A Novel Fault Detection Method Based on One-Dimension Convolutional Adversarial Autoencoder (1DAAE)

Fault detection is an important and demanding problem in industry. Recently, many researchers have addressed the use of deep learning architectures for fault applications such as autoencoder. Traditional methods based on autoencoder usually complete by comparing reconstruction errors, ignore a lot useful information about distribution latent variables. To deal with this problem, paper proposes novel unsupervised method named one-dimension convolutional adversarial (1DAAE), which introduces two new ideas: convolution layers encoder to obtain better features thought impose variable z cluster into prior distribution. The proposed not only has powerful feature representation ability than traditional autoencoder, but also enhanced discrimination imposing variables cluster. Then, anomaly scores 1DAAE were detect samples, one other Finally, it was shown experiments that outperformed autoencoder-based, autoencoder-based generative network-based algorithms Tennessee Eastman process. Through experiments, we found both vector are helpful detection.