ABSTRACT
Generative adversarial networks (GANs) have proven to be an effective tool for image generation, particularly in the realm of data augmentation. This chapter delves into the implementation of an auxiliary classifier GAN-based framework designed to learn from mechanical sensor signals and create realistic one-dimensional raw data. The overarching goal of this approach is to generate synthesized signals accompanied by labels, making them suitable for use as augmented data in machine fault diagnosis applications. To assess the performance of this generative model, a comprehensive set of evaluation measures is introduced, encompassing statistical characteristics and experimental verification. The effectiveness of the proposed framework is examined using a dataset of induction motor vibration signals.
