Vibration-based multi-state rolling bearing fault diagnosis for a vertical cattle-manure briquetting machine using multi-domain feature fusion and SVM-RBF

To address the multi-class fault identification of rolling bearings in vertical cattle-manure briquetting machines operating at low speed under actual briquetting loads and strong background noise, a fault diagnosis method based on complete ensemble empirical mode decomposition with adaptive noise–wavelet threshold denoising–dung beetle optimizer (CEEMDAN-WTD-DBO), multi-domain feature fusion, and a support vector machine with a radial basis function kernel (SVM-RBF) is proposed. First, the vibration signals are decomposed, followed by sensitive intrinsic mode function (IMF) selection and denoising. Subsequently, 45-dimensional CSvDE features are extracted and fused with 10 time-domain, 10 frequency-domain, and 10 envelope-spectrum features to construct a 75-dimensional Fusion75 feature set. The top 40 features are then selected using the F-score criterion and fed into the SVM-RBF classifier to identify five bearing states. Experimental validation was conducted using measured data from 30220-type rolling bearings. During model-selection cross-validation on the g01 dataset containing 350 samples, the selected configuration yielded an Accuracy of 94.86% and a Macro-F1 score of 94.96%. On the completely held-out g02/g03 external batches comprising 700 samples, the Accuracy and Macro-F1 reached 95.86% and 95.80%, respectively. The independent external validation results demonstrate that the proposed method can effectively characterize bearing fault features and exhibits good cross-batch generalization capability.