Torsional vibrations are inherently present in every rotating powertrain. In resonant conditions, torsional vibrations can be significantly amplified. A typical method to reduce the torsional vibration particularly at resonance is to modify the torsional natural frequencies with the component design. Commonly, a straightforward method for the modification is the adjustment of the torsional stiffness of a coupling. This study presents a method to modify the torsional natural frequencies using a coupling design with continuously adjustable torsional stiffness. The presented coupling design is investigated with torsional analysis and experimental measurements. Torsional analysis was utilized to predict the effects of varying the coupling stiffness to the torsional natural frequencies of a powertrain. The experimental measurements were conducted by attaching the adjustable stiffness coupling to the powertrain and measuring the change in the torsional natural frequencies while the torsional stiffness of the coupling was adjusted. The torsional natural frequencies were determined from the measurements by identifying the resonance induced torsional vibrations from the vibration response of the powertrain. The torsional vibrations were excited to the system by a load motor. The measurements showed that the first torsional natural frequency of the powertrain changed from 15.6 Hz to 43.0 Hz as the torsional stiffness of the coupling was adjusted. The results of the torsional analysis and the experimental measurements were compared to determine the performance of the realized coupling. The results indicated that the torsional natural frequencies determined by torsional analysis agree well with the experimentally measured results. The prediction errors were generally below $\pm 5\%$, which is typically considered as a margin for accurate torsional analysis.

Intelligent fault diagnosis (IFD) models have the potential to increase the level of automation and the diagnosis accuracy of machine condition monitoring systems. Many of the latest IFD models rely on convolutional layers for feature extraction from vibration data. The majority of these models employ batch normalisation (BN) for centring and scaling the input for each neuron. This study includes a novel examination of a competitive approach for layer input normalisation in the scope of fault diagnosis. Network deconvolution (ND) is a technique that further decorrelates the layer inputs reducing redundancy among the learned features. Both normalisation techniques are implemented on three common 1D-CNN-based fault diagnosis models. The models with ND mostly outperform the baseline models with BN in three experiments concerning fault datasets from two different rotor systems. Furthermore, the models with ND significantly outperform the baseline models with BN in the common CWRU bearing fault tests with load domain shifts, if the data from drive-end and fan-end sensors are employed. The results show that whitened features can improve the performance of CNN-based fault diagnosis models.

Bending stiffness variation (BSV) is a common problem causing vibration in large rotating machinery. BSV describes lateral bending stiffness and its variation as a function of the rotational angle. It has been observed that BSV causes excitation exactly twice per revolution, which leads to vibration problems, especially at half critical speed. BSV is caused by rotor geometry errors if the material is assumed to be homogeneous and linearly elastic. Therefore, the study investigated BSV with harmonic roundness components, which are commonly used in industry to describe the geometry of a rotor. Hence, the results are easily applicable in the industry. The research was conducted primarily by analytical means, but also static simulations and numeric calculations were used. The results clearly showed that when the effect of single harmonic roundness components in rotor cross-sections were observed, only the second component could produce BSV. However, when component pairs were studied, they produced BSV also without the second component. If the second component was included, the profile produced BSV the most aggressively. A generated arbitrary roundness profile, including components 3–50 with random phases and amplitudes, indicated that BSV occurs always twice per revolution despite different components in the profile. The results improve the possibilities of eliminating excessive BSV in the industry, when certain components and component pairs can be avoided.