A transient model of an induction machine (IM) is developed in this work using an artificial neural network (ANN) surrogate model. The model is suitable to be used for direct-on-line IMs. The finite-element (FE)-based model of IM is used to generate the training, validation, and testing datasets. Different inputs and model configurations are investigated to find an optimal solution in developing the transient model. The proposed transient model is suitable to be used in digital twin services since it can estimate the current and torque accurately in real time based on only voltage and measured shaft speed.

Electromagnetic force and torque computations for a bearingless synchronous reluctance motor present multi-input complexity due to the levitating auxiliary winding located in the machine active parts. The machine model complexity is reduced through the proper orthogonal decomposition method. For this purpose, a snapshot matrix is needed to be constructed. This paper proposes a methodology of choosing efficient snapshots for the model order reduction. The reduced model is then applied to compute the torque and electromagnetic forces on the rotor of the machine. The method presented for selecting the snapshots is shown to be more efficient than uniformly distributed snapshots and it reduces the number of finite element method evaluations needed for the construction of the reduced model.

A bearingless machine should be designed to maximize the average and to minimize the ripple in the electromagnetic torque and levitating force. In a bearingless synchronous reluctance machine (BSynRM), the complex rotor design and the levitation winding create a rich spatial and temporal spectrum of the magnetic flux density in the airgap, which affects both the torque and force ripple. In this paper, a novel method is introduced to compute the electromagnetic forces from the spatial harmonics of the airgap flux density. A new indexing method is introduced to classify the spatial harmonic wavenumbers and temporal harmonic frequencies of the flux density in the electromagnetic force. The proposed model enables the understanding of the spatial harmonic footprint on the temporal frequencies of the electromagnetic force ripple. The study is carried out through finite element simulations, which are verified by prototype measurements.

This paper presents the methodology and computation results for the torque and iron losses in a bearingless synchronous reluctance machine. The aim is to compute the hysteresis torque related to the asynchronous operation of the machine under the effect of the levitation field rotating at twice the synchronous speed. For this purpose, and because it is impossible to compute the hysteresis torque with a single-valued B-H characteristic, a magneto-dynamic vector hysteresis model is used. The computations show that the hysteresis-related shaft power is considerable under no-load operation of the machine. It is also shown that the asynchronous torque can be approximated by the classical equivalent circuit of an induction machine.

A robust open-source cloud-based workflow is developed for finite element (FE) data generation for active learning (AL) -based surrogate modelling. Special attention is paid to making the FE solution procedure as robust and fast as possible without human intervention by, e.g., implementing special convergence criteria, reliable parallel computation, and variable timestep length. In AL, a surrogate model automatically improves itself by iteratively querying more FE data. Using AL and large datasets generated with parallelised cloud FE simulations, we develop a surrogate model to rapidly predict induction machine steady-state torque, torque ripple, total losses, and current harmonic distortion, as a function of motor frequency, voltage, and slip. Results show that AL performs better than grid sampling and on average works as well as random sampling, but with some outputs, the results vary less with AL. In addition, accurate ripple estimation requires a much larger training dataset than the other variables.

This article presents a novel data augmentation method that generates feature values for unmeasured loading levels based on limited measured and simulated loading level data. The incorporation of offline simulated data in the augmentation framework and the mapping of the error distribution over the loading levels greatly reduce the dependency on including a large number of loading levels in the curve fitting process. Furthermore, the proposed method shows high potential to minimize the deviation between measured and simulated data at the feature level. The method is applied to the induction machine to generate feature values at 25% and 50% loading levels for healthy, one, two, and three broken rotor bars conditions. An excellent agreement is observed between the augmented and actual feature values calculated from the measured data at 25% and 50% loading levels. The inclusion of this augmented data in the training phase aids in resolving the generalization issue and enhancing the average classification accuracy of the XGBoost algorithm by 9.4% and 4.4% at 25% and 50% loading levels, respectively.