en Other مقاله پژوهشي Research Paper <span style="font-size:11.0pt"><span new="" roman="" style="font-family:" times="">Stern-tube bearing failures can immobilize vessels and discharge lubricants into the sea, yet traditional threshold-based oil monitoring often overlooks the earliest signs of wear. We have developed an AI-driven predictive-maintenance framework that combines routine oil-analysis data with machine-learning models to detect incipient faults. Using a labelled dataset of 437 samples (54.2 % normal, 40.3 % warning, 5.5 % abnormal), we trained Random Forest and CatBoost classifiers; the Random Forest achieved an overall accuracy of 85 %. Applying the ADASYN over sampler raised recall for the rare abnormal class from 0.42 to 0.73. SHAP analysis identified copper and lead (bearing wear) as well as sodium and boron (seal-water ingress and additive depletion) as the most influential predictors. In a field trial on an AHTS-DP1 vessel, the model flagged rising boron six months before a dry-dock inspection confirmed seal damage caused by fishing-net entanglement, enabling timely repair and averting a potential MARPOL violation. Compared with the operator&rsquo;s existing rule-based alerts, the proposed system cut false alarms by 40 % while detecting subtle degradation earlier. These findings show that merging tribological expertise with data-driven analytics can markedly enhance stern-tube reliability and promote more sustainable ship operations.</span></span> Predictive maintenance,Stern tube seals,Oil analysis,Machine learning,SHAP values 20 30 http://ijmt.ir/browse.php?a_code=A-10-8311-1&slc_lang=en&sid=1 Mohammad Ali Mohaghegh m.ali.mohaghegh@gmail.com 10031947532846004576 10031947532846004576 No Faculty of Mechanical Engineering, Sharif University of Technology, Tehran, Iran Mehdi Behzad m_behzad@sharif.edu 10031947532846004575 10031947532846004575 Yes Faculty of Mechanical Engineering, Sharif University of Technology, Tehran, Iran Somaye Mohammadi somaye.mohammadi@sharif.edu 10031947532846004574 10031947532846004574 No Faculty of Mechanical Engineering, Sharif University of Technology, Tehran, Iran