Digital twin based monitoring and control for DC-DC converters

DC-DC converters are critical components in power electronics applications including electric vehicles, DC microgrids, and renewable energy systems. Their nonlinear characteristics make control and state monitoring challenging, motivating the need for online state estimation and robust control strategies. Digital twins (DTs), dynamic virtual replicas of physical systems, have emerged across smart manufacturing, thermal power plants, driver assistance systems, and digitized laboratories, and are increasingly studied in power electronics for condition and health monitoring. Prior DT efforts for converters have focused largely on monitoring component degradation and fault diagnosis rather than control. This study addresses the research question of whether a DT approach can provide effective real-time monitoring and control for DC-DC converters, including tracking performance, adaptation to system variations, controller failure handling, and generalization across different power devices. A DT-based buck converter system is explored to demonstrate these capabilities from a control perspective.

Several observer-based and fuzzy state estimation methods have been applied to nonlinear circuits to address nonlinearity and obtain accurate state estimates (e.g., fuzzy observer-based estimation with slack variables and switching multi-instant fuzzy observers). DT technologies have seen broad adoption in Industry 4.0, thermal power, human-robot interaction, and digitized laboratories. In power electronics, DTs have primarily focused on condition monitoring, including health indicator estimation for components like capacitors and MOSFETs, and fault diagnosis in photovoltaic systems and buck converters. Surveys have identified DT concepts, applications, challenges, and trends for power-electronics-based energy conversion. However, less attention has been paid to DTs from a control perspective for DC-DC converters, particularly for dynamic tracking, adaptation to parameter variations (e.g., input voltage fluctuations in renewables), and controller redundancy/replacement. This work positions itself to fill this gap by providing a DT-based control framework and validation across practical scenarios.

System overview: The proposed DT system comprises a physical buck converter (hardware) and a digital twin implemented in real time using RT-LAB. Unlike HIL approaches that only emulate the converter, the DT is fed by live data from the physical system and can both monitor and control the physical plant when needed.

Physical model and state-space formulation: A standard buck converter is modeled via state-space averaging across switch ON/OFF states. States are inductor current i_L and capacitor voltage u_c, with input voltage V_in, inductance L, capacitance C, and load resistance R. The continuous-time averaged model is: d/dt [i_L; u_c] = [[0, -1/L]; [1/C, -1/(RC)]] [i_L; u_c] + [1/L; 0] V_in.

Hybrid twin modeling: To handle parameter variations (e.g., fluctuating V_in; drift in R, L, C), the continuous model is discretized to a time-varying form where A_d and B_d (including V_in) may change over time. A mechanism–data hybrid strategy is adopted: the mechanism model captures physical principles; DT parameters are adapted online using a Kalman filter-based recursive parameter/state estimation algorithm. The estimator updates parameter vector θ in real time: θ̂(t) = θ̂(t−1) + K(t)(y(t) − ŷ(t)), with gain K computed via Q(t) and covariance updates dependent on noise statistics (R1, R2). This provides noise-robust estimation and low computational complexity for the two-variable model; for higher-order nonlinear systems, feedback linearization can reduce complexity.

Data flow and synchronization: The physical system streams i_L, u_c, and control input u (duty cycle) to the DT. The DT updates model parameters online to maintain synchronization with the physical plant, accommodating parameter drift due to aging or operating condition changes.

Controller redundancy and switching: The architecture features two closed loops—one for the physical controller (DSP-based) and one for the DT controller (running in RT-LAB). The DT continuously monitors for controller failure (e.g., u = 0). Upon detection, the DT controller is activated to control the physical converter, ensuring continuity of operation and serving as a redundant controller.

Simulation setup: Four cases are simulated: (I) reference tracking, (II) input voltage variation, (III) controller failure and DT takeover, (IV) generalization with SiC MOSFETs and higher switching frequency. Parameters: V_in: 100 → 90 → 80 V; L = 2 mH; C = 3300 μF; R = 20 Ω; reference 48 V; IGBT switching 10 kHz (Cases I–III), SiC MOSFET switching 100 kHz (Case IV). DT time step: 0.1 ms.

Experimental setup: Physical controller: DSP TMS320F28335. Power source: IT-M3123. Switch devices: IGBT intelligent power module PM50RLA120 (Cases I–III) and SiC MOSFET NTH4L040N120SC1 (Case IV). The mechanism model and DT run in RT-LAB; the physical system is a buck converter under mechanism-model control unless DT takeover is triggered. The DT uses the same 0.1 ms time step as simulations. Three-dimensional CAD is not required for DT operation as the essential dynamics are captured via measured signals.

• Case I (reference tracking): When the reference voltage is changed from 0 V to 48 V and then to 20 V, the DT system closely tracks the physical buck converter dynamics, outperforming the mechanism model during transients, which exhibits oscillations likely due to unmodeled parasitics. Both simulation and experimental results confirm superior DT tracking.
• Case II (system model variation): With input voltage stepped 100 → 90 → 80 V, the DT adapts its duty cycle and model parameters to maintain output regulation and match the physical system, whereas the mechanism model (built at fixed V_in) fails to adapt, leading to steady-state errors. This demonstrates DT’s online adaptation capability.
• Case III (controller failure): Upon detecting controller failure (control input μ = 0), the DT controller rapidly takes over control, preserving converter operation. Simulation and experimental results validate effective failure detection and seamless controller replacement.
• Case IV (generalization with SiC MOSFET, 100 kHz): Replacing IGBT (10 kHz) with SiC MOSFET (100 kHz) shows the DT maintains dynamic tracking and performs effective controller takeover. Experimental results indicate output voltage error between physical and DT systems of approximately 1.0 V in the SiC case, improved versus about 1.6 V in the IGBT case, evidencing robust generalization across devices and switching frequencies.
• Overall: DT achieves dynamic synchronization with the physical system, tracks parameter variations, and enhances reliability by providing a redundant controller. DT runs with a 0.1 ms time step in both simulation and experiments.

The findings address the core research question by demonstrating that a DT can provide reliable real-time monitoring and control for DC-DC converters. The DT’s hybrid modeling and Kalman filter-based adaptation enable it to track the physical converter’s dynamics and accommodate parameter changes such as fluctuating input voltage. The ability to detect physical controller failure and replace it with the DT controller enhances system reliability and availability. Experimental results align with simulations, supporting the approach’s validity. Generalization to different semiconductor devices (IGBT vs. SiC MOSFET) and switching frequencies underscores the method’s robustness and applicability to practical power electronics scenarios. Compared to mechanism-only models, the DT reduces sensitivity to unmodeled parasitics and parameter drift, providing improved transient performance and monitoring fidelity.

This work designs and implements a digital twin for DC-DC buck converters that enables real-time monitoring, adaptive control, and controller redundancy. Using a mechanism–data hybrid model with Kalman filter-based parameter adaptation, the DT synchronizes with the physical converter, maintains performance under reference changes and input voltage variations, and seamlessly replaces a failed controller. Simulations and experiments across IGBT (10 kHz) and SiC MOSFET (100 kHz) platforms verify effectiveness and generalization. Future research directions suggested by the authors include handling sensor failures, advancing system condition diagnosis and prognosis, exploring optimal control strategies, and optimizing DT controller design for more complex nonlinear systems.

• Scope: Validation focuses on a buck converter topology and two switching device types (IGBT and SiC MOSFET); broader converter topologies and operating conditions were not reported.
• Comparison: Direct, quantitative comparison with alternative DT/HIL/IoT-based methods is noted as challenging, limiting cross-method benchmarking.
• Complexity: While computational cost is low for the two-state model, the approach acknowledges that complexity grows with the number of estimated states; more complex nonlinear systems may require techniques (e.g., feedback linearization) to remain tractable.
• Fault coverage: The study addresses controller failure but does not experimentally cover other faults (e.g., sensor failures), which are proposed for future work.