ADS-Based Hybrid IRBwSA–Vision Transformer Framework for Intelligent Fault Diagnosis of Industrial Centrifugal Pumps
Centrifugal pumps are essential components in industrial fluid transport and energy conversion systems, where reliable operation is critical for safety and efficiency. However, vibration signals generated by these systems are often noisy and highly nonstationary due to complex fluid-structure interactions, making accurate fault diagnosis challenging. Conventional signal processing and deep learning approaches frequently fail to capture transient fault characteristics, as they typically focus either on local feature extraction or global contextual modeling. Moreover, most deep learning architectures are computationally demanding and rely on large labeled datasets, whereas real-world industrial fault data are inherently limited and imbalanced. To overcome these limitations, this study proposes a hybrid fault diagnosis framework based on augmented diagnostic scalograms. Raw vibration signals are transformed into continuous wavelet transform-based time-frequency representations, followed by data augmentation to enhance robustness under limited data conditions. The proposed architecture integrates an inverted residual bottleneck network with feature-level self-attention for efficient local feature extraction and a vision transformer for capturing global dependencies. The complementary features are fused through an optimization-driven feature selection strategy to reduce redundancy and improve discriminative capability, and the final classification is performed using a shallow neural network. The effectiveness of the proposed method is validated using real industrial vibration data collected from a multistage centrifugal pump operating under multiple pressure conditions, achieving classification accuracies of 98.60%, 98.33%, and 99.44%, respectively. These results demonstrate that the proposed framework provides a robust, efficient, and reliable solution for intelligent fault diagnosis under varying industrial conditions.