26 April 2022 in IEEE Transactions on Industrial Informatics
In recent years, the signal diagnosis on device operating under industrial environment has attracted increasingly attention. Most data-driven signal diagnosis methods are based on closed set assumption that class sets of training and test data are consistent. However, in industrial scenarios, during the running process of the device, the operating environment and condition may change over time, continuing generating data belong to unknown classes with new characteristics and distribution. The unknown classes usually reflect new modes or faults of the device need to be captured. They are unavailable in training phase, contradicting the closed set assumption. Existing methods are inappropriate to this type of open set classification requiring to classify known classes and recognize unknown classes. To address this challenging problem, this paper proposes a generic open set signal classification method. First, we apply Fourier transform to convert the sensor signals from time domain to frequency domain, then data in the time and frequency domains are fused. Next, a variational encoder-classifier network is proposed to classify known classes and learn the distribution of feature space to extract robust latent features. Finally, based on extreme value theory and entropy, a pair of discriminators determine whether samples belong to unknown or not. The experimental results on two vibration signal datasets from bearings and nuclear reactor demonstrate the effectiveness and superiority of our proposed open set signal classification method, especially in practical applications.
3 June 2025 in IEEE Sensors Journal
Variational Mode Extraction (VME) introduces a new criterion based on Variational Mode Decomposition (VMD): the residual signal after extracting a specific mode should have minimal or no energy at that mode’s center frequency. This allows VME to effectively extract modes near the center frequency, reducing the mode aliasing effect seen in VMD. Recursive variational mode extraction (RVME) successfully applies VME to bearing fault diagnosis. However, VME does not fundamentally improve VMD’s estimation of mode bandwidth, leading to biases in the extracted signals and limiting the full extraction of bearing fault characteristics. To address this issue, this paper proposes an Improved Variational Mode Extraction (IVME) method. IVME uses fractional-order derivatives to adjust the L2-norm estimation of mode bandwidth, allowing it to dynamically adapt to signals with varying bandwidths. An empirical formula is provided to relate the penalty factor to the fractional-order derivative. Additionally, IVME adaptively determines the initial center frequency of the target mode based on the convergence trend of VMD’s center frequency. This enables adaptive extraction of bearing fault characteristics. The proposed method is tested on both synthetic and experimental bearing fault signals, with its performance compared to other classical fault feature extraction methods. The results show that IVME outperforms VMD, VME, and fast SK in extracting fault features.