Cyber–Physical Production Systems usher in a new era of smart manufacturing, with Digital Twin being a pivotal concept driving their adoption. Most literature concentrates on global Digital Twin deployment, typically enterprise servers or the cloud. However, manufacturing relies on real-time control and decision-making, demanding swift system response. To address this, we introduce a novel Local Digital Twin architecture, bringing processing to the edge to meet strict timing constraints. We also introduce a Five-dimension GDT/LDTs model aligned with RAMI4.0 for practical application and integration. The LDT was implemented on a multinational company’s Brazilian subsidiary assembly line, predicting optimal assembly components. Machine learning techniques were employed, improving the First Pass Yield by 1.3 to 2.5% in a test machine, affirming the feasibility of the LDT approach.