Physics-informed machine learning (PIML) has been emerging as a promising tool for applications with domain knowledge and physical models. To uncover its potentials in power electronics, this article proposes a PIML-based parameter estimation method demonstrated by a case study of dc–dc Buck converter. A deep neural network and the dynamic models of the converter are seamlessly coupled. It overcomes the challenges related to training data, accuracy, and robustness which a typical data-driven approach has. This exemplary application envisions to provide a new perspective for tailoring existing machine learning tools for power electronics.

One of the future challenges in modular multilevel converters (MMCs) is how to size key components with compromised costs and design margins, while fulfilling specific reliability targets. It demands better thermal modeling compared to the state-of-the-arts in terms of both accuracy and simplicity. Different from two-level power converters, MMCs have inherent dc-bias in arm currents and the power device conduction time is affected by operational parameters. A time-wise thermal modeling for the power devices in MMCs is, therefore, an iteration process and time-consuming. This paper, thus, proposes a simple analytical thermal modeling method, which adopts equivalent periodic power loss profiles. More importantly, time-domain simulations are not required in the proposed method. Benchmarking of the proposed methods with the prior-art solutions is performed in terms of parameter sensitivity and model accuracy with a case study on a 30-MW MMC system. Experiments are carried out on a specifically designed scaled-down system to verify the electrothermal aspects.

A proper thermal network model (TNM) facilitates the thermal analysis of multichip power modules and is beneficial to the reliability evaluation of power converters. However, many TNMs are struggling to provide the accurate thermal analysis while maintaining acceptable computational efficiency. To address this, a TNM based on the Cauer model is proposed in this article, where the Cauer model is obtained with the assistance of the structure function. Then, the thermal coupling effects in multichip power modules are fully characterized. More specifically, the thermal coupling in the proposed TNM is equivalent to the injected power losses from the specific nodes, and an enumeration method is accordingly used to identify the injected nodes and the injected power losses. Moreover, a discretization approach based on the forward Euler method is implemented to enhance the computational efficiency. Subsequently, the junction temperature of the multichip power module is estimated through the signal-flow graph method. Finally, the proposed TNM uses the typical insulated-gate bipolar transistor modules as a case study, and extensive simulations and experimental tests are conducted to validate the effectiveness of the proposed TNM.

This letter proposes an artificial intelligence-aided thermal model for power electronic devices/systems considering thermal cross-coupling effects. Since multiple heat sources can be applied simultaneously in the thermal system, the proposed method is able to characterize model parameters more conveniently compared to existing methods where only single heat source is allowed at a time. By employing simultaneous cooling curves, linear-to-logarithmic data re-sampling, and differentiated power losses, the proposed artificial neural network-based thermal model can be trained with better data richness and diversity while using fewer measurements. Finally, experimental verifications are conducted to validate the model capabilities.

This article proposes a mission profile-based reliability prediction method for modular multilevel converters (MMCs). It includes key modeling steps, such as long-term mission profile, analytical power loss models, system-level and component-level thermal modeling, lifetime modeling, Monte Carlo analysis, and redundancy analysis. Thermal couplings and uneven thermal stresses among submodules are considered. A case study of a 15-kVA down-scale MMC has been used to demonstrate the proposed method and validate the theoretical analysis. The outcomes serve as a first step for developing realistic reliability analysis and model-based design methods for full-scale MMCs in practical applications.

Junction temperature monitoring (JTM) is essential for reliability evaluation and health management for insulated-gate bipolar transistor (IGBT) modules, and thus is extensively focused on in power electronics converters. However, many JTM methods for IGBT modules are criticized for providing inaccurate junction temperature information with strong load current dependence. To address this, a JTM method based on the turn- off voltage (TOV) and convolutional neural networks (CNN) is proposed in this article. In this method, the TOV is used as the junction temperature indicator, and the characterization behavior of the TOV during turn- off transient process is thoroughly analyzed. Then, the parameter dependence of the TOV is investigated. Considering the proposed JTM method may be subject to the issue of load current dependence and show an undesirable performance, the CNN is adopted to maintain the accuracy of junction temperature prediction due to its excellent global and local feature recognition capability. With this regards, the proposed JTM method enables to provide accurate junction temperature information under different conditions. Finally, the double-pulse tests and experimental tests are carried out to validate the effectiveness of the proposed JTM method.