Transforming Predictive Maintenance With Machine Learning: A Comprehensive Review of Innovations and Applications in Centrifugal Pumps

Centrifugal pumps are critical assets across food and beverages, oil and gas, water treatment, and power generation, where reliability and efficiency are essential to minimizing downtime and economic losses. Traditional maintenance strategies, corrective and preventive, often fail to capture real‑time equipment conditions, leading to unexpected failures and reduced availability. Predictive maintenance (PdM), powered by machine learning (ML), offers a data‑driven alternative by enabling early fault detection, optimized scheduling, and improved equipment lifespan. This review synthesizes recent innovations in ML techniques applied to centrifugal pump diagnostics, including supervised, unsupervised, and hybrid models. Methods such as Support Vector Machines (SVM), Random Forests (RF), Artificial Neural Networks (ANN), Principal Component Analysis (PCA), and Long Short‑Term Memory (LSTM) networks are examined alongside signal processing approaches like Complementary Ensemble Empirical Mode Decomposition (CEEMD) and Sample Entropy (SampEn) for feature extraction and fault characterization. From 146 publications retrieved, 22 peer‑reviewed studies were selected for detailed analysis based on relevance, methodological clarity, and performance metrics. Findings reveal that ML models achieve high accuracy in fault detection and remaining useful life prediction, yet adoption remains limited by data scarcity, integration challenges in brownfield systems, and the “black‑box” nature that reduces operator trust. Emerging Explainable AI (XAI) techniques improve interpretability, while Internet of Things (IoT)‑enabled retrofitting strategies expand applicability to legacy equipment. A taxonomy is developed to classify ML applications into diagnostic domains, revealing research gaps and guiding future directions. The combination of enhanced sensor data, scalable hybrid architectures, and Total Productive Maintenance (TPM) frameworks, along with XAI and IoT retrofitting, is suggested as a way to make pump systems smarter, easier to understand, and more reliable. This review underscores the transformative potential of ML‑driven predictive maintenance in enhancing reliability, reducing downtime, and mitigating economic losses, while highlighting promising directions for future research.