Data-driven vibration monitoring for predictive maintenance of polymeric components using machine learning

Purpose The purpose of this study is to develop and evaluate a data-driven framework for vibration-based condition monitoring aimed at supporting predictive maintenance of polymeric components in rotating mechanical systems. The work seeks to demonstrate how machine learning techniques can be used to automatically identify crack-related faults based on vibration signals, contributing to improved reliability, early fault detection, and maintenance decision-making in engineering applications. Design/methodology/approach An experimental setup was developed to acquire multi-axial vibration signals from a rotating system containing polymeric components under controlled operating conditions. The collected signals were normalized and processed using principal component analysis for dimensionality reduction. A random forest classifier was then trained to distinguish between intact and crack-induced damaged conditions. Multiple datasets were analyzed to evaluate the robustness and repeatability of the proposed machine learning-based monitoring framework. Findings The results show that the proposed approach can reliably identify crack-related damage based on vibration data, achieving high classification accuracy across multiple experimental datasets. The framework demonstrated robustness to signal variability and noise, indicating its suitability for practical condition monitoring applications. The findings confirm that vibration measurements combined with machine learning can effectively support automated fault detection in polymeric components within rotating systems. Research limitations/implications This study is limited to controlled laboratory experiments and focuses on binary classification of intact and damaged conditions. The proposed framework does not explicitly model crack propagation or fatigue mechanisms. Future research may extend the approach to different materials, operating conditions, damage severities and multi-class fault scenarios, as well as investigate scalability to industrial environments. Practical implications From a maintenance engineering perspective, the proposed framework provides a low-cost and automated solution for vibration-based condition monitoring. By enabling early identification of crack-related faults, the method supports predictive maintenance strategies, reduces unplanned downtime and enhances maintenance planning and asset reliability in rotating machinery applications. Social implications The adoption of data-driven predictive maintenance strategies can contribute to safer and more reliable industrial operations by reducing the likelihood of unexpected failures. Improved maintenance efficiency also supports more sustainable use of resources, lower maintenance costs and reduced environmental impact associated with premature component replacement. Originality/value This study provides an original contribution by integrating vibration analysis and machine learning into a practical framework tailored for predictive maintenance of polymeric components. The value of the work lies in demonstrating how data-driven methods can enhance condition monitoring and maintenance decision-making without requiring complex physical modeling, offering a practical solution for maintenance engineering applications.