Intelligent Fault Diagnosis of Variable Air-Gap Rotational Resolver Based on Convolutional Neural Networks

Abstract

Resolvers are widely employed in industrial drives and precision motion control systems due to their robustness and high reliability. Nevertheless, the electrical and mechanical faults in resolver windings or mechanical alignment can lead to signal distortion, reduced accuracy, and potential system failure. The reliable and automatic fault diagnosis methods are therefore essential for condition monitoring and preventive maintenance. This article proposes an intelligent fault diagnosis method based on a convolutional neural network (CNN) using raw sine and cosine output signals of the resolver. A dataset containing multiple fault conditions, including eccentricities and short circuit fault, was generated through detailed simulations. The trained model performs inference on raw resolver signals using a sliding-window voting strategy, enabling reliable multiclass fault identification without additional signal processing. Finally, the prototype of the variable reluctance resolver was experimentally tested. Although the model was not trained with experimental data, it showed good performance on unseen experimental results, demonstrating the accuracy of the model.