Digital twin and artificial intelligence (AI) technologies are considered key enablers for Industry 4.0 and sustainable intelligent manufacturing. The usage of digital twins in manufacturing processes leads to an enhancement in process understanding due to the continuous processing and evaluation of production data. Hence, when properly implemented, it empowers the ability to increase the agility, traceability, and resilience of production systems. Despite these merits, the scalability and fidelity of digital twins at the manufacturing field level are limited by heterogeneous data sources as well as modeling uncertainties, e.g., due to coarse approximation and imprecise parameter identification. This study proposes a novel approach to overcome these deficiencies by exploiting machine learning on aggregated data throughout the product development process. A procedure is established to contextualize metadata sources from the construction, planning, manufacturing, and quality stages into a quality feature-based digital thread. On this basis, a hybrid approach for quantifying modeling uncertainties across the model chain is developed by incorporating the production domain knowledge into AI techniques. As a result, the digital twin takes into account disturbances from the physical manufacturing environment and estimates the reliability of the modeled outcomes. An exemplary implementation of this concept is demonstrated by using a cutting-edge digital process twin for machining quality monitoring.