Digital twin (DT) has gained increasing attention in different areas due to its capability of replicating the behaviors and performance of the physical system. It is emerging as a promising tool for optimizing the design, control, and maintenance of power electronic systems. This article proposes a machine learning (ML)-enabled framework for DT model development and noninvasive parameter estimation of voltage source converters (VSCs). A DT model of the converter system is first established, and extensive parameter sweeping simulations under various operating conditions are conducted to generate a comprehensive dataset for an artificial neural network (ANN). The trained ANN identifies key component parameters directly from measurements without additional hardware, thereby bridging the DT model and the physical system. These parameters are then used to update the DT model, ensuring high fidelity and synchronization with the physical counterpart. Experimental results confirm the effectiveness of the proposed framework, achieving accurate parameter estimation and high-fidelity DT. Finally, a DT-based fault diagnosis case study is demonstrated, highlighting its practical applicability.