Shi Huifang, Miao Yonghao, Wang Xun, Xie Jiaxin
School of Reliability and Systems Engineering, Beihang University, Beijing 100191, China.
School of Mathematical Sciences, Beihang University, Beijing 100191, China.
ISA Trans. 2025 May;160:218-236. doi: 10.1016/j.isatra.2025.02.029. Epub 2025 Feb 28.
Fault diagnosis in complex industrial systems often encounters significant challenges, including high noise levels, stochastic interference and coupled multi-fault features, especially for multi-channel signal processing. To address these issues, this study proposes multi-dimensional synchronous feature mode decomposition (MSFMD) method, a novel approach that combines multi-channel signal synergy with advanced decomposition and feature extraction techniques. The MSFMD method operates through a systematic framework comprising three key steps: custom-designed spectral segmentation strategy based on order statistic filter, synchronized decomposition of multi-channel signals with spectral alignment constraint, adaptive mode screening based on time-frequency correlation coefficients and envelope spectral kurtosis. Tailored for the channel signal, initial filter banks are decided. Then, the same fault-feature-oriented modes keep the spectral alignment constraint across channels, capturing inter-channel correlations while reducing noise and redundant modes. The adaptive screening strategy selectively retains fault-relevant modes, significantly improving the robustness and interpretability of the extracted features. MSFMD is able to effectively amplify weak fault features, handle complex multi-fault conditions, and improve computational efficiency under high-noise environments. Compared to traditional methods such as feature mode decomposition (FMD) and multivariable variational mode decomposition (MVMD), MSFMD demonstrates superior performance, such as susceptibility to noise, redundancy, and inefficiency in multi-fault scenarios. Validation through complex fault experiments confirms MSFMD's capability to provide accurate and reliable diagnostics.
复杂工业系统中的故障诊断常常面临重大挑战，包括高噪声水平、随机干扰和耦合多故障特征，尤其是在多通道信号处理方面。为解决这些问题，本研究提出了多维度同步特征模式分解（MSFMD）方法，这是一种将多通道信号协同与先进的分解和特征提取技术相结合的新颖方法。MSFMD方法通过一个包含三个关键步骤的系统框架来运行：基于顺序统计滤波器的定制频谱分割策略、具有频谱对齐约束的多通道信号同步分解、基于时频相关系数和包络谱峭度的自适应模式筛选。针对通道信号进行定制，确定初始滤波器组。然后，相同的面向故障特征的模式在各通道间保持频谱对齐约束，在降低噪声和冗余模式的同时捕捉通道间的相关性。自适应筛选策略有选择地保留与故障相关的模式，显著提高了所提取特征的鲁棒性和可解释性。MSFMD能够有效放大微弱故障特征，处理复杂的多故障情况，并在高噪声环境下提高计算效率。与传统方法如特征模式分解（FMD）和多变量变分模式分解（MVMD）相比，MSFMD在多故障场景中表现出优越的性能，如对噪声的敏感性、冗余性和低效率等方面。通过复杂故障实验进行验证，证实了MSFMD提供准确可靠诊断的能力。
