The existing fault diagnosis of harmonic drives is difficult to deploy vibration sensors, and the diagnosis accuracy is insufficient only using current signal. Therefore, we propose an intelligent fault diagnosis method using a multimodal collaborative meta network (MCMN) with severely missing modality. First, multimodal data of the harmonic drive were collected to analyze. Second, a feature reconstruction network is used to achieve a unified model to handle missing modalities in testing. Besides, the uncertainty assessment is used as a feature regularization network to overcome data bias. Third, the inference of the meta-learning is used to obtain the generated weights, and carried out the joint optimization. Finally, MCMN is optimized by using the approximate lower bound of Monte Carlo. Experimental results show that the accuracy of MCMN is 93.75% in the case of full-modalities training data and the severely missing modalities testing data, which is better than the existing method.

In actual industrial production, the load and speed of bearings are complex and changeable. However, most existing research on bearing fault diagnosis is based on constant speed conditions, and studies on bearing fault diagnosis at time-varying speeds are limited. Additionally, the scarcity of fault data further hinders practical applications of theoretical models developed so far. Thus, CapsFormer, a novel bearing intelligent fault diagnosis framework with negligible speed change under small-sample conditions, is proposed in this study. This framework combines the power of capsule network (CapsNet) and Transformer. It converts 1D time-domain samples into 2D time-frequency representations (TFRs) through short-time Fourier transform (STFT). Then it employs the idea of CapsNet to extract ordered spatial features from the TFRs of samples. On this basis, combined with the self-attention learning mechanism, it excavates deep fault features to promote the correct identification of bearing fault types by the model. Through experiments conducted under constant speed and time-varying speed conditions, the model was validated, demonstrating its superior performance compared to six other deep learning methods in bearing fault diagnosis under small sample sizes. These results strongly indicate the robustness of CapsFormer in addressing speed changes during bearing fault diagnosis.

In this paper, we presented a method for online oil monitoring and wear condition warning within the wind turbine gearboxes. Initially, an innovative wear particles sensor was engineered, utilizing a combination of electromagnetic and permanent magnet hybrid excitation, enabling the detection of ferromagnetic particles with diameters greater than 40μm. The precision of this sensor was thoroughly confirmed through a series of meticulous experimental analysis. Subsequently, the sensor was strategically positioned within a bypass circuit of the gearbox, in conjunction with electromagnetic induction and machine vision sensors, to establish an encompassing and coordinated monitoring framework. Lastly, a Multidimensional Transformer Network (Md-Transformer) was developed to integrate and process real-time sensor data for diagnostic analysis. From the data analysis of the two units selected from the engineering site for the past three years, the accuracy of Root Mean Square Error (RMSE) analysis was found to be 0.0002 and 0.0013; the accuracy of Mean Absolute Percentage Error (MAPE) analysis were 0.1831 and 0.1527; the R ² (Coefficient of Determination) stability analysis were 0.8912 and 0.9002; the accuracy of multi-sensor fusion fault analysis was increased to 83.33%, the Md-Transformer consistently outperformed established models such as NAR (Non-linear Autoregressive Model), GRU (Gated Recurrent Unit) and TCN (Temporal Convolutional Networks), underscoring its robustness and efficacy in the surveillance of wind turbine gearbox health.

This paper presents a wear condition warning method based on oil on-line monitoring for wind turbine gearboxes. The method comprises a high-performance machine vision sensor capable of detecting and tracking both magnetic and non-magnetic colored wear metal particles larger than 4μm in lubricating oil. Following the detection, image feature extraction is conducted utilizing the U-Net network in conjunction with the watershed algorithm, designing to acquire and categorize particulate matter into distinct size fractions: 4-6μm, 6-14μm, 14-21μm, 21-38μm, 38-70μm and sizes exceeding 70μm. Furthermore, the study proposed a novel approach termed learnable multi-scale convolutional neural network (LMCNN), which integrates online real-time monitoring with offline laboratory analysis. This method effectively extracts oil characteristics under diverse operational conditions, demonstrating robust performance. The fault detection accuracy achieved with the testing dataset stands at 97.78%, surpassing conventional methods such as K-Nearest Neighbors (KNN), Decision Trees (DT), Random Forests (RF) and Multilayer Perceptrons (MLP). Moreover, when combined with historical data from the preceding three years, the system’s high rate of successful warning accuracy substantiates its robustness and reliability in the context of predictive maintenance for wind turbine gearboxes.

The paper presents a wear condition warning method for wind turbine gearboxes. Initially, a diagnostic analysis was conducted to detect ferromagnetic wear particles larger than 4μm using machine vision monitoring sensor, and to identify metallic wear particles larger than 70μm using electromagnetic induction sensor. The development of this synchronous monitoring method represents a pioneering application within the wind turbine gearboxes, with significant engineering implications. Subsequently, a machine vision-based approach was proposed to quantify wear conditions by establishing a wear index, incorporating parameters such as the pixel area of wear particles and the Structural Similarity Index (SSIM). To validate the effectiveness of this method, experimental investigations were conducted on a sample of five wind turbines. Ultimately, an integrated early warning system for monitoring the wear conditions of wind turbine gearboxes was proposed, incorporating the Multiscale Temporal Convolutional Network (MTCN). The MTCN model was evaluated using wear index data and electromagnetic induction sensor data from comparative experimental analyses. The results indicated that the MTCN model achieved a Root Mean Square Error (RMSE) of 1.1713, a Mean Absolute Percentage Error (MAPE) of 0.3416, and a coefficient of determination (R²) of 0.8912, demonstrating superior performance compared to both the BiLSTM and BiGRU methodologies.

The large petrochemical pelletizer gearbox is a critical component in synthetic chemicals, accurate forecasting of its time series from online oil monitoring is of utmost importance for safe operation. However, existing time series forecasting methods face challenges in handling oil monitoring scenarios characterized by disturbances, multimodality, and large time spans. To address these issues, we propose an intelligent forecasting method named Multiscale Inverted Transform Network (MITN). Firstly, the multimodal online oil monitoring time series collected from moisture sensors, viscosity sensors, abrasive image sensors, and metal particle sensors are utilized to perform correlation analysis and identify the key nonlinear variables. In addition, the multiscale module is further employed to obtain comprehensive redundant characteristics. Secondly, the inverted perspective is designed for modeling from the time dimension and the variable dimension. Thirdly, the multiple multi-attention mechanism module is utilized with the feedforward network to learn the semantic meaning of variable time series and produce accurate forecasting results. Finally, layer normalization is used to improve the training stability and convergence to eliminate the distribution difference between variables. We used online oil monitoring time series from a petrochemical pelletizer gearbox from July 2018 to May 2023 for validation. The results show that MITN not only can obtain smaller forecasting errors than the existing time series forecasting networks such as LSTM, GRU, TCN, and Transform, but also can effectively generalize to the unknown variables. The proposed MITN pioneers ideal multivariate time series forecasting for complex online oil monitoring, with the potential to enhance the operational safety of large petrochemical plants.