Data-driven methods have shown promising performance in power converter fault diagnosis. However, the existing methods are bothered by the inadequacy of accuracy and robustness, due to the feature variation in the different operating points, especially in nonstationary fault signals. This article proposes a robust data-driven fault diagnosis method combining automatic feature extraction (AFE) and improved random forests (IRFs). In AFE, this method uses optimal discrete wavelet transform, optimized by the defined information component function, to extract multiresolution time-frequency features from nonstationary fault signals, and the improved ReliefF algorithm is used to select important fault features considering both the independent and interactive weights to overcome nonstationary influence. To enhance the robustness, the IRFs adjusts the weight of decision trees by robustness testing based on out-of-bag data, which extracts the mapping relationship between fault features and fault labels for fault location. The comparative experimental results show that the proposed method achieves 99.94% fault diagnosis accuracy, and has superior fault diagnosis robustness compared to existing methods.

Inverter systems play a crucial role in aerospace, defense, transportation, modern industry, and power systems, leading to extensive efforts from scholars and engineers in fault diagnosis. Data-based methods are widely utilized with the accessible history data instead of complex math modeling for this issue but they are incompetent for obstinate incipient fault. Therefore, this article proposes an intensive multiorder feature extractor (IMFE) for the incipient fault detection of inverter system, with intensively extracting deep statistical features and reducing harmful perturbations. First, a dense structure with short paths between nonadjacent layers is adopted for multiorder knowledge reutilization. Then, the acquired features are refined and the low-quality information is discarded. In addition, the effectiveness of IMFE is demonstrated through rigorous mathematical derivation with sensitivity and complexity analysis. Finally, a three-phase inverter system platform based on current and voltage dual control is established to verify the superiority of the proposed method. Experimental results show that the proposed approach significantly improves fault detection performance, achieving a 3.1% higher fault detection rate compared to existing state-of-the-art methods.

Bearings are important part of aerospace equipments. Therefore, bearings fault diagnosis and fault detection is necessary for the safe operation. Generally, the bearing are core component and runs in a complex and heavy background noise environment. Thence, the bearing signals will be overwhelmed by noise. At present, the diagnosis methods of bearings are based on the experience of experts which costs a lot of labor and time-consuming. Consequently, to address these inadequacies, a novel method based on variational mode decomposition (VMD) algorithms and Alexnet neural network has been presented. Firstly, it decomposes the nonstationary bearing signals into intrinsic mode functions adaptively. However, background noise will seriously affect the number of signal decompositions and reduce accuracy even over-decomposition. Thence, this paper proposed the hierarchical fuzzy entropy method to adaptively extract weak fault characteristics of bearings. Then Alexnet neural network has been utilized to learn the relationship of fault features and bearings health conditions. The architecture of Alexnet diagnosis model is convenient and efficient. Finally, two tentative dataset are adopted to verify the effectiveness and feasibility of the method proposed in this article. The results show that the presented method has a higher fault diagnosis accuracy rate than traditional convolutional neural networks.

As the core component of modern power electronic systems, power converters have widespread applications in renewable energy generation, smart grids, and aerospace. The open-circuit fault of power converters is highly concealed and easily derived from secondary faults, whose diagnosis technology has become a key research direction to ensure safe operation. Considering that power converters usually need to operate under different load conditions and complex environments, it is of great significance to achieve efficient open-circuit fault diagnosis while protecting data privacy. Therefore, this article proposes an open-circuit fault diagnosis approach for power converters from the perspective of source-free domain adaptation. In detail, this article proposes an attention-guided power converter fault diagnosis model to enhance the ability to capture key features. Without the need for source domain data to participate in joint training, this article proposes a dynamic clustering loss to achieve pseudolabel generation and iterative self-training, so that the model can make full use of target domain data. Furthermore, this article proposes uncertainty entropy loss and diversity entropy loss to decrease the uncertainty of fault diagnosis output while preventing the model from being over-confident. To demonstrate the reliability of the open-circuit fault diagnosis methodology, this research implements a functional power conversion system.

The data-driven fault diagnosis method, which eliminates the need for additional sensors while preserving the system's original structure, offers a promising approach to enhancing converter reliability. However, diagnosing multiple device faults presents significant challenges due to difficulties in fault sample acquisition, detectability uncertainty, and unexplained decisions. This article addresses these challenges by presenting a detectability-based data-driven fault diagnosis method. First, a model-based fault detectability analysis method is proposed to establish the measurement conditions necessary for reliably detecting various fault types. Utilizing these measurement condition constraints, a mechanism-enhanced neural network is designed to locate faults by fitting the changes in fault parameters. The consistency between the fitting fault parameters and the actual fault process ensures the interpretability of the diagnosis results. Additionally, by guaranteeing the identification of fault parameters, the fault circuit model assists in training, significantly reducing the number of actual fault samples required for mechanism-enhanced neural network training. Finally, experiments on a representative converter are conducted to verify the effectiveness of the proposed method. Comparisons with state-of-the-art techniques show the proposed scheme's superiority in terms of diagnostic accuracy, decision explainability, sample dependence.