A Novel Multivariate Degradation Data Generation Method Based on Flow Model

With the growing popularity of multivariate degradation data in the field of Prognostics and Health Management (PHM), the ability in generating or imputating high-accuracy and high-reliability data becomes critical. This paper proposes a multivariate degradation data generation model based on a flow model. Firstly, the JS-NICE model is constructed based on the flow model and JS divergence. Then, the generative model is trained to learn the rules of the original data, and the data are generated by sampling. Finally, the experimental results validates the excellent quality and accuracy of the generated data by the model proposed, which indicates that the proposed method has learned the distribution law of the original data, and at the same time, it provides an important advance check information for tasks such as fault detection and remaining useful life prediction.