Advanced Methods in Rotating Machines

The motions of power sources in industrial applications were always provided by electromechanical systems, which use around 70% of the gross energy consumption of industrialized economies. The rotating machines are key components used in those electromechanical systems. The working conditions of the rotating machines should greatly affect the power performances of the electromechanical systems. The condition monitoring and fault diagnosis methods are essential technologies for the state maintenance of the rotating machines. In this context, advanced methods, such as condition monitoring, fault diagnosis, artificial intelligence and dynamic modelling, are used for the working state monitoring, vibration data cleaning, mechanical vibration data diagnosis, composite fault signal decomposition and reconstruction, vibration and noise simulation in the mechanical and electrical rotating machines. Advanced methods make the rotating machines become “smarter” and gain “intelligence”. In this Special Issue entitled “Advanced methods in rotating machines”, we give a comprehensive view of different research contribution from condition monitoring and detection methods for mechanical and electrical rotor systems.

This Special Issue on “Advanced methods in rotating machines” includes eleven papers ranging over different methods of the mechanical and electrical rotor systems. Those papers include a comprehensive literature review for the current and novel works for the condition monitoring and fault diagnosis methods of the mechanical [1] and electrical [2] rotating machines, condition monitoring of the bearing geometrical changes [3], integrate vibration data cleaning method of rolling element bearings [4], deep learning method for the bearing and irreversible-demagnetization faults [5], mechanical vibration diagnosis method based on the shallow neural networks [6], composite fault signal decomposition and reconstruction method for gear-bearing system [7], automated fault classification method for rotating machines [8], simulation method for electrical motor vibration and noises [9], dynamic modelling methods of rolling element bearings [10] and rotor transmission equipment in the locomotive and track structure system [11].

A brief summary of the content associated with each of the chosen papers belonging to this Special Issue is listed as follows:

Advanced methods for obtaining the accuracy operational status of the rotating machines should be helpful for their maintenance and optimization design. Many works were introduced to develop different methods for the accurate diagnosis and prognosis. In the study conducted by Mushtaq et al. [1], the authors introduced a comprehensive literature review of the data-driven fault diagnosis framework including the machine learning methods (such as support vector machines, artificial neural networks, and k-nearest neighbors, etc.) and deep leaning algorithms (such as deep convolutional network, deep belief network, auto-encoder-based deep neural network, deep recurrent neural network) used for the bearing fault diagnosis for rotating machines. The authors also listed the available public datasets (such as Paderborn University Bearing, PRONOSTIA, Case Western Reserve University, and Intelligent Maintenance Systems). Moreover, Frosini [2] updated the literature review for current diagnosis methods for electrical rotating machines with different sizes and types, such as load and auxiliaries, stator and rotor windings, bearings, magnets, airgap, stator and rotor laminated core, etc.

When the faults were caused in the rotating machines, some small surface failures or operational performance changes should occur. To diagnose the bearing geometrical changes, Wang et al. [3] conducted an online bearing clearance monitoring method to detect the bearing clearance changes and runout errors. They used the Modulation Signal Bispectrum to reduce the noise effect. The Gini index was applied to the relative peakness in the spectrum. Moreover, an interpretable indicator from 0 and 1 was introduced to describe the bearing status. The performance of the developed method was partly verified by the datasets from a bearing test and run-to-failure gearbox test. In their results, the bearing clearance change of 20 µm may be indicated by the developed method.

The intelligent diagnosis accuracy of rotating machines is determined by the vibration data quality and the state identification model accuracy. To improve the data quality and identification model’s chosen accuracy, Qin et al. [4] proposed an integrated data-cleaning method to choose the effective learning samples and optimize the structural parameters selection in the deep belief network model (DBNM). In their model, to calculate the energy fluctuation parameter of finite number in the intrinsic function components, the variational modal decomposition of vibration data of rolling element bearing was used. To optimize the nodes’ number in the hidden layer in the DBNM and calculate the optimal structural parameters of diagnosis model, the particle swarm method was used. Moreover, the learning samples with the highest quality were defined as the input of intelligent identification method for bearing faults. In their results, the proposed method can improve the identification accuracy of bearing state by 3%.

Although many works were reported to train the deeper networks on the relatively fewer and non-uniform test data of electric rotating machines, it was still a study for the applications in electrical vehicles. Ullah et al. [5] proposed a deep learning method for the bearing and irreversible-demagnetization faults, which is based on a transfer-learning-based pre-trained visual geometry group (VGG). The stator current and vibration signals were selected for the feature extraction. To exploiting the transfer learning benefits, a confluence of vibration and current signals-based signal-to-image conversion method was developed. In their results, a state-of-the-art accuracy of 96.5% for the faults classification can be obtained.

To detect the faults in the rolling element bearings in the permanent magnet synchronous motors used in electric and hybrid vehicle drives, Ewert et al. [6] proposed a mechanical vibration diagnosis method based on the shallow neural networks to detect the mechanical damages in the support bearings. The Fast Fourier Transform and Hilbert transform were applied to extract the diagnostic symptoms in the vibration acceleration signals. Three neural network types including the multilayer perceptron (MLP), neural network with radial base function (RBF), and Kohonen map (self-organizing map, SOM) were applied to automate the diagnosis processing. The effectiveness of developed method was verified by the comparisons between different neural networks methods.

At present, many diagnosis methods only focused on vibration signals of individual drive components. However, compound faults of gears and bearings should be occurred extremely frequently in the rotating machines. In Han et al. [7], a gear-bearing composite fault signal decomposition and reconstruction method was proposed based on the marine predator algorithm (MPA) and variational mode decomposition (VMD) approaches. Their method can separate the vibration characteristics of gear and bearing compound faults. The test results also verified the performance of the separation and denoising mixed gear-bearing fault signals.

Vibration analysis technologies contributed significantly towards understanding different faults in the rotating machines. However, the vibration fault diagnosis needed the huge vibration data at various measurement points. Thus, many new methods should be needed to reduce the faults detection downtime. To rationalize the vibration diagnosis data of faults in the rotating machines, Yunusa-Kaltungo and Cao [8] presented an automated fault classification method for rotating machines to retain information that offer maximum variability including the principal components analysis and coherent composite spectrum. Their results provided that the artificial neural networks can offer the accurate and consistent classification outcomes for their studied cases. Moreover, the results demonstrate some opportunity for the automated diagnosis process of rotating machines.

Vibrations and acoustic noises can be used to diagnose the working status of the rotating machine. For the electric motors, the acoustic noise is the complex nature and origin phenomenon, which can be caused by the electromagnetic and mechanical conditions. Sathyan et al. [9] discussed the electromagnetic, mechanical and aerodynamic vibration and noises. They used the finite element and boundary element methods to simulate the vibration and noise of the electrical motor. The vibration noise characteristics can also be used for the condition monitoring and fault diagnosis of the electrical motors.

In addition: vibration-model-based methods are also helpful for the condition monitoring and fault diagnosis of the rotating machines. Accurate vibration modeling methods are beneficial for understanding the vibrations of rotating machines. For example, Luo et al. [10] proposed a dynamic model of a rolling element bearing with the cage slip, roller motion, friction and damping. The effects of the cage slip, rotational speed, friction coefficient, and radial load on the vibration energy of the rolling element bearing were discussed. Moreover, to simulate the working performances and vibrations from a more complex rotating machine, Guo et al. [11] proposed a vibration model of the rotor transmission equipment in the locomotive and track structure system. The effects of the flat length and relatively constant velocity on the vibrations of the rotor transmission equipment of the locomotive system were studied.