Research on a bridge crane rail fault diagnosis method integrating HO-VMD and RS-BP

Abstract

To address the challenges of small limited rail fault samples and difficulty in detecting bridge crane rail faults, we propose a fault diagnosis method that first employs Hippopotamus Optimization (HO)-optimized Variational Mode Decomposition (VMD) for signal processing and then integrates a diagnostic model combining Rough Set (RS) theory and a Back Propagation (BP) neural network. Initially, the collected vibration signals from rails with different faults are denoised using wavelet transform. Subsequently, the HO algorithm is employed to optimize the VMD parameters, determining the optimal combination of the number of modes and penalty factor. The Intrinsic Mode Function (IMF) with the smallest envelope entropy is then selected. Following this, time-domain and frequency-domain features are extracted. Based on these features, RS theory is applied for feature reduction, and a BP neural network is trained to classify and diagnose bridge crane rail faults. Experimental results demonstrate that the proposed HO-VMD method yields higher diagnostic efficiency and accuracy than the direct analysis of the original signal. Furthermore, the RS-BP model achieves a diagnostic accuracy of 98.3% even with a small dataset of fault samples. Therefore, the proposed method is highly feasible and effective for bridge crane rail fault diagnosis and is also promising for fault diagnosis in other related fields.