In the progressive context of industrial operations, the uninterrupted functioning of machines is crucial for ensuring optimal productivity. The estimation of Remaining Useful Life (RUL) in the field of Predictive Maintenance (PdM) is a vital task to minimize downtime, enhance operational efficiency, and reduce unexpected equipment failure. Predictive maintenance utilizes advanced analytic tools for analyzing operational and repair records to forecast potential failures in machine equipment before they occur.

However, due to the high quality of equipment, failure times are often undetected during observation, resulting in a large amount of censored data. Machine Learning-based Survival Analysis (ML-based- SA) models have emerged in recent years to address the challenge of censored data and predict the survival time of an instance. Nevertheless, the inherent black-box nature of ML models presents a significant challenge for domain experts in understanding the rationale behind the predictions. To address this issue, eXplainable AI (XAI) has gained increasing attention in academia and industry to enhance transparency and interpretability in the predictive models.

To the best of our knowledge, there is a scarcity in the application of XAI to ML-based-SA prediction models, especially within the PdM domain. To address this research gap, this study aims to enhance the transparency and interpretability of predictive maintenance models by integrating XAI techniques with ML-based-SA. The research utilized industrial real-world data from SCANIA AB to identify the best-performing ML-based survival analysis model for predicting the RUL of heavy vehicle components. Additionally, the study aims to identify the critical factors influencing RUL predictions through XAI techniques, thereby enhancing the transparency of predictive maintenance models in industrial practices.

Utilizing the experimental research methodology, three ML-based-SA were developed and evaluated by Concordance Index as evaluation metric. The results revealed that Random Survival Forest (RSF) outperformed Gradient Boosting Survival Analysis (GBSA) and Cox Proportional Hazards (CPH). Therefore RSF undergoes explainability analysis. In this study, we utilized SHAP analysis as XAI technique, that is capable of performing both global and local explanations in addition to offering insights into the direction, amount, and dependency of features. The influential features significantly affecting the overall predictions of RSF model are identified and their impact on the prediction output of two individual vehicles were discussed. These findings contribute to enhancing trust, understanding, and ultimately improving maintenance strategies in industrial settings.