the help of a case study involving failures in components of a submersible pump used in a power plant. The model's probability of critical failure (‘1′) prediction is 86.3% and 83.4% with training and test data respectively. Furthermore, the model showed a higher overall accuracy of 91% with training data. Also, sensitivity analysis, specificity, and ROC curve are carried out to validate the model. The proposed methodology shows PFM 4 (‘O’ ring) with the highest probability of failure, followed by PFM 9 (gasket).