As Industry 4.0 enablers, digital twins of manufacturing systems have led to multiple interaction levels among processes, systems, and workers across the factory. However, open issues still exist when addressing cyber–physical convergence in traditional manufacturing small and medium-sized enterprises. The problem for both traditional operators and the existing infrastructure is how to adapt knowledge to the increasing business needs of manufacturing plants that demand high efficiency, while reducing production costs. In this paper, a framework that implements the novel concept of Digital Twin Learning Ecosystem in traditional manufacturing is presented. The objective is to facilitate the integration of human-machine knowledge in different industrial cyber–physical contexts and eliminate existing technological and workforce barriers. This adaptive approach is particularly important in meeting the requirements to help small and medium-sized enterprises build their own interconnected Digital Twin Learning Ecosystem. The contribution of this work lies in a single digital twin learning framework for different traditional manufacturing scenarios that can work from scratch using a light infrastructure, reusing the knowledge and common condition-based methods well-known by skilled workers to rapidly and flexibly integrate existing legacy resources in a non-intrusive manner. The solution was tested using real data from a milling machine and a currently operating induction furnace with a maximum power of 12 MW in a foundry plant. In both cases, the proposed solution proved its benefits: first, by providing augmented methods for maintenance operations on the milling machine and second, by improving the power efficiency of the induction furnace by approximately 9 percent.