Noise-Resistant Structural Damage Identification via Multisensor Time–Frequency Feature Fusion and Interpretable Multiscale Pooling-Free Network

To address reliability degradation in structural damage identification (SDI) under complex operational noise interference (e.g., industrial machinery environments and bridge traffic vibrations), this study proposes a noise-resistant diagnostic framework integrating multisensor time–frequency feature engineering and a multiscale pooling-free network (MSPN). The framework specifically tackles two critical challenges: damage feature extraction and preservation in measurement systems under strong noise interference. The main contributions include the following. First, a time–frequency data construction method based on multisensor signal collaboration, generating physically interpretable noise-resistant feature descriptors through 3-D fusion of temporal waveforms, amplitude–frequency characteristics, and wavelet scattering coefficients. Compared to traditional single-domain feature methods, it maintains 80.63% SDI accuracy under 60% noise interference. Second, the proposed MSPN architecture for civil and mechanical infrastructure employs heterogeneous convolutional kernels and a parallel topology to eliminate feature degradation caused by pooling operations. Experimental results demonstrate that its noise robustness significantly outperforms baseline methods. Third, an interpretability analysis framework combining t-distributed stochastic neighbor embedding (t-SNE) and SHapley Additive exPlanations (SHAP) to evaluate the model’s focus on multilevel characteristic responses in damaged data, enhancing decision transparency. Validated through a laboratory-scale structure and a public dataset, the proposed method achieves 89.16% and 77.46% accuracy under 60% noise contamination. This work offers a promising low false-alarm-rate solution for real-time monitoring of civil infrastructure and industrial machinery in noisy operational environments, thereby facilitating the implementation of preventive maintenance strategies.