In the field of remaining useful life (RUL) prediction, accurately evaluating incipient faults in bearings by using conventional health indicators poses challenges, while traditional neural network models fail to provide reliable uncertainty distributions for credible output. Therefore, a cutting-edge deep learning method based on fault feature gain (FFG) is proposed, which aims to accurately predict the RUL of rolling bearings while quantifying the associated uncertainty distribution. Firstly, combined with the adaptive spectrum mode extraction (ASME) theory, FFG is proposed to quantitatively assess the degree of bearing damage. Secondly, a mechanism for identifying incipient faults is established to determine the optimal time for making the first prediction. Subsequently, a deep learning model combining gated recurrent unit (GRU) and Bayesian neural network (BNN) is constructed to predict the RUL of bearings and quantify the uncertainty distribution. Finally, experimental results obtained from an accelerated degradation test bench for rolling bearings validate the effectiveness and advantages of the proposed method. The results demonstrate that FFG enables accurate assessment of bearing health status while providing crucial insights into the underlying failure modes. Furthermore, the GRU-BNN model performs more accurately in RUL prediction, and can better quantify the uncertainty of RUL.

The acoustic pressure signal generated by blades is one of the key indicators for condition monitoring and fault diagnosis in the field of turbines. Generally, the working conditions of the turbine are harsh, resulting in a large amount of interference and noise in the measured acoustic pressure signal. Therefore, denoising the acoustic pressure signal is the basis of the subsequent research. In this paper, a denoising method of micro-turbine acoustic pressure signal based on the Complete Ensemble Empirical Mode Decomposition with Adaptive Noise (CEEMDAN) and Variable step-size Normalized Least Mean Square (VSS-NLMS) algorithms is proposed. Firstly, the CEEMDAN algorithm is used to decompose the original signal into multiple intrinsic mode functions (IMFs), based on the cross-correlation coefficient and continuous mean square error (CMSE) criterion; the obtained IMFs are divided into clear IMFs, noise-dominated IMFs, and noise IMFs. Finally, the improved VSS-NLMS algorithm is adopted to denoise the noise-dominated IMFs and combined with the clear IMF for reconstruction to obtain the final denoised signal. Adopting the above principles, the acoustic pressure signals generated by a micro-turbine with different rotation speeds and different states (normal turbine and fractured turbine) are denoised, respectively, and the results are compared with the axial flow fan test (ideal interference-free signal). The results show that the denoising method proposed in this paper has a good denoising effect, and the denoised signal is smooth and the important features are well preserved, which is conducive to the extraction of acoustic pressure signal characteristics.

Addressing the critical challenges of limited labeled data and operational variability in bearing fault diagnosis, a Multiscale Recalibration Attention Network (MRAN) integrated with adaptive grouped convolution is proposed in this paper. Firstly, a statistical-aware recalibration module that dynamically fuses mean and standard deviation information to suppress noise and amplify discriminative features under data scarcity. Secondly, a kernel-adaptive grouped convolution mechanism that intelligently assigns computational resources based on kernel size to optimize efficiency; Thirdly, a hierarchical fine-tuning strategy that selectively adapts deep layers to resolve cross-domain feature misalignment. Extensive experiments on BJTU-RAO and cross-device datasets demonstrate the superior performance of MRAN: it achieves an accuracy of 98.67% without transfer learning, which is significantly better than the state-of-the-art CNNs and Vision Transformers. Crucially, using only 30% of the target domain data, MRAN has an accuracy of 94.57% in cross-conditional tasks and an accuracy of 92% in demanding cross-device diagnostics, while greatly reducing training time and computational burden. This work establishes MRAN as a groundbreaking, efficient, and highly accurate solution for practical industrial bearing fault diagnosis under data and operational constraints.

To address the challenges in understanding the raceway failure mechanisms of bearings under dynamic radial excitations, this study proposes a vibration evolution analysis method based on multi-source data fusion and a Granger causality test. Firstly, a vertical bearing vibration test bench that can simulate the dynamic excitation in engineering practice is built, and the bearing acceleration, inner ring displacement and cage data are collected at the same time. Subsequently, the evolution law and correlation relationship of bearing vibration signals during the expansion process of bearing raceway damage were studied. Based on this, a multi-source vibration data fusion method was proposed, and the effectiveness of different data fusion schemes in characterizing raceway damage expansion was compared. Finally, the Granger causality test was applied to analyze the causal relationship between the evolution of various vibration behaviors during the damage propagation process. Research results demonstrate that under complex loading conditions during sustained operation, the “False Brinelling” indentation gradually develops into raceway surface damage. The vibration behavior of bearings exhibits distinct stage-specific characteristics under dynamic radial excitations. Notably, variations in vibration behavior amplitude and transition timing between different operational phases demonstrate significant discrepancies. Significant alterations in causal relationships between vibration behaviors were observed throughout different degradation phases. The combined approach proposed in this paper, encompassing complex load simulation, multi-source data fusion, and causal analysis, offers a new understanding of the raceway failure mechanism of bearings under real-world operating conditions.