Fault Diagnosis Techniques for Power Electronics Systems

Fault diagnosis in power electronics systems has become a critical research area in response to the growing demand for reliable converters and inverters in applications ranging from renewable energy to electric vehicles and industrial drives. Traditional approaches rely on signal-processing and model-based methods, exploiting discrepancies between measured and estimated waveforms to identify open-circuit and short-circuit faults in semiconductor switches. In recent years, data-driven techniques have come to the fore, applying machine learning and deep learning algorithms to extract diagnostic features directly from voltage and current measurements. Hybrid strategies that combine physics-based models with intelligent classifiers have further improved robustness under variable operating conditions. Overall, these methods aim to achieve early detection, precise localisation and rapid classification of faults to enable prompt mitigation, reduce unplanned downtime and enhance the lifetime of power electronic equipment.

Research from Nature Portfolio

No recent Nature Portfolio content available.

Research from all publishers

Recent advances from diverse outlets have demonstrated the power of deep learning for real-time diagnosis. One study introduced a modified convolutional neural network incorporating a global average pooling layer to replace traditional fully connected layers. By transforming one-dimensional current signals into two-dimensional feature maps and applying successive convolution and pooling operations, the model achieved diagnostic accuracies of up to 99.95 % for open-circuit faults in DC–DC inverters, while reducing computation time by over 15 %. A second effort developed a hybrid-CNN architecture for three-phase inverters, fusing one-dimensional and two-dimensional convolutional branches to capture temporal and spatial signatures under varying load and speed conditions; this approach yielded superior classification accuracy across multiple fault modes. A further contribution applied a deep CNN to hybrid active neutral-point-clamped inverters, training on distorted current profiles to identify open-switch faults with 99.6 % accuracy and sub-millisecond latency, without additional sensing hardware. Collectively, these studies highlight the global shift towards automated, high-fidelity fault diagnosis capable of scaling to the complexity of modern power conversion topologies.

Fault Diagnosis Techniques for Power Electronics Systems publication trend

The graph below shows the total number of articles in fault diagnosis techniques for power electronics systems across all publications each year (not limited to Nature Index journals).

Technical terms

Inverter: A power electronic converter that transforms direct current to alternating current, commonly used in motor drives and grid-connected renewable sources.

IGBT (Insulated Gate Bipolar Transistor): A semiconductor switch combining high-input impedance and high-current capability, widely employed in inverter legs.

Convolutional Neural Network (CNN): A deep learning architecture that automatically extracts hierarchical features from input signals or images via convolution and pooling layers.

Wavelet Transform: A signal-processing technique that decomposes time-series data into time–frequency components for fault feature extraction.

Global Average Pooling (GAP): A neural network layer that reduces spatial dimensions by averaging each feature map, lowering parameter count and overfitting risk.

References

1. A Data-Driven-Based Fault Diagnosis Approach for Electrical Power DC-DC Inverter by Using Modified Convolutional Neural Network With Global Average Pooling and 2-D Feature Image. IEEE Access (2020).
2. Adaptive feature extraction and fault diagnosis for three-phase inverter based on hybrid-CNN models under variable operating conditions. Complex & Intelligent Systems (2021).
3. Fault Detection Method Using a Convolution Neural Network for Hybrid Active Neutral-Point Clamped Inverters. IEEE Access (2020).