The Application of Machine Learning and Reliability Block Diagram (RBD) in Prognostic Health Management (PHM) for Pumps

Abstract

This study proposes an integrated Prognostics and Health Management (PHM) framework for centrifugal pumps by combining performance data (vibration, temperature, pressure) and functional failure data using Artificial Neural Networks (ANN) and Reliability Block Diagram (RBD). The dataset obtained from operational operating plant reports was preprocessed using PCA and SMOTE to address dimensionally and class imbalance. A NAR-based ANN model was developed to predict time-to-failure and the predicted outputs were integrated into an RBD model for availability estimation. The model achieved an MSE of 0.124 and high correlation with observed degradation trends. The RBD analysis revealed an estimated system availability of 34%, indicating significant downtime and the need for reliability improvement strategies. The proposed hybrid PHM framework demonstrates potential for early degradation detection and maintenance optimization in rotating machinery.