Answers are grounded in this paper and cite their sources.
A multimodal approach improves fault diagnosis accuracy in transformers using advanced algorithms.
Aiming at the limitations of traditional dissolved gas analysis methods in transformer fault diagnosis, a multimodal fault diagnosis method integrating quaternions and time‐series imaging is proposed. First, the Raman angle sum field is adopted to capture the global trend and phase relationship, and the Markov transition field is used to encode the local state transition and nonstationary dynamics, achieving the joint modeling of the global and local time structures. Then, a four‐branch feature extractor combining a multi‐scale pulse‐coupled neural network and multi‐scale spatial channel attention is constructed to learn multi‐scale saliency and spatial channel attention representations. Second, to simulate the geometric coupling between branches, quaternion representation and quaternion convolution are introduced, and multi‐head cross‐attention is adopted to enhance cross‐modal interaction and noise suppression. Finally, the hybrid strategy optimizes hyperparameters, the distributed adaptive Dragonfly algorithm performs global parameter search, while Bayesian optimization refines key configurations. The experimental results show that this method improves the accuracy and effectiveness of fault diagnosis.
No takes yet. Share an insight, caveat, or question.
Zhang et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: