Open‐Circuit Fault Diagnosis of Three‐Level T‐Type Converters Using a 1DCNN‐Transformer Framework With Adaptive Cross‐Domain Feature Interaction

ABSTRACT T‐type three‐level neutral‐point‐clamped (T‐NPC) inverters often operate under complex industrial conditions, where acquired electrical signals are heavily contaminated by background noise, nonlinear load disturbances, and electromagnetic interference. These challenges make it difficult to extract subtle fault signatures, resulting in low diagnostic accuracy for IGBT open‐circuit faults. To address this issue, this paper proposes a fault diagnosis model based on dual‐channel parallel feature extraction in both the time and frequency domains, coupled with adaptive feature fusion. In the time‐domain stream, an improved complete ensemble empirical mode decomposition with adaptive noise (ICEEMDAN) algorithm is employed for multiscale signal decomposition, followed by a one‐dimensional convolutional neural network (1DCNN) to extract deep local temporal features. In the frequency‐domain stream, a global frequency encoding (GFE) layer is introduced to replace the self‐attention mechanism of conventional transformers, reducing model parameters while effectively capturing global dependencies and harmonic distortion characteristics in the frequency domain. A cross‐domain fault‐adaptive recalibration mechanism dynamically adjusts the weights of the two branches, enabling complementary enhancement of multisource heterogeneous features. Experimental results demonstrate that the proposed model can reliably diagnose both single‐ and dual‐IGBT open‐circuit faults, achieving a diagnostic accuracy of 99.96% under both fault scenarios.