Condition monitoring and fault diagnosis are of significance to improve the safety and reliability of motors, given their widespread applications in virtually every branch of the industry. Sequential data modeling based on recurrent neural network (RNN) and its variants have drawn increasing attention because the temporal nature of motor signals can be well leveraged for motor analysis. One common drawback of prior research is that signals measured on motors are typically analyzed with a fixed time window, making it difficult to trade off between global state estimation and local feature extraction. This paper presents a deep learning-based model termed Multi-Resolution & multi-Sensor Fusion Network (MRSFN) for motor fault diagnosis, through multi-scale analysis of motor vibration and stator current signals. Specifically, vibration and current signals are first segmented by analysis windows of varying lengths to create a new data stream for the joint representation and temporal encoding of the original sensor signals, based on two network structures: convolutional neural network (CNN) and long short-term memory (LSTM). The advantage of the developed method is that it automatically learns the discriminative features through the network training process, without requiring manual feature selection as is typically the case in prior methods. By considering the temporal dependence of the signals being analyzed, the developed multi-resolution fusion technique not only improves the effectiveness of feature extraction but is also adaptive to varying motor speed. Two case studies demonstrate the advantages of the developed method.

The Differential Evolution (DE) population-based algorithm is an optimal algorithm with powerful global searching capability, but it is usually in low convergence speed and presents bad searching capability in the later evolution stage. A new Chaos Differential Evolution algorithm (CDE) based on the cat map is proposed which combines DE and chaotic searching algorithm. Firstly, the chaotic distributed superiority of the cat map is analyzed in this paper. Secondly, the detailed implementation of CDE is introduced. Finally, the effectiveness of CDE is verified in the numerical tests. The Support Vector Regression machine (SVR) is an effective tool to solve the problem of nonlinear prediction, but its prediction accuracy and generalization performances depend on the selection of parameters greatly. So, the CDE is applied to SVR to build an optimized prediction model called CDE-SVR. Then the new prediction model is applied to the short-time regression prediction of the chaotic time series and the boundary extension of the mechanical vibration signals. The results of the two experiments demonstrate the effectiveness of the CDE-SVR

The ball valve is an essential piece of equipment in an oil and gas pipeline. The sand particles transported through the pipeline can cause erosion and wear to the ball valve, thus causing it to fail, leading to serious safety hazards. In this paper, the self-designed erosion experiment method was combined with computational fluid dynamics (CFD), while the Euler-Lagrange method was also introduced to optimize the Oka erosion model and Ford particle-wall rebound model. The erosion mechanism and characteristics of the ball valve sealing surface in gas-solid two-phase flow were simulated, while the erosion condition of the specimen was analyzed and compared when exposed to different factors, such as different particle velocities, impact angle, particle size, and specimen materials. The experimental data conformed well to the CFD erosion simulation data, verifying the accuracy of the CFD simulation analysis. The results indicated that the worn surface was caused by various wear mechanisms, while a “stagnation zone” was identified at the center of the specimen. The maximum erosion area, which was U-shaped, was also located at the center. The erosion rate increased in conjunction with an increase in the particle velocity and size, both of which failed to affect the erosion pattern. The erosion rate initially increased, after which it decreased with the impact angle, reaching the maximum value at an impact angle of 30°. This paper summarizes the erosion failure mechanism and characteristics in gas–solid two-phase flow and provides both technical support and a theoretical basis for the on-site maintenance of essential vulnerable parts in the pipeline, such as ball valves.

Blade tip timing (BTT) technology is considered the most promising method for blade vibration measurements due to the advantages of its simplicity and non-contact measurement capacity. Nevertheless, BTT technology still suffers from two problems, which are (1) the requirements of domain expertise and prior knowledge of BTT signals analysis due to severe under-sampling; and (2) that the traditional BTT method can only judge whether there is a defect in the blade but it cannot judge the severity and the location of the defect. Thus, how to overcome the above drawbacks has become a big challenge. Aiming at under-sampled BTT signals, a feature learning method using a convolutional neural network (CNN) is introduced. In this way, some new fault-sensitive features can be adaptively learned from raw under-sampled data and it is therefore no longer necessary to rely on prior knowledge. At the same time, research has found that tip clearance (TC) is also very sensitive to the blade state, especially regarding defect severity and location. A novel analysis method fusing TC and BTT signals is proposed in this paper. The goal of this approach is to integrate tip clearance information with tip timing information for blade fault detection. The method consists of four key steps: First, we extract the TC and BTT signals from raw pulse data; second, TC statistical features and BTT deep learning features will be extracted and fused using the kernel principal component analysis (KPCA) method; then, model training and selection are carried out; and finally, 16 sets of experiments are carried out to validate the feasibility of the proposed method and the classification accuracy achieves 95%, which is far higher than the traditional diagnostic method.

Recent years have witnessed a booming interest in the data-driven paradigm for fault diagnosis. However, it is usually difficult to collect sufficient faulty data for model training in practical applications, thus limiting the application of these intelligent diagnosis methods. In this article, we develop a novel method named meta-learning with intraclass and interclass optimization (MLIIO), which targets training an effective metric-based fault classifier using limited data. On the one hand, an intraclass aggregation loss function is proposed to enable sample features from the same class to gather together. This yields a compact representation manifesting the central tendency for the same categories. On the other hand, an interclass discriminative loss function is proposed to enforce sample features from the different classes to maintain a large margin, which further ensures that the metric space has a clearer discriminative boundary. By applying the episodic training mechanism to optimize the proposed losses, general, and discriminative feature representations can be learned to more efficiently identify new failure scenarios with scarce data. Experimental results on a public rolling bearing dataset and a real-world railway turnout dataset showcase that the proposed MLIIO approach outperforms several state-of-the-art methodologies.