Designing Health Monitoring System for fault prediction in Centrifugal Pump using ML and DL algorithm

Abstract

The operational dependability of centrifugal pumps holds paramount significance within industrial applications, thereby necessitating the implementation of sophisticated health monitoring systems to proactively identify potential failures. This investigation introduces an innovative Health Monitoring System (HMS) meticulously engineered for centrifugal pumps, utilizing Machine Learning (ML) and Deep Learning (DL) algorithms to augment predictive maintenance capabilities. Our framework amalgamates real-time data acquisition, rigorous feature extraction, and advanced predictive analytics to effectively evaluate pump health. By employing a comprehensive dataset that encompasses operational parameters alongside historical failure records, the research includes executing a variety of ML algorithms, which include Logistic Classifier, Decision Tree Classifier, XGB Classifier, Naïve Bayes Classifier, Random Forest, Support Vector Classifier, and Neural Network, for the classification of centrifugal pump health status. Moreover, the research accentuates the application of sophisticated deep learning techniques, particularly Artificial Neural Networks (ANN), to discern intricate patterns and temporal relationships within the data. Experimental results demonstrate that our proposed HMS markedly enhances fault detection accuracy, accomplishing classification rates that surpass 99%. This system not only enables timely maintenance interventions but also mitigates operational downtime and maintenance expenditures. The research elucidates the potential of advanced AI-driven methodologies in the formulation of intelligent monitoring solutions, ultimately contributing to the sustainability and reliability of centrifugal pump operations in industrial contexts.