SWT-KELM-based rolling bearing fault diagnosis method under noise conditions with different SNRs

Current deep-learning methods are often based on significantly large quantities of labeled fault data for supervised training. In practice, it is difficult to obtain samples of rolling bearing failures. In this paper, a transfer learning-based feature fusion convolutional neural network approach for bearing fault diagnosis is proposed. Specifically, the raw vibration signal features and the corresponding time-frequency image features of the input data are extracted by a one-dimensional convolutional neural network and a pre-trained ConvNeXt, respectively, and connected by a feature fusion strategy. Then, the fine-tuning method based on transfer learning can effectively reduce the reliance on labeled samples in the target domain. A wide convolution kernel is introduced in the time-domain signal feature extraction to increase the receptive field, which is combined with the channel attention mechanism to further optimize the feature quality. Finally, two common bearing datasets are utilized for fault diagnosis experiments. The experimental results show that the proposed model achieves an average accuracy of more than 98.63% in both cross-working conditions and cross-device diagnosis tasks. Meanwhile, anti-noise experiments and ablation experiments further validate the accuracy and robustness of the proposed method.

Section: Data Preprocessingmentioning

confidence: 99%

Deep transfer learning rolling bearing fault diagnosis method based on convolutional neural network feature fusion

Yu,

Fu,

Song

et al. 2023

“…g is the activation function. In this paper's case, in order to improve prediction accuracy, it is not only need to know whether the input signal has an important impact on the result after weighting and offsetting, but also need to identify whether the impact is positive or negative and how much they are, so comparing to previous works, a new form of activation function (Hyperbolic tangent function) was chosen in this paper [44][45][46][47][48][49]:…”

Section: Elmmentioning

confidence: 99%

A novel data-driven method for fuel-consumption prediction based on fast converged kernel extreme learning machine

Wang

2023

An accurate fuel consumption prediction is significant to fuel-economy-oriented optimization, which can improve vehicle fuel economy. Based on the data-driven prediction framework, this paper proposes a Real-time Fuel Consumption Prediction Model (RFCPM) using Kernel Extreme Learning Machine (KELM) which is optimized by Fast Converged Grey Wolf Algorithm (FCGWA). A new steering function of FCGWA and a new activation function for KELM are presented to ensure the fast converge speed and higher accuracy performance. First, the characteristic variables of RFCPM are selected by reference to the model-based fuel consumption prediction method. Second, a KELM is adopted to predict fuel consumption. Third, FCGWA is adopted to select the best parameters of KELM using k-fold cross-validation method. Finally, the best model is chosen through real-vehicle test. Test results are compared with the original ELM and the wildly used WNN. The experiment shows that the proposed method significantly outperforms the original ELM and WNN in terms of both prediction accuracy and training time.

“…In the process of seismic wave signal analysis, researchers have proposed numerous filtering methods, such as Kalman [29,30] filtering, wavelet transforms filtering [31,32], SVD [33,34], EMD [35][36][37][38][39] and so on. DMD is an emerging data dimensionality reduction method that can also be utilized for analyzing seismic wave signals.…”

Section: Introductionmentioning

confidence: 99%

A new rapid positioning strategy for the impact-point using dynamic mode decomposition and time difference of arrival ^*

Bi¹,

Huang²,

Wang³

et al. 2023

In outdoor shooting range experiments, locating the projectile impact point is typically done through manual search, which can be time-consuming and costly. To improve experimental efficiency and reduce costs, this paper proposes a new approach for localizing the impact point using the dynamic mode decomposition (DMD) extension method and time difference of arrival (TDOA) based on seismic waves. The DMD extension method is applied to one-dimensional seismic signals. Additionally, diagonal averaging is applied to recover the reconstructed matrix, which could achieve better noise reduction than conventional singular value decomposition and empirical mode decomposition methods. Numerical simulations are then conducted to analyze the impact point positioning model, resulting in an absolute location error within 20 m in a 1000 m square area. The work demonstrates that the joint positioning algorithm combining on DMD and TDOA achieves high positioning accuracy, making it a promising tool for future experiments on outdoor shooting ranges.