This thesis addresses critical challenges in train maintenance work when it comes to safety for production personnel carrying out maintenance tasks. The methodology of the study is based on a mixed method approach, combining elements of quantitative data, through incident reports, and qualitative data, through interviews and study visit. The study aims to investigate how risks in train maintenance workflow can be identified and mitigated through a real time risk identification tool. This tool uses a custom trained YOLO11-model for object detection in order to identify objects in a real-time video stream related to a maintenance task. When a risk is identified the model triggers a pinch warning of squeeze risk and then notifies when the risk is over. The risk identification is based on distance calculation with Euclidean distances. The study also covers possibilities of integrating Augmented Reality (AR) into the risk identification model to enhance a user’s situational awareness and thus mitigate risks associated with a maintenance task. Results from this study show that most of the accidents in the depots is a consequence of lack in safety mindset and an underestimating of risks related to a task. To assess this, the model is effective since it creates contextual understanding for the user to be more aware of the risks associated with the task at hand. Furthermore, AR has the potential of further enhance situational awareness for technicians in a specific task, possibly reducing the acceptance of risk and favors a safety culture of individual responsibility and safety awareness. This is of great interest since related studies identifies contextual awareness as a challenge in current AI-applications. This study shows that risk identification models using object detection combined with AR-enhanced feedback can indeed improve contextual awareness in train maintenance work.