Work accidents, especially in the construction sector, are still a serious problem, with fatigue as one of the main causes. Based on data from satudata.kemnaker, the author realizes the need for early prevention solutions to reduce the risk of accidents due to fatigue. One of the approaches proposed is the development of an automatic detection system to recognize workers' facial expressions, especially in detecting levels of freshness and sleepiness. The obstacles that are often faced are limited time and scale in manual monitoring, especially on large-scale construction projects. To overcome this, the You Only Look Once (YOLO) algorithm is used, which is able to detect objects quickly and accurately, to provide continuous monitoring of workers' conditions. This research focuses on the application of the YOLOv8n model in an automatic freshness and sleepiness facial expression detection system. The model is trained using a dataset that includes a variety of facial expressions in different situations, allowing the system to detect worker conditions in real-time and at scale. The evaluation results in this research show very good performance, with precision reaching 99.9%, recall 100%, mAP50 99.5%, and mAP50-90 97.9%. Although the model sometimes makes mistakes in object class recognition, the overall results still show a very high level of accuracy. With this system, it is hoped that it can improve work safety through early detection of signs of fatigue in workers, so that the potential for work accidents can be significantly minimized.