Data-driven mechanical fault diagnosis has been successfully developed in recent years, and the task of training and testing data from the same distribution has been well-solved. However, for some large machines with complex mechanical structures, such as reciprocating pumps, it is often not possible to obtain data from specific sensor locations. When the sensor position is changed, the distribution of the features of the signal data also changes and the fault diagnosis problem becomes more complicated. In this paper, a cross-sensor transfer diagnosis method is proposed, which utilizes the sharing of information collected by sensors between different locations of the machine to complete a more accurate and comprehensive fault diagnosis. To enhance the model’s perception ability towards the critical part of the fault signal, the local attention mechanism is embedded into the proposed method. Finally, the proposed method is validated by applying it to experimentally acquired vibration signal data of reciprocating pumps. Excellent performance is demonstrated in terms of fault diagnosis accuracy and sensor generalization capability. The transferability of practical industrial faults among different sensors is confirmed.

In recent years, with the advent of machine learning technology, data-driven fault diagnosis has emerged as a prominent research area. However, since the necessity of training and test data has to follow an identical distribution, the conventional data-driven diagnostic approach is not well-suited to engineering diagnosis. To address this shortcoming, the diagnosis method named FOCAL-TFAM-ResNet-LSTM is proposed to solve the problem of inconsistent distributions. This diagnosis method is mainly composed of two aspects. Firstly, the mapping-based deep transfer learning method with a novel adaptive loss function construction method is proposed. Specifically, the feature-oriented correlation alignment layer (FOCAL) metric function can significantly improve the classification accuracy of transfer learning by alignment of nonlinear distributional differences between the source and target domains in a high-dimensional reproducing kernel Hilbert space. Meanwhile, a novel adaptive metric function based on gradient dynamic weight optimization is proposed to balance the weight allocation relationship between domain and diagnosis, which can simultaneously minimize the domain difference and decrease the classification error. Secondary, the time-frequency attention module (TFAM) is innovatively applied in the ResNet-LSTM network to avoid the gradient disappearance among long-term sequence signals training. Finally, the algorithm is validated by using various experiments during the reciprocating pump dataset and the Case Western Reserve University public dataset. Under extremely adverse conditions (e.g., an SNR of - 5 dB), the FOCAL-TFAM-ResNet-LSTM method attains a classification accuracy of 96.83% and a loss of 0.1024 in reciprocating pump dataset. In comparison to conventional domain alignment techniques such as MMD and MK-MMD, it achieves a notable 8.42% gain in accuracy and a 0.3594 reduction in loss.

As one of the rotating components in the reluctance motor type control rod drive mechanism (CRDM), the life of the scroll wheel is closely related to the service life of the CRDM. In addition, the prediction of the remaining useful life of the scroll wheel helps to optimize the maintenance process of the CRDM. This paper proposes a hybrid method to predict its remaining useful life when the available degradation data are rare and the failure threshold cannot be accurately defined. First, the particle filtering algorithm, whose state transfer equation is established on the segmental damage physical model, is used to predict the degradation state of the scroll wheel. Second, the proportional hazard model for the relationship between the scroll wheel damage characteristics and reliability model is established to predict the remaining useful life of it. The proposed method focuses on the establishment of segmental damage physical model and the clustering analysis of damage characteristics extracted from vibration signals, which can be used to predict the remaining useful life of the scroll wheel. In addition, the results provide an opportunity for the condition-based preventive maintenance of the CRDM.

Fault diagnosis is of great significance for the maintenance of rotating machinery. Deep learning is an intelligent diagnostic technique that is receiving increasing attention. To address the issues of industrial data with small samples and varying working conditions, a residual convolutional neural network based on the attention mechanism is put forward for the fault diagnosis of rotating machinery. The method incorporates channel attention and spatial attention simultaneously, implementing channel-wise recalibration for frequency-dependent feature adjustment and performing spatial context aggregation across receptive fields. Subsequently, a residual module is introduced to address the vanishing gradient problem of the model in deep network structures. In addition, LSTM is used to realize spatiotemporal feature fusion. Finally, label smoothing regularization (LSR) is proposed to balance the distributional disparities among labeled samples. The effectiveness of the method is evaluated by its application to the vibration signal data from the safe injection pump and the Case Western Reserve University (CWRU). The results show that the method has superb diagnostic accuracy and strong robustness.

Aiming at the difficult problem of predicting the running state of the rotor of a Control Rod Drive Mechanism (CRDM) in a floating nuclear reactor, this paper proposes a Remaining Useful Life (RUL) prediction method based on Variational Mode Decomposition and Bidirectional Long Short-Term Memory (VMD-BiLSTM). Firstly, a bench experiment of the CRDM is carried out to collect the full operational cycle (full-stroke) vibration signals of the CRDM. Secondly, the collected data are decomposed based on the VMD, and the typical vibration signals at different stages of the experiment are used to verify this method and comprehensively mine the degradation characteristics. At the same time, the time-frequency domain feature analysis is carried out on the original vibration data, and the changing trends of the extracted features are carefully analyzed. Five feature quantities closely related to the degradation trend of the rotor of the CRDM are screened out, and the corresponding health indicators are constructed in combination with the stroke. Finally, the life prediction of the rotor of the CRDM is realized through the BiLSTM method. Then, the comparison experiments with other methods are carried out, and the experimental results show that the method proposed in this paper has high accuracy and reliability and can effectively solve the RUL prediction problem of CRDM, which provides a strong support to ensure the safe and stable operation of floating nuclear reactors.

Valves are critical actuation components for the reactor coolant pressure boundary, and their reliability directly governs the safety of offshore floating nuclear power plants. However, nuclear-grade valves face scarce fault samples and unlabeled operational data, making supervised learning infeasible. This paper presents an unsupervised fault detection and grading method based on OCSVM-IF-PCA, combining physical feature fusion and a pruning-weighted voting mechanism. Five physically interpretable features are extracted from differential pressure signals based on the valve switching mechanism. Three base detectors, namely OCSVM, Isolation Forest, and PCA reconstruction error, are selected via performance optimization. K-Means is adopted for unsupervised fault severity grading of detected anomalies. Tests are conducted on the ZeMA hydraulic dataset with 20 dB simulated radiation-induced noise. Results show that the proposed method achieves 100% average recall, 99.93% average accuracy, and 0.86% average false alarm rate. It enables reliable, interpretable fault diagnosis without fault labels under strong radiation noise, offering a technically feasible solution for reactor valve condition monitoring.