Industry 5.0 promotes the development of human-centered industrial operations fueled by a fresh wave of disruptive technologies that encourage synergistic human-machine integration. Its focus is on understanding how human cognition contributes to a more secure and harmonious coexistence between humans and machines in industrial scenarios, employing solutions that prioritize fundamental worker demands while preserving or enhancing industrial productivity. In this context, the ability to assess fatigue objectively is crucial for occupational health and safety because it can reduce cognitive and motor function, ultimately lowering productivity and raising the risk of harm to human operators. To this end, wearable systems provide a promising solution for continuous, non-intrusive, and long-term monitoring of biological signals for fatigue detection. However, the adoption of these devices presents unique challenges, such as inter-individual variability that renders traditional one-size-fits-all machine learning models unsuitable. This paper provides an analysis of the current state-of-the-art for wearable device monitoring, including ongoing issues and current knowledge gaps. In addition, an experimental analysis is presented, employing a pattern discovery pipeline based on unsupervised learning on a real-world dataset. Our analysis provides experimental evidence of the limitations of one of the classical approaches to fatigue assessment, thus highlighting the need for more advanced models.