融合BWOSP-VMD-TOPSIS降噪和深度学习的旋转机械故障诊断

Fault diagnosis of rotating machinery using a fusion of BWOSP-VMD-TOPSIS denoising and deep learning techniques

提出了一种融合改进白鲸优化算法(Beluga Whale Optimization with Stranding Phase,BWOSP)、变分模态分解(Variational Mode Decomposition,VMD)和理想排序法(Technique for Order of Preference by Similarity to Ideal Solution,TOPSIS)构建综合指标,并结合深度学习的旋转机械故障诊断方法。首先,通过加入“搁浅”阶段建立了一种新型BWOSP算法;其次,利用BWOSP-VMD得到(K,α)最优参数组合;再次,考虑各本征模态分量的中心频率、相关性系数、峭度指标和包络熵通过TOPSIS构建综合指标并进行筛选、重构;最后,将BWOSP-VMD-TOPSIS降噪方法与多种深度学习模型相结合,以某轴承故障为例计算了故障诊断准确率和F1值,并与多种方法对比验证了方法的有效性和泛化性。结果表明,基于BWOSP-VMD-TOPSIS和深度学习的故障诊断方法能对含有强噪声干扰的旋转机械故障信号有效降噪并准确进行故障诊断,具有较强的泛化能力。

A novel approach is proposed to overcome the limitations of fault diagnosis methods for rotating machinery in noisy environments.This method integrates Beluga Whale Optimization with Stranding Phase(BWOSP),Variational Mode Decomposition(VMD),and Technique for Order of Preference by Similarity to Ideal Solution(TOPSIS)to develop a comprehensive evaluation index for Intrinsic Mode Function(IMF).Additionally,it employs a deep learning model for fault diagnosis in rotating machinery.The method proceeds as follows:First,adding a“stranded”phase to the original Beluga algorithm,resulting in a new BWOSP algorithm.The basic steps of the BWOSP algorithm are as follows:Firstly,determine whether the algorithm is in the exploration or exploitation phase based on whether B_f>0.5(exploration)or B_f≤0.5(exploitation).During the exploration phase,the algorithm explores new potential solutions by leveraging the stranded phase,enhancing the diversity of search.Conversely,in the exploitation phase,the algorithm focuses on refining promising solutions to achieve optimal performance.The second stage involves determining whether the algorithm is in the whale-fall stage or the stranding stage.This is determined by comparing the equilibrium factor(B_f),whale fall probability(W_f),and stranding probability(G_f).If G_f<B_f<W_f,the algorithm is in the whale-fall stage.Conversely,if B_f≤G_f,the algorithm enters the stranding stage.Secondly,the envelope entropy was utilized as the fitness function to optimize the parameters of the VMD method using BWOSP,resulting in the identification of the optimal parameter combination(K,α).Thirdly,the center frequency of each IMF component is taken into account,and a comprehensive index is constructed by TOPSIS based on the correlation coefficient,kurtosis index,and envelope entropy of each IMF component.The specific steps are as follows:Firstly,the IMF components are extracted.The first step involved removing■K/2■high-frequency components from K IMF components(where■denotes rounding down).In the second step,the correlation coefficient,kurtosis index,and envelope entropy of the remaining IMF components were calculated.Thirdly,a comprehensive index was constructed using the TOPSIS algorithm.Fourthly,the signal was reconstructed based on the selected IMF components.Finally,by integrating the BWOSP-VMD-TOPSIS noise reduction method with various deep learning neural network models,the fault diagnosis accuracy and F_(1)value were evaluated,focusing on a bearing fault as an example.The effectiveness and generalizability of the proposed method were validated through comparative analysis with multiple existing methods.The results demonstrate that the fault diagnosis method,which combines BWOSP-VMD-TOPSIS for noise reduction with deep learning techniques,effectively diagnoses fault signals in rotating machinery even in the presence of strong noise interference.This approach not only enhances the accuracy of the deep learning diagnosis model but also exhibits robust generalizability across various fault scenarios.