<p>The concept of the Digital Twin (DT) has evolved with the advancement of Industry 4.0 technologies. With its ability to facilitate autonomous operations, health monitoring, fault detection, and predictive maintenance, it has tremendous potential in smart manufacturing. Industrial robots (IRs) are the backbone of modern manufacturing shop floors in enabling autonomous operations. Fault diagnostics of engineered systems, including IR, is essential for reliable operations within an unmanned manufacturing shop floor. Reducers, which are designed for precise operations and speed reduction, are one of the core components that significantly impact performance in the presence of faults such as wear, pitting, and cracks. Hence, it is essential to monitor the presence or absence of faults and the corresponding level of severity to reduce machine downtime and plan for predictive maintenance. This article presents a cloud-enabled DT framework demonstrating real-time fault diagnostics of the reducer of IR. The flexspline of the harmonic reducer is a highly flexible component that undergoes loading and is prone to failure. Therefore, this article presents a comprehensive fault diagnosis architecture, particularly for the reducer flexspline in the presence of a defect. A two-stage fault diagnostics model has been demonstrated, where the first stage performs fault classification and the second stage determines the severity of the associated fault using multi-domain features extracted from the motor current signature. The random forest classifier demonstrated 100% prediction accuracy for fault classification, using the top 5 features ranked using feature ranking technique. For severity estimation, a hybrid ensemble model demonstrated an accuracy of 98.21% and 93.35% for teeth height wear and flank wear, respectively. These developed models were embedded within the comprehensive DT framework, achieving real-time, accurate fault diagnostic capability in operational environments.</p>