A Novel Model Based on Deep Convolutional Generation of Adversarial Networks Using Meta-Learning (MAML-DCGAN)

Aircraft fault diagnosis is the basis for fault-tolerant control of aircraft. Deep learning is an emerging data-based approach to fault diagnosis. Deep convolutional generative adversarial network (DCGAN) is a data augmentation model. Model-Agnostic Meta-Learning (MAML) allows the model to obtain good initialization parameters, which produces better results. This study solves the problem of data augmentation through a data generation algorithm based on MAML and DCGAN. The algorithm uses the MAML model to optimize DCGAN so that DCGAN obtains the best initialization parameters to generate higher quality data. In this paper, vehicle failure data is used as the original data. Comparing the Fréchet inception distance score (FID) of the data generated by DCGAN and MAML-DCGAN, the experiments show that the data generated by the MAML-DCGAN is more similar to the original data and is more advantageous for fault diagnosis and analysis.