The incorporation of digital twins (DTs) with advanced computational approaches is transforming contemporary manufacturing by enabling real-time condition tracking, fault detection, and process optimization. However, many data-driven models used in such systems often lack clarity, raising concerns about their dependability when applied to edge devices that need quick and comprehensive outcomes. This chapter presents a lightweight interpretability-focused framework designed for digital twin systems operating on edge hardware in industrial environments. The techniques employ transparent models like decision trees, logistic regression, and a gradient boosting method enhanced with feature attribution methods like SHAP, all optimized for performance on low-power devices. The architecture is tested utilizing an open industrial dataset, with evaluations centered on prediction quality, interpretability, and computational performance. The outputs suggest that these streamlined models provide a functional balance between system insight and responsiveness, making them suitable for practical deployment in next-generation manufacturing systems. The proposed solution enhances trust, transparency, and adaptability in smart factory applications.