This comprehensive review explores data-driven methodologies that facilitate the prognostics and health management (PHM) of centrifugal pumps (CPs) while utilizing both vibration and non-vibration sensor data. This review investigates common fault types in CPs, while placing a specific emphasis on artificial intelligence (AI) approaches, including machine learning (ML) and deep learning (DL) techniques, for fault diagnosis and prognosis. A key innovation of this review is its in-depth analysis of cutting-edge methods, such as adaptive thresholding, hybrid models, and advanced neural network architectures, aimed at accurately predicting the remaining useful life (RUL) of CPs under varying operational conditions. This review also addresses the limitations and challenges of the current AI-driven methodologies, offering insights into potential solutions. By synthesizing these methodologies and presenting practical applications through case studies, this review provides a forward-looking perspective to empower industry professionals and researchers with effective strategies to ensure the reliability and efficiency of centrifugal pumps. These findings could contribute to optimizing industrial processes and advancing health management strategies for critical components.

The hydraulic solenoid valve is an essential electromechanical component used in various industries to control the flow rate, pressure, and direction of hydraulic fluid. However, these valves can fail due to factors like electrical issues, mechanical wear, contamination, seal failure, or improper assembly; these failures can lead to system downtime and safety risks. To address hydraulic solenoid valve failure, and its related impacts, this study aimed to develop a nondestructive diagnostic technology for rapid and accurate diagnosis of valve failures. The proposed approach is based on a data-driven model that uses voltage and current signals measured from normal and faulty valve samples. The algorithm utilizes a convolutional autoencoder and hypersphere-based clustering of the latent variables. This clustering approach helps to identify patterns and categorize the samples into distinct groups, normal and faulty. By clustering the data into groups of hyperspheres, the algorithm identifies the specific fault type, including both known and potentially new fault types. The proposed diagnostic model successfully achieved an accuracy rate of 98% in classifying the measurement data, which were augmented with white noise across seven distinct fault modes. This high accuracy demonstrates the effectiveness of the proposed diagnosis method for accurate and prompt identification of faults present in actual hydraulic solenoid valves.