Extended Reality (XR) technologies, including Virtual Reality (VR), Augmented Reality (AR), and Mixed Reality (MR), enhance industrial training by enabling immersive, interactive learning. Unlike traditional methods, XR allows for safe, repeatable simulations of complex scenarios. However, ensuring adaptability across diverse user skills, cognitive preferences, and hardware configurations remains a challenge. This paper presents a modular system architecture for adaptive XR training environments that dynamically adjust training scenarios based on user profiles, interaction patterns, and hardware capabilities. The proposed system integrates semantic reasoning techniques to analyze user data and modify scene elements accordingly. Key contributions include a classification-based adaptation mechanism that personalizes training content, a semantic reasoning engine for optimizing training scenarios, and a flexible architecture that ensures accessibility across VR and non-VR setups. A case study in an industrial training context demonstrates the effectiveness of this approach in enhancing user experience and learning outcomes. Future work aims to integrate AI-driven automation to further refine adaptive XR training environments.