Abstract:With the development of commercial spaceflight, the hot-fire testing of reusable liquid rocket engines (LREs) has also been growing, leading to a significant increase in the amount of data. However, among this large amount of data, the proportion of fault data is very small, and the coverage of fault modes is limited, which poses a challenge to the development of fault diagnosis systems. In addition, due to the high level of confidentiality of the test data, the data is usually not shared with system developers, which becomes a longstanding obstacle to data-driven research in the space industry. In order to address the shortage of partial data, there is a great deal of research on simulation-based method. But the simulation data generating from models are different from that of real environment, which makes little effort to data-driven anomaly detection research. Thus, we explored how Generative Adversarial Networks (GAN) can be utilized for complementing real datasets with time-dynamic property. At the same time, since the synthetic data has the same characteristics as the real data but does not contain complete information, it opens up the possibility of providing confidential data to system developers. As a specific target, the focus of this paper is on synthesizing time series data. The largest challenge is to ensure that the synthetic data has engineering values, such as the relationships on synthetic sequences should be consistent with the original variables. Especially for LREs, there is a strong correlation between the data monitored by some sensors of some components, which cannot be ignored when performing data analysis. Considering this unique temporal correlation, Time Series Generative Adversarial Network (TimeGAN) with three loss functions for objective optimization is developed for this effort. In addition to generate and discriminate modules in traditional GAN, it contains embedding and recovery modules, and introduces hidden latent. We use TimeGAN to learn a test data containing eight sequences and generate realistic “fake data”. We used autoregressive moving average models to test the residuals of the individual synthetic sequences and the results demonstrate that this method is well suited for data synthesis of sensory data from LREs.

Time-series Transformer Generative Adversarial Networks

TTS-GAN: A Transformer-based Time-Series Generative Adversarial Network

Time-series Generative Adversarial Networks

GT-GAN: General Purpose Time Series Synthesis with Generative Adversarial Networks

A Survey of Transformer Enabled Time Series Synthesis

Time-Transformer: Integrating Local and Global Features for Better Time Series Generation

GAT-GAN : A Graph-Attention-based Time-Series Generative Adversarial Network

SeriesGAN: Time Series Generation via Adversarial and Autoregressive Learning

TTS-CGAN: A Transformer Time-Series Conditional GAN for Biosignal Data Augmentation

Generative Adversarial Networks in Time Series: A Systematic Literature Review

Fully Embedded Time-Series Generative Adversarial Networks

Representation Learning of Multivariate Time Series using Attention and Adversarial Training

DCT-GAN: Dilated Convolutional Transformer-Based GAN for Time Series Anomaly Detection

Generative adversarial transformers

Generative adversarial networks in time series: A survey and taxonomy

Directed Chain Generative Adversarial Networks

TSA-GAN: A Robust Generative Adversarial Networks for Time Series Augmentation

Towards Generating Real-World Time Series Data

TS-GAN: Time-series GAN for Sensor-based Health Data Augmentation

Synthetic Data Generation of Complex Machines with Time Series Generative Adversarial Network